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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128d           c6grid_00;
  94     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  95     real             *vdwgridparam;
  96     __m128d           one_half = _mm_set1_pd(0.5);
  97     __m128d           minus_one = _mm_set1_pd(-1.0);
  98     __m128i          ewitab;
  99     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     __m128d          dummy_mask,cutoff_mask;
 102     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 103     __m128d          one     = _mm_set1_pd(1.0);
 104     __m128d          two     = _mm_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_pd(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121     vdwgridparam     = fr->ljpme_c6grid;
 122     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 123     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 124     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 125
 126     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 132     rcutoff_scalar   = fr->rcoulomb;
 133     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 134     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 135
 136     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 137     rvdw             = _mm_set1_pd(fr->rvdw);
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 163
 164         fix0             = _mm_setzero_pd();
 165         fiy0             = _mm_setzero_pd();
 166         fiz0             = _mm_setzero_pd();
 167
 168         /* Load parameters for i particles */
 169         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 170         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_pd();
 174         vvdwsum          = _mm_setzero_pd();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_pd(ix0,jx0);
 192             dy00             = _mm_sub_pd(iy0,jy0);
 193             dz00             = _mm_sub_pd(iz0,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 197
 198             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 199
 200             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 201
 202             /* Load parameters for j particles */
 203             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 204             vdwjidx0A        = 2*vdwtype[jnrA+0];
 205             vdwjidx0B        = 2*vdwtype[jnrB+0];
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             if (gmx_mm_any_lt(rsq00,rcutoff2))
 212             {
 213
 214             r00              = _mm_mul_pd(rsq00,rinv00);
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq00             = _mm_mul_pd(iq0,jq0);
 218             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 219                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 220             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 221                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 222
 223             /* EWALD ELECTROSTATICS */
 224
 225             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 226             ewrt             = _mm_mul_pd(r00,ewtabscale);
 227             ewitab           = _mm_cvttpd_epi32(ewrt);
 228             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 229             ewitab           = _mm_slli_epi32(ewitab,2);
 230             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 231             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 232             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 233             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 234             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 235             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 236             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 237             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 238             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 239             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 240
 241             /* Analytical LJ-PME */
 242             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 243             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 244             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 245             exponent         = gmx_simd_exp_d(ewcljrsq);
 246             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 247             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 248             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 249             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 250             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 251             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),
 252                                _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));
 253             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 254             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);
 255
 256             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             velec            = _mm_and_pd(velec,cutoff_mask);
 260             velecsum         = _mm_add_pd(velecsum,velec);
 261             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 262             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 263
 264             fscal            = _mm_add_pd(felec,fvdw);
 265
 266             fscal            = _mm_and_pd(fscal,cutoff_mask);
 267
 268             /* Calculate temporary vectorial force */
 269             tx               = _mm_mul_pd(fscal,dx00);
 270             ty               = _mm_mul_pd(fscal,dy00);
 271             tz               = _mm_mul_pd(fscal,dz00);
 272
 273             /* Update vectorial force */
 274             fix0             = _mm_add_pd(fix0,tx);
 275             fiy0             = _mm_add_pd(fiy0,ty);
 276             fiz0             = _mm_add_pd(fiz0,tz);
 277
 278             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 279
 280             }
 281
 282             /* Inner loop uses 81 flops */
 283         }
 284
 285         if(jidx<j_index_end)
 286         {
 287
 288             jnrA             = jjnr[jidx];
 289             j_coord_offsetA  = DIM*jnrA;
 290
 291             /* load j atom coordinates */
 292             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 293                                               &jx0,&jy0,&jz0);
 294
 295             /* Calculate displacement vector */
 296             dx00             = _mm_sub_pd(ix0,jx0);
 297             dy00             = _mm_sub_pd(iy0,jy0);
 298             dz00             = _mm_sub_pd(iz0,jz0);
 299
 300             /* Calculate squared distance and things based on it */
 301             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 302
 303             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 304
 305             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 306
 307             /* Load parameters for j particles */
 308             jq0              = _mm_load_sd(charge+jnrA+0);
 309             vdwjidx0A        = 2*vdwtype[jnrA+0];
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             if (gmx_mm_any_lt(rsq00,rcutoff2))
 316             {
 317
 318             r00              = _mm_mul_pd(rsq00,rinv00);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq00             = _mm_mul_pd(iq0,jq0);
 322             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 323
 324             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 325
 326             /* EWALD ELECTROSTATICS */
 327
 328             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 329             ewrt             = _mm_mul_pd(r00,ewtabscale);
 330             ewitab           = _mm_cvttpd_epi32(ewrt);
 331             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 332             ewitab           = _mm_slli_epi32(ewitab,2);
 333             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 334             ewtabD           = _mm_setzero_pd();
 335             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 336             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 337             ewtabFn          = _mm_setzero_pd();
 338             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 339             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 340             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 341             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 342             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 343
 344             /* Analytical LJ-PME */
 345             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 346             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 347             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 348             exponent         = gmx_simd_exp_d(ewcljrsq);
 349             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 350             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 351             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 352             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 353             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 354             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),
 355                                _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));
 356             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 357             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);
 358
 359             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_and_pd(velec,cutoff_mask);
 363             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 364             velecsum         = _mm_add_pd(velecsum,velec);
 365             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 366             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 367             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 368
 369             fscal            = _mm_add_pd(felec,fvdw);
 370
 371             fscal            = _mm_and_pd(fscal,cutoff_mask);
 372
 373             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 374
 375             /* Calculate temporary vectorial force */
 376             tx               = _mm_mul_pd(fscal,dx00);
 377             ty               = _mm_mul_pd(fscal,dy00);
 378             tz               = _mm_mul_pd(fscal,dz00);
 379
 380             /* Update vectorial force */
 381             fix0             = _mm_add_pd(fix0,tx);
 382             fiy0             = _mm_add_pd(fiy0,ty);
 383             fiz0             = _mm_add_pd(fiz0,tz);
 384
 385             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 386
 387             }
 388
 389             /* Inner loop uses 81 flops */
 390         }
 391
 392         /* End of innermost loop */
 393
 394         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 395                                               f+i_coord_offset,fshift+i_shift_offset);
 396
 397         ggid                        = gid[iidx];
 398         /* Update potential energies */
 399         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 400         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 401
 402         /* Increment number of inner iterations */
 403         inneriter                  += j_index_end - j_index_start;
 404
 405         /* Outer loop uses 9 flops */
 406     }
 407
 408     /* Increment number of outer iterations */
 409     outeriter        += nri;
 410
 411     /* Update outer/inner flops */
 412
 413     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*81);
 414 }
 415 /*
 416  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 417  * Electrostatics interaction: Ewald
 418  * VdW interaction:            LJEwald
 419  * Geometry:                   Particle-Particle
 420  * Calculate force/pot:        Force
 421  */
 422 void
 423 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse4_1_double
 424                     (t_nblist                    * gmx_restrict       nlist,
 425                      rvec                        * gmx_restrict          xx,
 426                      rvec                        * gmx_restrict          ff,
 427                      t_forcerec                  * gmx_restrict          fr,
 428                      t_mdatoms                   * gmx_restrict     mdatoms,
 429                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 430                      t_nrnb                      * gmx_restrict        nrnb)
 431 {
 432     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 433      * just 0 for non-waters.
 434      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 435      * jnr indices corresponding to data put in the four positions in the SIMD register.
 436      */
 437     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 438     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 439     int              jnrA,jnrB;
 440     int              j_coord_offsetA,j_coord_offsetB;
 441     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 442     real             rcutoff_scalar;
 443     real             *shiftvec,*fshift,*x,*f;
 444     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 445     int              vdwioffset0;
 446     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 447     int              vdwjidx0A,vdwjidx0B;
 448     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 449     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 450     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 451     real             *charge;
 452     int              nvdwtype;
 453     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 454     int              *vdwtype;
 455     real             *vdwparam;
 456     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 457     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 458     __m128d           c6grid_00;
 459     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 460     real             *vdwgridparam;
 461     __m128d           one_half = _mm_set1_pd(0.5);
 462     __m128d           minus_one = _mm_set1_pd(-1.0);
 463     __m128i          ewitab;
 464     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 465     real             *ewtab;
 466     __m128d          dummy_mask,cutoff_mask;
 467     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 468     __m128d          one     = _mm_set1_pd(1.0);
 469     __m128d          two     = _mm_set1_pd(2.0);
 470     x                = xx[0];
 471     f                = ff[0];
 472
 473     nri              = nlist->nri;
 474     iinr             = nlist->iinr;
 475     jindex           = nlist->jindex;
 476     jjnr             = nlist->jjnr;
 477     shiftidx         = nlist->shift;
 478     gid              = nlist->gid;
 479     shiftvec         = fr->shift_vec[0];
 480     fshift           = fr->fshift[0];
 481     facel            = _mm_set1_pd(fr->epsfac);
 482     charge           = mdatoms->chargeA;
 483     nvdwtype         = fr->ntype;
 484     vdwparam         = fr->nbfp;
 485     vdwtype          = mdatoms->typeA;
 486     vdwgridparam     = fr->ljpme_c6grid;
 487     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 488     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 489     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 490
 491     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 492     ewtab            = fr->ic->tabq_coul_F;
 493     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 494     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 495
 496     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 497     rcutoff_scalar   = fr->rcoulomb;
 498     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 499     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 500
 501     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 502     rvdw             = _mm_set1_pd(fr->rvdw);
 503
 504     /* Avoid stupid compiler warnings */
 505     jnrA = jnrB = 0;
 506     j_coord_offsetA = 0;
 507     j_coord_offsetB = 0;
 508
 509     outeriter        = 0;
 510     inneriter        = 0;
 511
 512     /* Start outer loop over neighborlists */
 513     for(iidx=0; iidx<nri; iidx++)
 514     {
 515         /* Load shift vector for this list */
 516         i_shift_offset   = DIM*shiftidx[iidx];
 517
 518         /* Load limits for loop over neighbors */
 519         j_index_start    = jindex[iidx];
 520         j_index_end      = jindex[iidx+1];
 521
 522         /* Get outer coordinate index */
 523         inr              = iinr[iidx];
 524         i_coord_offset   = DIM*inr;
 525
 526         /* Load i particle coords and add shift vector */
 527         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 528
 529         fix0             = _mm_setzero_pd();
 530         fiy0             = _mm_setzero_pd();
 531         fiz0             = _mm_setzero_pd();
 532
 533         /* Load parameters for i particles */
 534         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 535         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 536
 537         /* Start inner kernel loop */
 538         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 539         {
 540
 541             /* Get j neighbor index, and coordinate index */
 542             jnrA             = jjnr[jidx];
 543             jnrB             = jjnr[jidx+1];
 544             j_coord_offsetA  = DIM*jnrA;
 545             j_coord_offsetB  = DIM*jnrB;
 546
 547             /* load j atom coordinates */
 548             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 549                                               &jx0,&jy0,&jz0);
 550
 551             /* Calculate displacement vector */
 552             dx00             = _mm_sub_pd(ix0,jx0);
 553             dy00             = _mm_sub_pd(iy0,jy0);
 554             dz00             = _mm_sub_pd(iz0,jz0);
 555
 556             /* Calculate squared distance and things based on it */
 557             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 558
 559             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 560
 561             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 562
 563             /* Load parameters for j particles */
 564             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 565             vdwjidx0A        = 2*vdwtype[jnrA+0];
 566             vdwjidx0B        = 2*vdwtype[jnrB+0];
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm_any_lt(rsq00,rcutoff2))
 573             {
 574
 575             r00              = _mm_mul_pd(rsq00,rinv00);
 576
 577             /* Compute parameters for interactions between i and j atoms */
 578             qq00             = _mm_mul_pd(iq0,jq0);
 579             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 580                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 581             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 582                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 583
 584             /* EWALD ELECTROSTATICS */
 585
 586             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 587             ewrt             = _mm_mul_pd(r00,ewtabscale);
 588             ewitab           = _mm_cvttpd_epi32(ewrt);
 589             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 590             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 591                                          &ewtabF,&ewtabFn);
 592             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 593             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 594
 595             /* Analytical LJ-PME */
 596             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 597             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 598             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 599             exponent         = gmx_simd_exp_d(ewcljrsq);
 600             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 601             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 602             /* f6A = 6 * C6grid * (1 - poly) */
 603             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 604             /* f6B = C6grid * exponent * beta^6 */
 605             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 606             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 607             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);
 608
 609             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 610
 611             fscal            = _mm_add_pd(felec,fvdw);
 612
 613             fscal            = _mm_and_pd(fscal,cutoff_mask);
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm_mul_pd(fscal,dx00);
 617             ty               = _mm_mul_pd(fscal,dy00);
 618             tz               = _mm_mul_pd(fscal,dz00);
 619
 620             /* Update vectorial force */
 621             fix0             = _mm_add_pd(fix0,tx);
 622             fiy0             = _mm_add_pd(fiy0,ty);
 623             fiz0             = _mm_add_pd(fiz0,tz);
 624
 625             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 626
 627             }
 628
 629             /* Inner loop uses 62 flops */
 630         }
 631
 632         if(jidx<j_index_end)
 633         {
 634
 635             jnrA             = jjnr[jidx];
 636             j_coord_offsetA  = DIM*jnrA;
 637
 638             /* load j atom coordinates */
 639             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 640                                               &jx0,&jy0,&jz0);
 641
 642             /* Calculate displacement vector */
 643             dx00             = _mm_sub_pd(ix0,jx0);
 644             dy00             = _mm_sub_pd(iy0,jy0);
 645             dz00             = _mm_sub_pd(iz0,jz0);
 646
 647             /* Calculate squared distance and things based on it */
 648             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 649
 650             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 651
 652             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 653
 654             /* Load parameters for j particles */
 655             jq0              = _mm_load_sd(charge+jnrA+0);
 656             vdwjidx0A        = 2*vdwtype[jnrA+0];
 657
 658             /**************************
 659              * CALCULATE INTERACTIONS *
 660              **************************/
 661
 662             if (gmx_mm_any_lt(rsq00,rcutoff2))
 663             {
 664
 665             r00              = _mm_mul_pd(rsq00,rinv00);
 666
 667             /* Compute parameters for interactions between i and j atoms */
 668             qq00             = _mm_mul_pd(iq0,jq0);
 669             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 670
 671             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 672
 673             /* EWALD ELECTROSTATICS */
 674
 675             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 676             ewrt             = _mm_mul_pd(r00,ewtabscale);
 677             ewitab           = _mm_cvttpd_epi32(ewrt);
 678             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 679             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 680             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 681             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 682
 683             /* Analytical LJ-PME */
 684             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 685             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 686             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 687             exponent         = gmx_simd_exp_d(ewcljrsq);
 688             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 689             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 690             /* f6A = 6 * C6grid * (1 - poly) */
 691             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 692             /* f6B = C6grid * exponent * beta^6 */
 693             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 694             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 695             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);
 696
 697             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 698
 699             fscal            = _mm_add_pd(felec,fvdw);
 700
 701             fscal            = _mm_and_pd(fscal,cutoff_mask);
 702
 703             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm_mul_pd(fscal,dx00);
 707             ty               = _mm_mul_pd(fscal,dy00);
 708             tz               = _mm_mul_pd(fscal,dz00);
 709
 710             /* Update vectorial force */
 711             fix0             = _mm_add_pd(fix0,tx);
 712             fiy0             = _mm_add_pd(fiy0,ty);
 713             fiz0             = _mm_add_pd(fiz0,tz);
 714
 715             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 716
 717             }
 718
 719             /* Inner loop uses 62 flops */
 720         }
 721
 722         /* End of innermost loop */
 723
 724         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 725                                               f+i_coord_offset,fshift+i_shift_offset);
 726
 727         /* Increment number of inner iterations */
 728         inneriter                  += j_index_end - j_index_start;
 729
 730         /* Outer loop uses 7 flops */
 731     }
 732
 733     /* Increment number of outer iterations */
 734     outeriter        += nri;
 735
 736     /* Update outer/inner flops */
 737
 738     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 739 }