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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecEwSh_VdwLJEwSh_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_ElecEwSh_VdwLJEwSh_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     /* 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             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 223                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 224
 225             /* EWALD ELECTROSTATICS */
 226
 227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 228             ewrt             = _mm_mul_pd(r00,ewtabscale);
 229             ewitab           = _mm_cvttpd_epi32(ewrt);
 230             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 231             ewitab           = _mm_slli_epi32(ewitab,2);
 232             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 233             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 234             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 235             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 236             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 237             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 238             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 239             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 240             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 241             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 242
 243             /* Analytical LJ-PME */
 244             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 245             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 246             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 247             exponent         = gmx_simd_exp_d(ewcljrsq);
 248             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 249             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 250             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 251             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 252             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 253             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),
 254                                _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));
 255             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 256             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);
 257
 258             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velec            = _mm_and_pd(velec,cutoff_mask);
 262             velecsum         = _mm_add_pd(velecsum,velec);
 263             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 264             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 265
 266             fscal            = _mm_add_pd(felec,fvdw);
 267
 268             fscal            = _mm_and_pd(fscal,cutoff_mask);
 269
 270             /* Calculate temporary vectorial force */
 271             tx               = _mm_mul_pd(fscal,dx00);
 272             ty               = _mm_mul_pd(fscal,dy00);
 273             tz               = _mm_mul_pd(fscal,dz00);
 274
 275             /* Update vectorial force */
 276             fix0             = _mm_add_pd(fix0,tx);
 277             fiy0             = _mm_add_pd(fiy0,ty);
 278             fiz0             = _mm_add_pd(fiz0,tz);
 279
 280             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 281
 282             }
 283
 284             /* Inner loop uses 81 flops */
 285         }
 286
 287         if(jidx<j_index_end)
 288         {
 289
 290             jnrA             = jjnr[jidx];
 291             j_coord_offsetA  = DIM*jnrA;
 292
 293             /* load j atom coordinates */
 294             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 295                                               &jx0,&jy0,&jz0);
 296
 297             /* Calculate displacement vector */
 298             dx00             = _mm_sub_pd(ix0,jx0);
 299             dy00             = _mm_sub_pd(iy0,jy0);
 300             dz00             = _mm_sub_pd(iz0,jz0);
 301
 302             /* Calculate squared distance and things based on it */
 303             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 304
 305             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 306
 307             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 308
 309             /* Load parameters for j particles */
 310             jq0              = _mm_load_sd(charge+jnrA+0);
 311             vdwjidx0A        = 2*vdwtype[jnrA+0];
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             if (gmx_mm_any_lt(rsq00,rcutoff2))
 318             {
 319
 320             r00              = _mm_mul_pd(rsq00,rinv00);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq00             = _mm_mul_pd(iq0,jq0);
 324             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 325
 326             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 327
 328             /* EWALD ELECTROSTATICS */
 329
 330             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 331             ewrt             = _mm_mul_pd(r00,ewtabscale);
 332             ewitab           = _mm_cvttpd_epi32(ewrt);
 333             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 334             ewitab           = _mm_slli_epi32(ewitab,2);
 335             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 336             ewtabD           = _mm_setzero_pd();
 337             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 338             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 339             ewtabFn          = _mm_setzero_pd();
 340             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 341             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 342             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 343             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 344             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 345
 346             /* Analytical LJ-PME */
 347             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 348             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 349             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 350             exponent         = gmx_simd_exp_d(ewcljrsq);
 351             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 352             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 353             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 354             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 355             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 356             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),
 357                                _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));
 358             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 359             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);
 360
 361             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velec            = _mm_and_pd(velec,cutoff_mask);
 365             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 366             velecsum         = _mm_add_pd(velecsum,velec);
 367             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 368             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 369             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 370
 371             fscal            = _mm_add_pd(felec,fvdw);
 372
 373             fscal            = _mm_and_pd(fscal,cutoff_mask);
 374
 375             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm_mul_pd(fscal,dx00);
 379             ty               = _mm_mul_pd(fscal,dy00);
 380             tz               = _mm_mul_pd(fscal,dz00);
 381
 382             /* Update vectorial force */
 383             fix0             = _mm_add_pd(fix0,tx);
 384             fiy0             = _mm_add_pd(fiy0,ty);
 385             fiz0             = _mm_add_pd(fiz0,tz);
 386
 387             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 388
 389             }
 390
 391             /* Inner loop uses 81 flops */
 392         }
 393
 394         /* End of innermost loop */
 395
 396         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 397                                               f+i_coord_offset,fshift+i_shift_offset);
 398
 399         ggid                        = gid[iidx];
 400         /* Update potential energies */
 401         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 402         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 403
 404         /* Increment number of inner iterations */
 405         inneriter                  += j_index_end - j_index_start;
 406
 407         /* Outer loop uses 9 flops */
 408     }
 409
 410     /* Increment number of outer iterations */
 411     outeriter        += nri;
 412
 413     /* Update outer/inner flops */
 414
 415     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*81);
 416 }
 417 /*
 418  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 419  * Electrostatics interaction: Ewald
 420  * VdW interaction:            LJEwald
 421  * Geometry:                   Particle-Particle
 422  * Calculate force/pot:        Force
 423  */
 424 void
 425 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 426                     (t_nblist                    * gmx_restrict       nlist,
 427                      rvec                        * gmx_restrict          xx,
 428                      rvec                        * gmx_restrict          ff,
 429                      t_forcerec                  * gmx_restrict          fr,
 430                      t_mdatoms                   * gmx_restrict     mdatoms,
 431                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 432                      t_nrnb                      * gmx_restrict        nrnb)
 433 {
 434     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 435      * just 0 for non-waters.
 436      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 437      * jnr indices corresponding to data put in the four positions in the SIMD register.
 438      */
 439     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 440     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 441     int              jnrA,jnrB;
 442     int              j_coord_offsetA,j_coord_offsetB;
 443     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 444     real             rcutoff_scalar;
 445     real             *shiftvec,*fshift,*x,*f;
 446     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 447     int              vdwioffset0;
 448     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 449     int              vdwjidx0A,vdwjidx0B;
 450     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 451     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 452     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 453     real             *charge;
 454     int              nvdwtype;
 455     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 456     int              *vdwtype;
 457     real             *vdwparam;
 458     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 459     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 460     __m128d           c6grid_00;
 461     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 462     real             *vdwgridparam;
 463     __m128d           one_half = _mm_set1_pd(0.5);
 464     __m128d           minus_one = _mm_set1_pd(-1.0);
 465     __m128i          ewitab;
 466     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 467     real             *ewtab;
 468     __m128d          dummy_mask,cutoff_mask;
 469     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 470     __m128d          one     = _mm_set1_pd(1.0);
 471     __m128d          two     = _mm_set1_pd(2.0);
 472     x                = xx[0];
 473     f                = ff[0];
 474
 475     nri              = nlist->nri;
 476     iinr             = nlist->iinr;
 477     jindex           = nlist->jindex;
 478     jjnr             = nlist->jjnr;
 479     shiftidx         = nlist->shift;
 480     gid              = nlist->gid;
 481     shiftvec         = fr->shift_vec[0];
 482     fshift           = fr->fshift[0];
 483     facel            = _mm_set1_pd(fr->epsfac);
 484     charge           = mdatoms->chargeA;
 485     nvdwtype         = fr->ntype;
 486     vdwparam         = fr->nbfp;
 487     vdwtype          = mdatoms->typeA;
 488     vdwgridparam     = fr->ljpme_c6grid;
 489     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 490     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 491     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 492
 493     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 494     ewtab            = fr->ic->tabq_coul_F;
 495     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 496     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 497
 498     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 499     rcutoff_scalar   = fr->rcoulomb;
 500     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 501     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 502
 503     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 504     rvdw             = _mm_set1_pd(fr->rvdw);
 505
 506     /* Avoid stupid compiler warnings */
 507     jnrA = jnrB = 0;
 508     j_coord_offsetA = 0;
 509     j_coord_offsetB = 0;
 510
 511     outeriter        = 0;
 512     inneriter        = 0;
 513
 514     /* Start outer loop over neighborlists */
 515     for(iidx=0; iidx<nri; iidx++)
 516     {
 517         /* Load shift vector for this list */
 518         i_shift_offset   = DIM*shiftidx[iidx];
 519
 520         /* Load limits for loop over neighbors */
 521         j_index_start    = jindex[iidx];
 522         j_index_end      = jindex[iidx+1];
 523
 524         /* Get outer coordinate index */
 525         inr              = iinr[iidx];
 526         i_coord_offset   = DIM*inr;
 527
 528         /* Load i particle coords and add shift vector */
 529         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 530
 531         fix0             = _mm_setzero_pd();
 532         fiy0             = _mm_setzero_pd();
 533         fiz0             = _mm_setzero_pd();
 534
 535         /* Load parameters for i particles */
 536         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 537         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 538
 539         /* Start inner kernel loop */
 540         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 541         {
 542
 543             /* Get j neighbor index, and coordinate index */
 544             jnrA             = jjnr[jidx];
 545             jnrB             = jjnr[jidx+1];
 546             j_coord_offsetA  = DIM*jnrA;
 547             j_coord_offsetB  = DIM*jnrB;
 548
 549             /* load j atom coordinates */
 550             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 551                                               &jx0,&jy0,&jz0);
 552
 553             /* Calculate displacement vector */
 554             dx00             = _mm_sub_pd(ix0,jx0);
 555             dy00             = _mm_sub_pd(iy0,jy0);
 556             dz00             = _mm_sub_pd(iz0,jz0);
 557
 558             /* Calculate squared distance and things based on it */
 559             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 560
 561             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 562
 563             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 564
 565             /* Load parameters for j particles */
 566             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 567             vdwjidx0A        = 2*vdwtype[jnrA+0];
 568             vdwjidx0B        = 2*vdwtype[jnrB+0];
 569
 570             /**************************
 571              * CALCULATE INTERACTIONS *
 572              **************************/
 573
 574             if (gmx_mm_any_lt(rsq00,rcutoff2))
 575             {
 576
 577             r00              = _mm_mul_pd(rsq00,rinv00);
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             qq00             = _mm_mul_pd(iq0,jq0);
 581             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 582                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 583             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 584                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 585
 586             /* EWALD ELECTROSTATICS */
 587
 588             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 589             ewrt             = _mm_mul_pd(r00,ewtabscale);
 590             ewitab           = _mm_cvttpd_epi32(ewrt);
 591             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 592             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 593                                          &ewtabF,&ewtabFn);
 594             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 595             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 596
 597             /* Analytical LJ-PME */
 598             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 599             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 600             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 601             exponent         = gmx_simd_exp_d(ewcljrsq);
 602             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 603             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 604             /* f6A = 6 * C6grid * (1 - poly) */
 605             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 606             /* f6B = C6grid * exponent * beta^6 */
 607             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 608             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 609             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);
 610
 611             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 612
 613             fscal            = _mm_add_pd(felec,fvdw);
 614
 615             fscal            = _mm_and_pd(fscal,cutoff_mask);
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm_mul_pd(fscal,dx00);
 619             ty               = _mm_mul_pd(fscal,dy00);
 620             tz               = _mm_mul_pd(fscal,dz00);
 621
 622             /* Update vectorial force */
 623             fix0             = _mm_add_pd(fix0,tx);
 624             fiy0             = _mm_add_pd(fiy0,ty);
 625             fiz0             = _mm_add_pd(fiz0,tz);
 626
 627             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 628
 629             }
 630
 631             /* Inner loop uses 62 flops */
 632         }
 633
 634         if(jidx<j_index_end)
 635         {
 636
 637             jnrA             = jjnr[jidx];
 638             j_coord_offsetA  = DIM*jnrA;
 639
 640             /* load j atom coordinates */
 641             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 642                                               &jx0,&jy0,&jz0);
 643
 644             /* Calculate displacement vector */
 645             dx00             = _mm_sub_pd(ix0,jx0);
 646             dy00             = _mm_sub_pd(iy0,jy0);
 647             dz00             = _mm_sub_pd(iz0,jz0);
 648
 649             /* Calculate squared distance and things based on it */
 650             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 651
 652             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 653
 654             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 655
 656             /* Load parameters for j particles */
 657             jq0              = _mm_load_sd(charge+jnrA+0);
 658             vdwjidx0A        = 2*vdwtype[jnrA+0];
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             if (gmx_mm_any_lt(rsq00,rcutoff2))
 665             {
 666
 667             r00              = _mm_mul_pd(rsq00,rinv00);
 668
 669             /* Compute parameters for interactions between i and j atoms */
 670             qq00             = _mm_mul_pd(iq0,jq0);
 671             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 672
 673             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 674
 675             /* EWALD ELECTROSTATICS */
 676
 677             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 678             ewrt             = _mm_mul_pd(r00,ewtabscale);
 679             ewitab           = _mm_cvttpd_epi32(ewrt);
 680             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 681             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 682             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 683             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 684
 685             /* Analytical LJ-PME */
 686             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 687             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 688             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 689             exponent         = gmx_simd_exp_d(ewcljrsq);
 690             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 691             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 692             /* f6A = 6 * C6grid * (1 - poly) */
 693             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 694             /* f6B = C6grid * exponent * beta^6 */
 695             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 696             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 697             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);
 698
 699             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 700
 701             fscal            = _mm_add_pd(felec,fvdw);
 702
 703             fscal            = _mm_and_pd(fscal,cutoff_mask);
 704
 705             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 706
 707             /* Calculate temporary vectorial force */
 708             tx               = _mm_mul_pd(fscal,dx00);
 709             ty               = _mm_mul_pd(fscal,dy00);
 710             tz               = _mm_mul_pd(fscal,dz00);
 711
 712             /* Update vectorial force */
 713             fix0             = _mm_add_pd(fix0,tx);
 714             fiy0             = _mm_add_pd(fiy0,ty);
 715             fiz0             = _mm_add_pd(fiz0,tz);
 716
 717             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 718
 719             }
 720
 721             /* Inner loop uses 62 flops */
 722         }
 723
 724         /* End of innermost loop */
 725
 726         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 727                                               f+i_coord_offset,fshift+i_shift_offset);
 728
 729         /* Increment number of inner iterations */
 730         inneriter                  += j_index_end - j_index_start;
 731
 732         /* Outer loop uses 7 flops */
 733     }
 734
 735     /* Increment number of outer iterations */
 736     outeriter        += nri;
 737
 738     /* Update outer/inner flops */
 739
 740     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 741 }