src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #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_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sse2_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_sse2_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
 221             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 222                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 223
 224             /* EWALD ELECTROSTATICS */
 225
 226             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 227             ewrt             = _mm_mul_pd(r00,ewtabscale);
 228             ewitab           = _mm_cvttpd_epi32(ewrt);
 229             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 230             ewitab           = _mm_slli_epi32(ewitab,2);
 231             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 232             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 233             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 234             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 235             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 236             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 237             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 238             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 239             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 240             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 241
 242             /* Analytical LJ-PME */
 243             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 244             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 245             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 246             exponent         = gmx_simd_exp_d(ewcljrsq);
 247             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 248             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 249             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 250             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 251             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 252             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),
 253                                _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));
 254             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 255             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);
 256
 257             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 258
 259             /* Update potential sum for this i atom from the interaction with this j atom. */
 260             velec            = _mm_and_pd(velec,cutoff_mask);
 261             velecsum         = _mm_add_pd(velecsum,velec);
 262             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 263             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 264
 265             fscal            = _mm_add_pd(felec,fvdw);
 266
 267             fscal            = _mm_and_pd(fscal,cutoff_mask);
 268
 269             /* Calculate temporary vectorial force */
 270             tx               = _mm_mul_pd(fscal,dx00);
 271             ty               = _mm_mul_pd(fscal,dy00);
 272             tz               = _mm_mul_pd(fscal,dz00);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm_add_pd(fix0,tx);
 276             fiy0             = _mm_add_pd(fiy0,ty);
 277             fiz0             = _mm_add_pd(fiz0,tz);
 278
 279             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 280
 281             }
 282
 283             /* Inner loop uses 82 flops */
 284         }
 285
 286         if(jidx<j_index_end)
 287         {
 288
 289             jnrA             = jjnr[jidx];
 290             j_coord_offsetA  = DIM*jnrA;
 291
 292             /* load j atom coordinates */
 293             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 294                                               &jx0,&jy0,&jz0);
 295
 296             /* Calculate displacement vector */
 297             dx00             = _mm_sub_pd(ix0,jx0);
 298             dy00             = _mm_sub_pd(iy0,jy0);
 299             dz00             = _mm_sub_pd(iz0,jz0);
 300
 301             /* Calculate squared distance and things based on it */
 302             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 303
 304             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 305
 306             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 307
 308             /* Load parameters for j particles */
 309             jq0              = _mm_load_sd(charge+jnrA+0);
 310             vdwjidx0A        = 2*vdwtype[jnrA+0];
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm_any_lt(rsq00,rcutoff2))
 317             {
 318
 319             r00              = _mm_mul_pd(rsq00,rinv00);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq00             = _mm_mul_pd(iq0,jq0);
 323             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 324
 325             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 330             ewrt             = _mm_mul_pd(r00,ewtabscale);
 331             ewitab           = _mm_cvttpd_epi32(ewrt);
 332             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 333             ewitab           = _mm_slli_epi32(ewitab,2);
 334             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 335             ewtabD           = _mm_setzero_pd();
 336             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 337             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 338             ewtabFn          = _mm_setzero_pd();
 339             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 340             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 341             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 342             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 343             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 344
 345             /* Analytical LJ-PME */
 346             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 347             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 348             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 349             exponent         = gmx_simd_exp_d(ewcljrsq);
 350             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 351             poly             = _mm_mul_pd(exponent,_mm_add_pd(_mm_sub_pd(one,ewcljrsq),_mm_mul_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 352             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 353             vvdw6            = _mm_mul_pd(_mm_sub_pd(c6_00,_mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly))),rinvsix);
 354             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 355             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),
 356                                _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));
 357             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 358             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);
 359
 360             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velec            = _mm_and_pd(velec,cutoff_mask);
 364             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 365             velecsum         = _mm_add_pd(velecsum,velec);
 366             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 367             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 368             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 369
 370             fscal            = _mm_add_pd(felec,fvdw);
 371
 372             fscal            = _mm_and_pd(fscal,cutoff_mask);
 373
 374             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm_mul_pd(fscal,dx00);
 378             ty               = _mm_mul_pd(fscal,dy00);
 379             tz               = _mm_mul_pd(fscal,dz00);
 380
 381             /* Update vectorial force */
 382             fix0             = _mm_add_pd(fix0,tx);
 383             fiy0             = _mm_add_pd(fiy0,ty);
 384             fiz0             = _mm_add_pd(fiz0,tz);
 385
 386             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 387
 388             }
 389
 390             /* Inner loop uses 82 flops */
 391         }
 392
 393         /* End of innermost loop */
 394
 395         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 396                                               f+i_coord_offset,fshift+i_shift_offset);
 397
 398         ggid                        = gid[iidx];
 399         /* Update potential energies */
 400         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 401         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 402
 403         /* Increment number of inner iterations */
 404         inneriter                  += j_index_end - j_index_start;
 405
 406         /* Outer loop uses 9 flops */
 407     }
 408
 409     /* Increment number of outer iterations */
 410     outeriter        += nri;
 411
 412     /* Update outer/inner flops */
 413
 414     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*82);
 415 }
 416 /*
 417  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse2_double
 418  * Electrostatics interaction: Ewald
 419  * VdW interaction:            LJEwald
 420  * Geometry:                   Particle-Particle
 421  * Calculate force/pot:        Force
 422  */
 423 void
 424 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sse2_double
 425                     (t_nblist                    * gmx_restrict       nlist,
 426                      rvec                        * gmx_restrict          xx,
 427                      rvec                        * gmx_restrict          ff,
 428                      t_forcerec                  * gmx_restrict          fr,
 429                      t_mdatoms                   * gmx_restrict     mdatoms,
 430                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 431                      t_nrnb                      * gmx_restrict        nrnb)
 432 {
 433     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 434      * just 0 for non-waters.
 435      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 436      * jnr indices corresponding to data put in the four positions in the SIMD register.
 437      */
 438     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 439     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 440     int              jnrA,jnrB;
 441     int              j_coord_offsetA,j_coord_offsetB;
 442     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 443     real             rcutoff_scalar;
 444     real             *shiftvec,*fshift,*x,*f;
 445     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 446     int              vdwioffset0;
 447     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 448     int              vdwjidx0A,vdwjidx0B;
 449     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 450     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 451     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 452     real             *charge;
 453     int              nvdwtype;
 454     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 455     int              *vdwtype;
 456     real             *vdwparam;
 457     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 458     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 459     __m128d           c6grid_00;
 460     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 461     real             *vdwgridparam;
 462     __m128d           one_half = _mm_set1_pd(0.5);
 463     __m128d           minus_one = _mm_set1_pd(-1.0);
 464     __m128i          ewitab;
 465     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 466     real             *ewtab;
 467     __m128d          dummy_mask,cutoff_mask;
 468     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 469     __m128d          one     = _mm_set1_pd(1.0);
 470     __m128d          two     = _mm_set1_pd(2.0);
 471     x                = xx[0];
 472     f                = ff[0];
 473
 474     nri              = nlist->nri;
 475     iinr             = nlist->iinr;
 476     jindex           = nlist->jindex;
 477     jjnr             = nlist->jjnr;
 478     shiftidx         = nlist->shift;
 479     gid              = nlist->gid;
 480     shiftvec         = fr->shift_vec[0];
 481     fshift           = fr->fshift[0];
 482     facel            = _mm_set1_pd(fr->epsfac);
 483     charge           = mdatoms->chargeA;
 484     nvdwtype         = fr->ntype;
 485     vdwparam         = fr->nbfp;
 486     vdwtype          = mdatoms->typeA;
 487     vdwgridparam     = fr->ljpme_c6grid;
 488     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 489     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 490     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 491
 492     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 493     ewtab            = fr->ic->tabq_coul_F;
 494     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 495     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 496
 497     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 498     rcutoff_scalar   = fr->rcoulomb;
 499     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 500     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 501
 502     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 503     rvdw             = _mm_set1_pd(fr->rvdw);
 504
 505     /* Avoid stupid compiler warnings */
 506     jnrA = jnrB = 0;
 507     j_coord_offsetA = 0;
 508     j_coord_offsetB = 0;
 509
 510     outeriter        = 0;
 511     inneriter        = 0;
 512
 513     /* Start outer loop over neighborlists */
 514     for(iidx=0; iidx<nri; iidx++)
 515     {
 516         /* Load shift vector for this list */
 517         i_shift_offset   = DIM*shiftidx[iidx];
 518
 519         /* Load limits for loop over neighbors */
 520         j_index_start    = jindex[iidx];
 521         j_index_end      = jindex[iidx+1];
 522
 523         /* Get outer coordinate index */
 524         inr              = iinr[iidx];
 525         i_coord_offset   = DIM*inr;
 526
 527         /* Load i particle coords and add shift vector */
 528         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 529
 530         fix0             = _mm_setzero_pd();
 531         fiy0             = _mm_setzero_pd();
 532         fiz0             = _mm_setzero_pd();
 533
 534         /* Load parameters for i particles */
 535         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 536         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 537
 538         /* Start inner kernel loop */
 539         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 540         {
 541
 542             /* Get j neighbor index, and coordinate index */
 543             jnrA             = jjnr[jidx];
 544             jnrB             = jjnr[jidx+1];
 545             j_coord_offsetA  = DIM*jnrA;
 546             j_coord_offsetB  = DIM*jnrB;
 547
 548             /* load j atom coordinates */
 549             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 550                                               &jx0,&jy0,&jz0);
 551
 552             /* Calculate displacement vector */
 553             dx00             = _mm_sub_pd(ix0,jx0);
 554             dy00             = _mm_sub_pd(iy0,jy0);
 555             dz00             = _mm_sub_pd(iz0,jz0);
 556
 557             /* Calculate squared distance and things based on it */
 558             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 559
 560             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 561
 562             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 563
 564             /* Load parameters for j particles */
 565             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 566             vdwjidx0A        = 2*vdwtype[jnrA+0];
 567             vdwjidx0B        = 2*vdwtype[jnrB+0];
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             if (gmx_mm_any_lt(rsq00,rcutoff2))
 574             {
 575
 576             r00              = _mm_mul_pd(rsq00,rinv00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq00             = _mm_mul_pd(iq0,jq0);
 580             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 581                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 582
 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_cvtepi32_pd(ewitab));
 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_cvtepi32_pd(ewitab));
 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 }