src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwCSTab_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_ElecEw_VdwCSTab_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_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     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  96     real             *vftab;
  97     __m128i          ewitab;
  98     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  99     real             *ewtab;
 100     __m128d          dummy_mask,cutoff_mask;
 101     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 102     __m128d          one     = _mm_set1_pd(1.0);
 103     __m128d          two     = _mm_set1_pd(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_pd(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     vftab            = kernel_data->table_vdw->data;
 122     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 123
 124     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm_setzero_pd();
 155         fiy0             = _mm_setzero_pd();
 156         fiz0             = _mm_setzero_pd();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 187
 188             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 189
 190             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             r00              = _mm_mul_pd(rsq00,rinv00);
 202
 203             /* Compute parameters for interactions between i and j atoms */
 204             qq00             = _mm_mul_pd(iq0,jq0);
 205             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 206                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 207
 208             /* Calculate table index by multiplying r with table scale and truncate to integer */
 209             rt               = _mm_mul_pd(r00,vftabscale);
 210             vfitab           = _mm_cvttpd_epi32(rt);
 211             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 212             vfitab           = _mm_slli_epi32(vfitab,3);
 213
 214             /* EWALD ELECTROSTATICS */
 215
 216             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 217             ewrt             = _mm_mul_pd(r00,ewtabscale);
 218             ewitab           = _mm_cvttpd_epi32(ewrt);
 219             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 220             ewitab           = _mm_slli_epi32(ewitab,2);
 221             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 222             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 223             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 224             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 225             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 226             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 227             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 228             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 229             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 230             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 231
 232             /* CUBIC SPLINE TABLE DISPERSION */
 233             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 234             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 235             GMX_MM_TRANSPOSE2_PD(Y,F);
 236             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 237             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 238             GMX_MM_TRANSPOSE2_PD(G,H);
 239             Heps             = _mm_mul_pd(vfeps,H);
 240             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 241             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 242             vvdw6            = _mm_mul_pd(c6_00,VV);
 243             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 244             fvdw6            = _mm_mul_pd(c6_00,FF);
 245
 246             /* CUBIC SPLINE TABLE REPULSION */
 247             vfitab           = _mm_add_epi32(vfitab,ifour);
 248             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 249             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 250             GMX_MM_TRANSPOSE2_PD(Y,F);
 251             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 252             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 253             GMX_MM_TRANSPOSE2_PD(G,H);
 254             Heps             = _mm_mul_pd(vfeps,H);
 255             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 256             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 257             vvdw12           = _mm_mul_pd(c12_00,VV);
 258             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 259             fvdw12           = _mm_mul_pd(c12_00,FF);
 260             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 261             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velecsum         = _mm_add_pd(velecsum,velec);
 265             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = _mm_add_pd(felec,fvdw);
 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             /* Inner loop uses 75 flops */
 282         }
 283
 284         if(jidx<j_index_end)
 285         {
 286
 287             jnrA             = jjnr[jidx];
 288             j_coord_offsetA  = DIM*jnrA;
 289
 290             /* load j atom coordinates */
 291             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 292                                               &jx0,&jy0,&jz0);
 293
 294             /* Calculate displacement vector */
 295             dx00             = _mm_sub_pd(ix0,jx0);
 296             dy00             = _mm_sub_pd(iy0,jy0);
 297             dz00             = _mm_sub_pd(iz0,jz0);
 298
 299             /* Calculate squared distance and things based on it */
 300             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 301
 302             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 303
 304             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 305
 306             /* Load parameters for j particles */
 307             jq0              = _mm_load_sd(charge+jnrA+0);
 308             vdwjidx0A        = 2*vdwtype[jnrA+0];
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             r00              = _mm_mul_pd(rsq00,rinv00);
 315
 316             /* Compute parameters for interactions between i and j atoms */
 317             qq00             = _mm_mul_pd(iq0,jq0);
 318             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 319
 320             /* Calculate table index by multiplying r with table scale and truncate to integer */
 321             rt               = _mm_mul_pd(r00,vftabscale);
 322             vfitab           = _mm_cvttpd_epi32(rt);
 323             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 324             vfitab           = _mm_slli_epi32(vfitab,3);
 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(rinv00,velec));
 342             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 343
 344             /* CUBIC SPLINE TABLE DISPERSION */
 345             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 346             F                = _mm_setzero_pd();
 347             GMX_MM_TRANSPOSE2_PD(Y,F);
 348             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 349             H                = _mm_setzero_pd();
 350             GMX_MM_TRANSPOSE2_PD(G,H);
 351             Heps             = _mm_mul_pd(vfeps,H);
 352             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 353             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 354             vvdw6            = _mm_mul_pd(c6_00,VV);
 355             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 356             fvdw6            = _mm_mul_pd(c6_00,FF);
 357
 358             /* CUBIC SPLINE TABLE REPULSION */
 359             vfitab           = _mm_add_epi32(vfitab,ifour);
 360             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 361             F                = _mm_setzero_pd();
 362             GMX_MM_TRANSPOSE2_PD(Y,F);
 363             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 364             H                = _mm_setzero_pd();
 365             GMX_MM_TRANSPOSE2_PD(G,H);
 366             Heps             = _mm_mul_pd(vfeps,H);
 367             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 368             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 369             vvdw12           = _mm_mul_pd(c12_00,VV);
 370             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 371             fvdw12           = _mm_mul_pd(c12_00,FF);
 372             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 373             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 377             velecsum         = _mm_add_pd(velecsum,velec);
 378             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 379             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 380
 381             fscal            = _mm_add_pd(felec,fvdw);
 382
 383             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm_mul_pd(fscal,dx00);
 387             ty               = _mm_mul_pd(fscal,dy00);
 388             tz               = _mm_mul_pd(fscal,dz00);
 389
 390             /* Update vectorial force */
 391             fix0             = _mm_add_pd(fix0,tx);
 392             fiy0             = _mm_add_pd(fiy0,ty);
 393             fiz0             = _mm_add_pd(fiz0,tz);
 394
 395             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 396
 397             /* Inner loop uses 75 flops */
 398         }
 399
 400         /* End of innermost loop */
 401
 402         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 403                                               f+i_coord_offset,fshift+i_shift_offset);
 404
 405         ggid                        = gid[iidx];
 406         /* Update potential energies */
 407         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 408         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 409
 410         /* Increment number of inner iterations */
 411         inneriter                  += j_index_end - j_index_start;
 412
 413         /* Outer loop uses 9 flops */
 414     }
 415
 416     /* Increment number of outer iterations */
 417     outeriter        += nri;
 418
 419     /* Update outer/inner flops */
 420
 421     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*75);
 422 }
 423 /*
 424  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse4_1_double
 425  * Electrostatics interaction: Ewald
 426  * VdW interaction:            CubicSplineTable
 427  * Geometry:                   Particle-Particle
 428  * Calculate force/pot:        Force
 429  */
 430 void
 431 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse4_1_double
 432                     (t_nblist                    * gmx_restrict       nlist,
 433                      rvec                        * gmx_restrict          xx,
 434                      rvec                        * gmx_restrict          ff,
 435                      t_forcerec                  * gmx_restrict          fr,
 436                      t_mdatoms                   * gmx_restrict     mdatoms,
 437                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 438                      t_nrnb                      * gmx_restrict        nrnb)
 439 {
 440     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 441      * just 0 for non-waters.
 442      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 443      * jnr indices corresponding to data put in the four positions in the SIMD register.
 444      */
 445     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 446     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 447     int              jnrA,jnrB;
 448     int              j_coord_offsetA,j_coord_offsetB;
 449     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 450     real             rcutoff_scalar;
 451     real             *shiftvec,*fshift,*x,*f;
 452     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 453     int              vdwioffset0;
 454     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 455     int              vdwjidx0A,vdwjidx0B;
 456     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 457     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 458     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 459     real             *charge;
 460     int              nvdwtype;
 461     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 462     int              *vdwtype;
 463     real             *vdwparam;
 464     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 465     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 466     __m128i          vfitab;
 467     __m128i          ifour       = _mm_set1_epi32(4);
 468     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 469     real             *vftab;
 470     __m128i          ewitab;
 471     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 472     real             *ewtab;
 473     __m128d          dummy_mask,cutoff_mask;
 474     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 475     __m128d          one     = _mm_set1_pd(1.0);
 476     __m128d          two     = _mm_set1_pd(2.0);
 477     x                = xx[0];
 478     f                = ff[0];
 479
 480     nri              = nlist->nri;
 481     iinr             = nlist->iinr;
 482     jindex           = nlist->jindex;
 483     jjnr             = nlist->jjnr;
 484     shiftidx         = nlist->shift;
 485     gid              = nlist->gid;
 486     shiftvec         = fr->shift_vec[0];
 487     fshift           = fr->fshift[0];
 488     facel            = _mm_set1_pd(fr->epsfac);
 489     charge           = mdatoms->chargeA;
 490     nvdwtype         = fr->ntype;
 491     vdwparam         = fr->nbfp;
 492     vdwtype          = mdatoms->typeA;
 493
 494     vftab            = kernel_data->table_vdw->data;
 495     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 496
 497     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 498     ewtab            = fr->ic->tabq_coul_F;
 499     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 500     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 501
 502     /* Avoid stupid compiler warnings */
 503     jnrA = jnrB = 0;
 504     j_coord_offsetA = 0;
 505     j_coord_offsetB = 0;
 506
 507     outeriter        = 0;
 508     inneriter        = 0;
 509
 510     /* Start outer loop over neighborlists */
 511     for(iidx=0; iidx<nri; iidx++)
 512     {
 513         /* Load shift vector for this list */
 514         i_shift_offset   = DIM*shiftidx[iidx];
 515
 516         /* Load limits for loop over neighbors */
 517         j_index_start    = jindex[iidx];
 518         j_index_end      = jindex[iidx+1];
 519
 520         /* Get outer coordinate index */
 521         inr              = iinr[iidx];
 522         i_coord_offset   = DIM*inr;
 523
 524         /* Load i particle coords and add shift vector */
 525         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 526
 527         fix0             = _mm_setzero_pd();
 528         fiy0             = _mm_setzero_pd();
 529         fiz0             = _mm_setzero_pd();
 530
 531         /* Load parameters for i particles */
 532         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 533         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 534
 535         /* Start inner kernel loop */
 536         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 537         {
 538
 539             /* Get j neighbor index, and coordinate index */
 540             jnrA             = jjnr[jidx];
 541             jnrB             = jjnr[jidx+1];
 542             j_coord_offsetA  = DIM*jnrA;
 543             j_coord_offsetB  = DIM*jnrB;
 544
 545             /* load j atom coordinates */
 546             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 547                                               &jx0,&jy0,&jz0);
 548
 549             /* Calculate displacement vector */
 550             dx00             = _mm_sub_pd(ix0,jx0);
 551             dy00             = _mm_sub_pd(iy0,jy0);
 552             dz00             = _mm_sub_pd(iz0,jz0);
 553
 554             /* Calculate squared distance and things based on it */
 555             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 556
 557             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 558
 559             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 560
 561             /* Load parameters for j particles */
 562             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 563             vdwjidx0A        = 2*vdwtype[jnrA+0];
 564             vdwjidx0B        = 2*vdwtype[jnrB+0];
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r00              = _mm_mul_pd(rsq00,rinv00);
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq00             = _mm_mul_pd(iq0,jq0);
 574             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 575                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 576
 577             /* Calculate table index by multiplying r with table scale and truncate to integer */
 578             rt               = _mm_mul_pd(r00,vftabscale);
 579             vfitab           = _mm_cvttpd_epi32(rt);
 580             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 581             vfitab           = _mm_slli_epi32(vfitab,3);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 586             ewrt             = _mm_mul_pd(r00,ewtabscale);
 587             ewitab           = _mm_cvttpd_epi32(ewrt);
 588             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 589             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 590                                          &ewtabF,&ewtabFn);
 591             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 592             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 593
 594             /* CUBIC SPLINE TABLE DISPERSION */
 595             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 596             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 597             GMX_MM_TRANSPOSE2_PD(Y,F);
 598             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 599             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 600             GMX_MM_TRANSPOSE2_PD(G,H);
 601             Heps             = _mm_mul_pd(vfeps,H);
 602             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 603             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 604             fvdw6            = _mm_mul_pd(c6_00,FF);
 605
 606             /* CUBIC SPLINE TABLE REPULSION */
 607             vfitab           = _mm_add_epi32(vfitab,ifour);
 608             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 609             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 610             GMX_MM_TRANSPOSE2_PD(Y,F);
 611             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 612             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 613             GMX_MM_TRANSPOSE2_PD(G,H);
 614             Heps             = _mm_mul_pd(vfeps,H);
 615             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 616             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 617             fvdw12           = _mm_mul_pd(c12_00,FF);
 618             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 619
 620             fscal            = _mm_add_pd(felec,fvdw);
 621
 622             /* Calculate temporary vectorial force */
 623             tx               = _mm_mul_pd(fscal,dx00);
 624             ty               = _mm_mul_pd(fscal,dy00);
 625             tz               = _mm_mul_pd(fscal,dz00);
 626
 627             /* Update vectorial force */
 628             fix0             = _mm_add_pd(fix0,tx);
 629             fiy0             = _mm_add_pd(fiy0,ty);
 630             fiz0             = _mm_add_pd(fiz0,tz);
 631
 632             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 633
 634             /* Inner loop uses 62 flops */
 635         }
 636
 637         if(jidx<j_index_end)
 638         {
 639
 640             jnrA             = jjnr[jidx];
 641             j_coord_offsetA  = DIM*jnrA;
 642
 643             /* load j atom coordinates */
 644             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 645                                               &jx0,&jy0,&jz0);
 646
 647             /* Calculate displacement vector */
 648             dx00             = _mm_sub_pd(ix0,jx0);
 649             dy00             = _mm_sub_pd(iy0,jy0);
 650             dz00             = _mm_sub_pd(iz0,jz0);
 651
 652             /* Calculate squared distance and things based on it */
 653             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 654
 655             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 656
 657             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 658
 659             /* Load parameters for j particles */
 660             jq0              = _mm_load_sd(charge+jnrA+0);
 661             vdwjidx0A        = 2*vdwtype[jnrA+0];
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 674             rt               = _mm_mul_pd(r00,vftabscale);
 675             vfitab           = _mm_cvttpd_epi32(rt);
 676             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 677             vfitab           = _mm_slli_epi32(vfitab,3);
 678
 679             /* EWALD ELECTROSTATICS */
 680
 681             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 682             ewrt             = _mm_mul_pd(r00,ewtabscale);
 683             ewitab           = _mm_cvttpd_epi32(ewrt);
 684             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 685             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 686             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 687             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 688
 689             /* CUBIC SPLINE TABLE DISPERSION */
 690             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 691             F                = _mm_setzero_pd();
 692             GMX_MM_TRANSPOSE2_PD(Y,F);
 693             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 694             H                = _mm_setzero_pd();
 695             GMX_MM_TRANSPOSE2_PD(G,H);
 696             Heps             = _mm_mul_pd(vfeps,H);
 697             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 698             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 699             fvdw6            = _mm_mul_pd(c6_00,FF);
 700
 701             /* CUBIC SPLINE TABLE REPULSION */
 702             vfitab           = _mm_add_epi32(vfitab,ifour);
 703             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 704             F                = _mm_setzero_pd();
 705             GMX_MM_TRANSPOSE2_PD(Y,F);
 706             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 707             H                = _mm_setzero_pd();
 708             GMX_MM_TRANSPOSE2_PD(G,H);
 709             Heps             = _mm_mul_pd(vfeps,H);
 710             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 711             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 712             fvdw12           = _mm_mul_pd(c12_00,FF);
 713             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 714
 715             fscal            = _mm_add_pd(felec,fvdw);
 716
 717             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 718
 719             /* Calculate temporary vectorial force */
 720             tx               = _mm_mul_pd(fscal,dx00);
 721             ty               = _mm_mul_pd(fscal,dy00);
 722             tz               = _mm_mul_pd(fscal,dz00);
 723
 724             /* Update vectorial force */
 725             fix0             = _mm_add_pd(fix0,tx);
 726             fiy0             = _mm_add_pd(fiy0,ty);
 727             fiz0             = _mm_add_pd(fiz0,tz);
 728
 729             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 730
 731             /* Inner loop uses 62 flops */
 732         }
 733
 734         /* End of innermost loop */
 735
 736         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 737                                               f+i_coord_offset,fshift+i_shift_offset);
 738
 739         /* Increment number of inner iterations */
 740         inneriter                  += j_index_end - j_index_start;
 741
 742         /* Outer loop uses 7 flops */
 743     }
 744
 745     /* Increment number of outer iterations */
 746     outeriter        += nri;
 747
 748     /* Update outer/inner flops */
 749
 750     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 751 }