src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwCSTab_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_sse2_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_sse2_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128i          ewitab;
 100     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 101     real             *ewtab;
 102     __m128d          dummy_mask,cutoff_mask;
 103     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 104     __m128d          one     = _mm_set1_pd(1.0);
 105     __m128d          two     = _mm_set1_pd(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_pd(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 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     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _mm_setzero_pd();
 157         fiy0             = _mm_setzero_pd();
 158         fiz0             = _mm_setzero_pd();
 159
 160         /* Load parameters for i particles */
 161         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 162         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_pd();
 166         vvdwsum          = _mm_setzero_pd();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_pd(ix0,jx0);
 184             dy00             = _mm_sub_pd(iy0,jy0);
 185             dz00             = _mm_sub_pd(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 189
 190             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 191
 192             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 196             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             r00              = _mm_mul_pd(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _mm_mul_pd(iq0,jq0);
 207             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 208                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 209
 210             /* Calculate table index by multiplying r with table scale and truncate to integer */
 211             rt               = _mm_mul_pd(r00,vftabscale);
 212             vfitab           = _mm_cvttpd_epi32(rt);
 213             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 214             vfitab           = _mm_slli_epi32(vfitab,3);
 215
 216             /* EWALD ELECTROSTATICS */
 217
 218             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 219             ewrt             = _mm_mul_pd(r00,ewtabscale);
 220             ewitab           = _mm_cvttpd_epi32(ewrt);
 221             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 222             ewitab           = _mm_slli_epi32(ewitab,2);
 223             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 224             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 225             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 226             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 227             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 228             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 229             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 230             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 231             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 232             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 233
 234             /* CUBIC SPLINE TABLE DISPERSION */
 235             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 236             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 237             GMX_MM_TRANSPOSE2_PD(Y,F);
 238             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 239             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 240             GMX_MM_TRANSPOSE2_PD(G,H);
 241             Heps             = _mm_mul_pd(vfeps,H);
 242             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 243             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 244             vvdw6            = _mm_mul_pd(c6_00,VV);
 245             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 246             fvdw6            = _mm_mul_pd(c6_00,FF);
 247
 248             /* CUBIC SPLINE TABLE REPULSION */
 249             vfitab           = _mm_add_epi32(vfitab,ifour);
 250             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 251             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 252             GMX_MM_TRANSPOSE2_PD(Y,F);
 253             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 254             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 255             GMX_MM_TRANSPOSE2_PD(G,H);
 256             Heps             = _mm_mul_pd(vfeps,H);
 257             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 258             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 259             vvdw12           = _mm_mul_pd(c12_00,VV);
 260             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 261             fvdw12           = _mm_mul_pd(c12_00,FF);
 262             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 263             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             velecsum         = _mm_add_pd(velecsum,velec);
 267             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 268
 269             fscal            = _mm_add_pd(felec,fvdw);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_pd(fscal,dx00);
 273             ty               = _mm_mul_pd(fscal,dy00);
 274             tz               = _mm_mul_pd(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_pd(fix0,tx);
 278             fiy0             = _mm_add_pd(fiy0,ty);
 279             fiz0             = _mm_add_pd(fiz0,tz);
 280
 281             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 282
 283             /* Inner loop uses 75 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             r00              = _mm_mul_pd(rsq00,rinv00);
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq00             = _mm_mul_pd(iq0,jq0);
 320             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 321
 322             /* Calculate table index by multiplying r with table scale and truncate to integer */
 323             rt               = _mm_mul_pd(r00,vftabscale);
 324             vfitab           = _mm_cvttpd_epi32(rt);
 325             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 326             vfitab           = _mm_slli_epi32(vfitab,3);
 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_cvtepi32_pd(ewitab));
 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(rinv00,velec));
 344             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 345
 346             /* CUBIC SPLINE TABLE DISPERSION */
 347             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 348             F                = _mm_setzero_pd();
 349             GMX_MM_TRANSPOSE2_PD(Y,F);
 350             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 351             H                = _mm_setzero_pd();
 352             GMX_MM_TRANSPOSE2_PD(G,H);
 353             Heps             = _mm_mul_pd(vfeps,H);
 354             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 355             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 356             vvdw6            = _mm_mul_pd(c6_00,VV);
 357             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 358             fvdw6            = _mm_mul_pd(c6_00,FF);
 359
 360             /* CUBIC SPLINE TABLE REPULSION */
 361             vfitab           = _mm_add_epi32(vfitab,ifour);
 362             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 363             F                = _mm_setzero_pd();
 364             GMX_MM_TRANSPOSE2_PD(Y,F);
 365             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 366             H                = _mm_setzero_pd();
 367             GMX_MM_TRANSPOSE2_PD(G,H);
 368             Heps             = _mm_mul_pd(vfeps,H);
 369             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 370             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 371             vvdw12           = _mm_mul_pd(c12_00,VV);
 372             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 373             fvdw12           = _mm_mul_pd(c12_00,FF);
 374             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 375             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 379             velecsum         = _mm_add_pd(velecsum,velec);
 380             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 381             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 382
 383             fscal            = _mm_add_pd(felec,fvdw);
 384
 385             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 386
 387             /* Calculate temporary vectorial force */
 388             tx               = _mm_mul_pd(fscal,dx00);
 389             ty               = _mm_mul_pd(fscal,dy00);
 390             tz               = _mm_mul_pd(fscal,dz00);
 391
 392             /* Update vectorial force */
 393             fix0             = _mm_add_pd(fix0,tx);
 394             fiy0             = _mm_add_pd(fiy0,ty);
 395             fiz0             = _mm_add_pd(fiz0,tz);
 396
 397             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 398
 399             /* Inner loop uses 75 flops */
 400         }
 401
 402         /* End of innermost loop */
 403
 404         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 405                                               f+i_coord_offset,fshift+i_shift_offset);
 406
 407         ggid                        = gid[iidx];
 408         /* Update potential energies */
 409         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 410         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 411
 412         /* Increment number of inner iterations */
 413         inneriter                  += j_index_end - j_index_start;
 414
 415         /* Outer loop uses 9 flops */
 416     }
 417
 418     /* Increment number of outer iterations */
 419     outeriter        += nri;
 420
 421     /* Update outer/inner flops */
 422
 423     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*75);
 424 }
 425 /*
 426  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse2_double
 427  * Electrostatics interaction: Ewald
 428  * VdW interaction:            CubicSplineTable
 429  * Geometry:                   Particle-Particle
 430  * Calculate force/pot:        Force
 431  */
 432 void
 433 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse2_double
 434                     (t_nblist                    * gmx_restrict       nlist,
 435                      rvec                        * gmx_restrict          xx,
 436                      rvec                        * gmx_restrict          ff,
 437                      t_forcerec                  * gmx_restrict          fr,
 438                      t_mdatoms                   * gmx_restrict     mdatoms,
 439                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 440                      t_nrnb                      * gmx_restrict        nrnb)
 441 {
 442     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 443      * just 0 for non-waters.
 444      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 445      * jnr indices corresponding to data put in the four positions in the SIMD register.
 446      */
 447     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 448     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 449     int              jnrA,jnrB;
 450     int              j_coord_offsetA,j_coord_offsetB;
 451     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 452     real             rcutoff_scalar;
 453     real             *shiftvec,*fshift,*x,*f;
 454     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 455     int              vdwioffset0;
 456     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 457     int              vdwjidx0A,vdwjidx0B;
 458     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 459     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 460     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 461     real             *charge;
 462     int              nvdwtype;
 463     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 464     int              *vdwtype;
 465     real             *vdwparam;
 466     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 467     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 468     __m128i          vfitab;
 469     __m128i          ifour       = _mm_set1_epi32(4);
 470     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 471     real             *vftab;
 472     __m128i          ewitab;
 473     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 474     real             *ewtab;
 475     __m128d          dummy_mask,cutoff_mask;
 476     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 477     __m128d          one     = _mm_set1_pd(1.0);
 478     __m128d          two     = _mm_set1_pd(2.0);
 479     x                = xx[0];
 480     f                = ff[0];
 481
 482     nri              = nlist->nri;
 483     iinr             = nlist->iinr;
 484     jindex           = nlist->jindex;
 485     jjnr             = nlist->jjnr;
 486     shiftidx         = nlist->shift;
 487     gid              = nlist->gid;
 488     shiftvec         = fr->shift_vec[0];
 489     fshift           = fr->fshift[0];
 490     facel            = _mm_set1_pd(fr->epsfac);
 491     charge           = mdatoms->chargeA;
 492     nvdwtype         = fr->ntype;
 493     vdwparam         = fr->nbfp;
 494     vdwtype          = mdatoms->typeA;
 495
 496     vftab            = kernel_data->table_vdw->data;
 497     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 498
 499     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 500     ewtab            = fr->ic->tabq_coul_F;
 501     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 502     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 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             r00              = _mm_mul_pd(rsq00,rinv00);
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             qq00             = _mm_mul_pd(iq0,jq0);
 576             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 577                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 578
 579             /* Calculate table index by multiplying r with table scale and truncate to integer */
 580             rt               = _mm_mul_pd(r00,vftabscale);
 581             vfitab           = _mm_cvttpd_epi32(rt);
 582             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 583             vfitab           = _mm_slli_epi32(vfitab,3);
 584
 585             /* EWALD ELECTROSTATICS */
 586
 587             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 588             ewrt             = _mm_mul_pd(r00,ewtabscale);
 589             ewitab           = _mm_cvttpd_epi32(ewrt);
 590             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 591             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 592                                          &ewtabF,&ewtabFn);
 593             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 594             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 595
 596             /* CUBIC SPLINE TABLE DISPERSION */
 597             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 598             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 599             GMX_MM_TRANSPOSE2_PD(Y,F);
 600             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 601             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 602             GMX_MM_TRANSPOSE2_PD(G,H);
 603             Heps             = _mm_mul_pd(vfeps,H);
 604             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 605             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 606             fvdw6            = _mm_mul_pd(c6_00,FF);
 607
 608             /* CUBIC SPLINE TABLE REPULSION */
 609             vfitab           = _mm_add_epi32(vfitab,ifour);
 610             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 611             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 612             GMX_MM_TRANSPOSE2_PD(Y,F);
 613             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 614             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 615             GMX_MM_TRANSPOSE2_PD(G,H);
 616             Heps             = _mm_mul_pd(vfeps,H);
 617             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 618             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 619             fvdw12           = _mm_mul_pd(c12_00,FF);
 620             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 621
 622             fscal            = _mm_add_pd(felec,fvdw);
 623
 624             /* Calculate temporary vectorial force */
 625             tx               = _mm_mul_pd(fscal,dx00);
 626             ty               = _mm_mul_pd(fscal,dy00);
 627             tz               = _mm_mul_pd(fscal,dz00);
 628
 629             /* Update vectorial force */
 630             fix0             = _mm_add_pd(fix0,tx);
 631             fiy0             = _mm_add_pd(fiy0,ty);
 632             fiz0             = _mm_add_pd(fiz0,tz);
 633
 634             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 635
 636             /* Inner loop uses 62 flops */
 637         }
 638
 639         if(jidx<j_index_end)
 640         {
 641
 642             jnrA             = jjnr[jidx];
 643             j_coord_offsetA  = DIM*jnrA;
 644
 645             /* load j atom coordinates */
 646             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 647                                               &jx0,&jy0,&jz0);
 648
 649             /* Calculate displacement vector */
 650             dx00             = _mm_sub_pd(ix0,jx0);
 651             dy00             = _mm_sub_pd(iy0,jy0);
 652             dz00             = _mm_sub_pd(iz0,jz0);
 653
 654             /* Calculate squared distance and things based on it */
 655             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 656
 657             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 658
 659             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 660
 661             /* Load parameters for j particles */
 662             jq0              = _mm_load_sd(charge+jnrA+0);
 663             vdwjidx0A        = 2*vdwtype[jnrA+0];
 664
 665             /**************************
 666              * CALCULATE INTERACTIONS *
 667              **************************/
 668
 669             r00              = _mm_mul_pd(rsq00,rinv00);
 670
 671             /* Compute parameters for interactions between i and j atoms */
 672             qq00             = _mm_mul_pd(iq0,jq0);
 673             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 674
 675             /* Calculate table index by multiplying r with table scale and truncate to integer */
 676             rt               = _mm_mul_pd(r00,vftabscale);
 677             vfitab           = _mm_cvttpd_epi32(rt);
 678             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 679             vfitab           = _mm_slli_epi32(vfitab,3);
 680
 681             /* EWALD ELECTROSTATICS */
 682
 683             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 684             ewrt             = _mm_mul_pd(r00,ewtabscale);
 685             ewitab           = _mm_cvttpd_epi32(ewrt);
 686             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 687             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 688             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 689             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 690
 691             /* CUBIC SPLINE TABLE DISPERSION */
 692             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 693             F                = _mm_setzero_pd();
 694             GMX_MM_TRANSPOSE2_PD(Y,F);
 695             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 696             H                = _mm_setzero_pd();
 697             GMX_MM_TRANSPOSE2_PD(G,H);
 698             Heps             = _mm_mul_pd(vfeps,H);
 699             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 700             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 701             fvdw6            = _mm_mul_pd(c6_00,FF);
 702
 703             /* CUBIC SPLINE TABLE REPULSION */
 704             vfitab           = _mm_add_epi32(vfitab,ifour);
 705             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 706             F                = _mm_setzero_pd();
 707             GMX_MM_TRANSPOSE2_PD(Y,F);
 708             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 709             H                = _mm_setzero_pd();
 710             GMX_MM_TRANSPOSE2_PD(G,H);
 711             Heps             = _mm_mul_pd(vfeps,H);
 712             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 713             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 714             fvdw12           = _mm_mul_pd(c12_00,FF);
 715             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 716
 717             fscal            = _mm_add_pd(felec,fvdw);
 718
 719             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 720
 721             /* Calculate temporary vectorial force */
 722             tx               = _mm_mul_pd(fscal,dx00);
 723             ty               = _mm_mul_pd(fscal,dy00);
 724             tz               = _mm_mul_pd(fscal,dz00);
 725
 726             /* Update vectorial force */
 727             fix0             = _mm_add_pd(fix0,tx);
 728             fiy0             = _mm_add_pd(fiy0,ty);
 729             fiz0             = _mm_add_pd(fiz0,tz);
 730
 731             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 732
 733             /* Inner loop uses 62 flops */
 734         }
 735
 736         /* End of innermost loop */
 737
 738         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 739                                               f+i_coord_offset,fshift+i_shift_offset);
 740
 741         /* Increment number of inner iterations */
 742         inneriter                  += j_index_end - j_index_start;
 743
 744         /* Outer loop uses 7 flops */
 745     }
 746
 747     /* Increment number of outer iterations */
 748     outeriter        += nri;
 749
 750     /* Update outer/inner flops */
 751
 752     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 753 }