src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_sse2_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  94     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_ps(fr->ic->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_vdw->data;
 121     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_ps();
 156         fiy0             = _mm_setzero_ps();
 157         fiz0             = _mm_setzero_ps();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_ps();
 165         vvdwsum          = _mm_setzero_ps();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178             j_coord_offsetC  = DIM*jnrC;
 179             j_coord_offsetD  = DIM*jnrD;
 180
 181             /* load j atom coordinates */
 182             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                               x+j_coord_offsetC,x+j_coord_offsetD,
 184                                               &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm_sub_ps(ix0,jx0);
 188             dy00             = _mm_sub_ps(iy0,jy0);
 189             dz00             = _mm_sub_ps(iz0,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 193
 194             rinv00           = sse2_invsqrt_f(rsq00);
 195
 196             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                               charge+jnrC+0,charge+jnrD+0);
 201             vdwjidx0A        = 2*vdwtype[jnrA+0];
 202             vdwjidx0B        = 2*vdwtype[jnrB+0];
 203             vdwjidx0C        = 2*vdwtype[jnrC+0];
 204             vdwjidx0D        = 2*vdwtype[jnrD+0];
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             r00              = _mm_mul_ps(rsq00,rinv00);
 211
 212             /* Compute parameters for interactions between i and j atoms */
 213             qq00             = _mm_mul_ps(iq0,jq0);
 214             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 215                                          vdwparam+vdwioffset0+vdwjidx0B,
 216                                          vdwparam+vdwioffset0+vdwjidx0C,
 217                                          vdwparam+vdwioffset0+vdwjidx0D,
 218                                          &c6_00,&c12_00);
 219
 220             /* Calculate table index by multiplying r with table scale and truncate to integer */
 221             rt               = _mm_mul_ps(r00,vftabscale);
 222             vfitab           = _mm_cvttps_epi32(rt);
 223             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 224             vfitab           = _mm_slli_epi32(vfitab,3);
 225
 226             /* COULOMB ELECTROSTATICS */
 227             velec            = _mm_mul_ps(qq00,rinv00);
 228             felec            = _mm_mul_ps(velec,rinvsq00);
 229
 230             /* CUBIC SPLINE TABLE DISPERSION */
 231             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 232             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 233             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 234             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 235             _MM_TRANSPOSE4_PS(Y,F,G,H);
 236             Heps             = _mm_mul_ps(vfeps,H);
 237             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 238             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 239             vvdw6            = _mm_mul_ps(c6_00,VV);
 240             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 241             fvdw6            = _mm_mul_ps(c6_00,FF);
 242
 243             /* CUBIC SPLINE TABLE REPULSION */
 244             vfitab           = _mm_add_epi32(vfitab,ifour);
 245             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 246             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 247             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 248             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 249             _MM_TRANSPOSE4_PS(Y,F,G,H);
 250             Heps             = _mm_mul_ps(vfeps,H);
 251             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 252             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 253             vvdw12           = _mm_mul_ps(c12_00,VV);
 254             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 255             fvdw12           = _mm_mul_ps(c12_00,FF);
 256             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 257             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 258
 259             /* Update potential sum for this i atom from the interaction with this j atom. */
 260             velecsum         = _mm_add_ps(velecsum,velec);
 261             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 262
 263             fscal            = _mm_add_ps(felec,fvdw);
 264
 265             /* Calculate temporary vectorial force */
 266             tx               = _mm_mul_ps(fscal,dx00);
 267             ty               = _mm_mul_ps(fscal,dy00);
 268             tz               = _mm_mul_ps(fscal,dz00);
 269
 270             /* Update vectorial force */
 271             fix0             = _mm_add_ps(fix0,tx);
 272             fiy0             = _mm_add_ps(fiy0,ty);
 273             fiz0             = _mm_add_ps(fiz0,tz);
 274
 275             fjptrA             = f+j_coord_offsetA;
 276             fjptrB             = f+j_coord_offsetB;
 277             fjptrC             = f+j_coord_offsetC;
 278             fjptrD             = f+j_coord_offsetD;
 279             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 280
 281             /* Inner loop uses 63 flops */
 282         }
 283
 284         if(jidx<j_index_end)
 285         {
 286
 287             /* Get j neighbor index, and coordinate index */
 288             jnrlistA         = jjnr[jidx];
 289             jnrlistB         = jjnr[jidx+1];
 290             jnrlistC         = jjnr[jidx+2];
 291             jnrlistD         = jjnr[jidx+3];
 292             /* Sign of each element will be negative for non-real atoms.
 293              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 294              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 295              */
 296             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 297             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 298             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 299             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 300             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 301             j_coord_offsetA  = DIM*jnrA;
 302             j_coord_offsetB  = DIM*jnrB;
 303             j_coord_offsetC  = DIM*jnrC;
 304             j_coord_offsetD  = DIM*jnrD;
 305
 306             /* load j atom coordinates */
 307             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 308                                               x+j_coord_offsetC,x+j_coord_offsetD,
 309                                               &jx0,&jy0,&jz0);
 310
 311             /* Calculate displacement vector */
 312             dx00             = _mm_sub_ps(ix0,jx0);
 313             dy00             = _mm_sub_ps(iy0,jy0);
 314             dz00             = _mm_sub_ps(iz0,jz0);
 315
 316             /* Calculate squared distance and things based on it */
 317             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 318
 319             rinv00           = sse2_invsqrt_f(rsq00);
 320
 321             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 322
 323             /* Load parameters for j particles */
 324             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 325                                                               charge+jnrC+0,charge+jnrD+0);
 326             vdwjidx0A        = 2*vdwtype[jnrA+0];
 327             vdwjidx0B        = 2*vdwtype[jnrB+0];
 328             vdwjidx0C        = 2*vdwtype[jnrC+0];
 329             vdwjidx0D        = 2*vdwtype[jnrD+0];
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r00              = _mm_mul_ps(rsq00,rinv00);
 336             r00              = _mm_andnot_ps(dummy_mask,r00);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq00             = _mm_mul_ps(iq0,jq0);
 340             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 341                                          vdwparam+vdwioffset0+vdwjidx0B,
 342                                          vdwparam+vdwioffset0+vdwjidx0C,
 343                                          vdwparam+vdwioffset0+vdwjidx0D,
 344                                          &c6_00,&c12_00);
 345
 346             /* Calculate table index by multiplying r with table scale and truncate to integer */
 347             rt               = _mm_mul_ps(r00,vftabscale);
 348             vfitab           = _mm_cvttps_epi32(rt);
 349             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 350             vfitab           = _mm_slli_epi32(vfitab,3);
 351
 352             /* COULOMB ELECTROSTATICS */
 353             velec            = _mm_mul_ps(qq00,rinv00);
 354             felec            = _mm_mul_ps(velec,rinvsq00);
 355
 356             /* CUBIC SPLINE TABLE DISPERSION */
 357             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 358             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 359             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 360             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 361             _MM_TRANSPOSE4_PS(Y,F,G,H);
 362             Heps             = _mm_mul_ps(vfeps,H);
 363             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 364             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 365             vvdw6            = _mm_mul_ps(c6_00,VV);
 366             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 367             fvdw6            = _mm_mul_ps(c6_00,FF);
 368
 369             /* CUBIC SPLINE TABLE REPULSION */
 370             vfitab           = _mm_add_epi32(vfitab,ifour);
 371             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 372             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 373             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 374             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 375             _MM_TRANSPOSE4_PS(Y,F,G,H);
 376             Heps             = _mm_mul_ps(vfeps,H);
 377             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 378             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 379             vvdw12           = _mm_mul_ps(c12_00,VV);
 380             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 381             fvdw12           = _mm_mul_ps(c12_00,FF);
 382             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 383             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velec            = _mm_andnot_ps(dummy_mask,velec);
 387             velecsum         = _mm_add_ps(velecsum,velec);
 388             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 389             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 390
 391             fscal            = _mm_add_ps(felec,fvdw);
 392
 393             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm_mul_ps(fscal,dx00);
 397             ty               = _mm_mul_ps(fscal,dy00);
 398             tz               = _mm_mul_ps(fscal,dz00);
 399
 400             /* Update vectorial force */
 401             fix0             = _mm_add_ps(fix0,tx);
 402             fiy0             = _mm_add_ps(fiy0,ty);
 403             fiz0             = _mm_add_ps(fiz0,tz);
 404
 405             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 406             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 407             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 408             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 409             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 410
 411             /* Inner loop uses 64 flops */
 412         }
 413
 414         /* End of innermost loop */
 415
 416         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 417                                               f+i_coord_offset,fshift+i_shift_offset);
 418
 419         ggid                        = gid[iidx];
 420         /* Update potential energies */
 421         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 422         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 423
 424         /* Increment number of inner iterations */
 425         inneriter                  += j_index_end - j_index_start;
 426
 427         /* Outer loop uses 9 flops */
 428     }
 429
 430     /* Increment number of outer iterations */
 431     outeriter        += nri;
 432
 433     /* Update outer/inner flops */
 434
 435     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*64);
 436 }
 437 /*
 438  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_single
 439  * Electrostatics interaction: Coulomb
 440  * VdW interaction:            CubicSplineTable
 441  * Geometry:                   Particle-Particle
 442  * Calculate force/pot:        Force
 443  */
 444 void
 445 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_single
 446                     (t_nblist                    * gmx_restrict       nlist,
 447                      rvec                        * gmx_restrict          xx,
 448                      rvec                        * gmx_restrict          ff,
 449                      struct t_forcerec           * gmx_restrict          fr,
 450                      t_mdatoms                   * gmx_restrict     mdatoms,
 451                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 452                      t_nrnb                      * gmx_restrict        nrnb)
 453 {
 454     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 455      * just 0 for non-waters.
 456      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 457      * jnr indices corresponding to data put in the four positions in the SIMD register.
 458      */
 459     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 460     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 461     int              jnrA,jnrB,jnrC,jnrD;
 462     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 463     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 464     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 465     real             rcutoff_scalar;
 466     real             *shiftvec,*fshift,*x,*f;
 467     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 468     real             scratch[4*DIM];
 469     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 470     int              vdwioffset0;
 471     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 472     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 473     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 474     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 475     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 476     real             *charge;
 477     int              nvdwtype;
 478     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 479     int              *vdwtype;
 480     real             *vdwparam;
 481     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 482     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 483     __m128i          vfitab;
 484     __m128i          ifour       = _mm_set1_epi32(4);
 485     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 486     real             *vftab;
 487     __m128           dummy_mask,cutoff_mask;
 488     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 489     __m128           one     = _mm_set1_ps(1.0);
 490     __m128           two     = _mm_set1_ps(2.0);
 491     x                = xx[0];
 492     f                = ff[0];
 493
 494     nri              = nlist->nri;
 495     iinr             = nlist->iinr;
 496     jindex           = nlist->jindex;
 497     jjnr             = nlist->jjnr;
 498     shiftidx         = nlist->shift;
 499     gid              = nlist->gid;
 500     shiftvec         = fr->shift_vec[0];
 501     fshift           = fr->fshift[0];
 502     facel            = _mm_set1_ps(fr->ic->epsfac);
 503     charge           = mdatoms->chargeA;
 504     nvdwtype         = fr->ntype;
 505     vdwparam         = fr->nbfp;
 506     vdwtype          = mdatoms->typeA;
 507
 508     vftab            = kernel_data->table_vdw->data;
 509     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 510
 511     /* Avoid stupid compiler warnings */
 512     jnrA = jnrB = jnrC = jnrD = 0;
 513     j_coord_offsetA = 0;
 514     j_coord_offsetB = 0;
 515     j_coord_offsetC = 0;
 516     j_coord_offsetD = 0;
 517
 518     outeriter        = 0;
 519     inneriter        = 0;
 520
 521     for(iidx=0;iidx<4*DIM;iidx++)
 522     {
 523         scratch[iidx] = 0.0;
 524     }
 525
 526     /* Start outer loop over neighborlists */
 527     for(iidx=0; iidx<nri; iidx++)
 528     {
 529         /* Load shift vector for this list */
 530         i_shift_offset   = DIM*shiftidx[iidx];
 531
 532         /* Load limits for loop over neighbors */
 533         j_index_start    = jindex[iidx];
 534         j_index_end      = jindex[iidx+1];
 535
 536         /* Get outer coordinate index */
 537         inr              = iinr[iidx];
 538         i_coord_offset   = DIM*inr;
 539
 540         /* Load i particle coords and add shift vector */
 541         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 542
 543         fix0             = _mm_setzero_ps();
 544         fiy0             = _mm_setzero_ps();
 545         fiz0             = _mm_setzero_ps();
 546
 547         /* Load parameters for i particles */
 548         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 549         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 550
 551         /* Start inner kernel loop */
 552         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 553         {
 554
 555             /* Get j neighbor index, and coordinate index */
 556             jnrA             = jjnr[jidx];
 557             jnrB             = jjnr[jidx+1];
 558             jnrC             = jjnr[jidx+2];
 559             jnrD             = jjnr[jidx+3];
 560             j_coord_offsetA  = DIM*jnrA;
 561             j_coord_offsetB  = DIM*jnrB;
 562             j_coord_offsetC  = DIM*jnrC;
 563             j_coord_offsetD  = DIM*jnrD;
 564
 565             /* load j atom coordinates */
 566             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 567                                               x+j_coord_offsetC,x+j_coord_offsetD,
 568                                               &jx0,&jy0,&jz0);
 569
 570             /* Calculate displacement vector */
 571             dx00             = _mm_sub_ps(ix0,jx0);
 572             dy00             = _mm_sub_ps(iy0,jy0);
 573             dz00             = _mm_sub_ps(iz0,jz0);
 574
 575             /* Calculate squared distance and things based on it */
 576             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 577
 578             rinv00           = sse2_invsqrt_f(rsq00);
 579
 580             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 581
 582             /* Load parameters for j particles */
 583             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 584                                                               charge+jnrC+0,charge+jnrD+0);
 585             vdwjidx0A        = 2*vdwtype[jnrA+0];
 586             vdwjidx0B        = 2*vdwtype[jnrB+0];
 587             vdwjidx0C        = 2*vdwtype[jnrC+0];
 588             vdwjidx0D        = 2*vdwtype[jnrD+0];
 589
 590             /**************************
 591              * CALCULATE INTERACTIONS *
 592              **************************/
 593
 594             r00              = _mm_mul_ps(rsq00,rinv00);
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq00             = _mm_mul_ps(iq0,jq0);
 598             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 599                                          vdwparam+vdwioffset0+vdwjidx0B,
 600                                          vdwparam+vdwioffset0+vdwjidx0C,
 601                                          vdwparam+vdwioffset0+vdwjidx0D,
 602                                          &c6_00,&c12_00);
 603
 604             /* Calculate table index by multiplying r with table scale and truncate to integer */
 605             rt               = _mm_mul_ps(r00,vftabscale);
 606             vfitab           = _mm_cvttps_epi32(rt);
 607             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 608             vfitab           = _mm_slli_epi32(vfitab,3);
 609
 610             /* COULOMB ELECTROSTATICS */
 611             velec            = _mm_mul_ps(qq00,rinv00);
 612             felec            = _mm_mul_ps(velec,rinvsq00);
 613
 614             /* CUBIC SPLINE TABLE DISPERSION */
 615             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 616             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 617             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 618             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 619             _MM_TRANSPOSE4_PS(Y,F,G,H);
 620             Heps             = _mm_mul_ps(vfeps,H);
 621             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 622             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 623             fvdw6            = _mm_mul_ps(c6_00,FF);
 624
 625             /* CUBIC SPLINE TABLE REPULSION */
 626             vfitab           = _mm_add_epi32(vfitab,ifour);
 627             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 628             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 629             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 630             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 631             _MM_TRANSPOSE4_PS(Y,F,G,H);
 632             Heps             = _mm_mul_ps(vfeps,H);
 633             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 634             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 635             fvdw12           = _mm_mul_ps(c12_00,FF);
 636             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 637
 638             fscal            = _mm_add_ps(felec,fvdw);
 639
 640             /* Calculate temporary vectorial force */
 641             tx               = _mm_mul_ps(fscal,dx00);
 642             ty               = _mm_mul_ps(fscal,dy00);
 643             tz               = _mm_mul_ps(fscal,dz00);
 644
 645             /* Update vectorial force */
 646             fix0             = _mm_add_ps(fix0,tx);
 647             fiy0             = _mm_add_ps(fiy0,ty);
 648             fiz0             = _mm_add_ps(fiz0,tz);
 649
 650             fjptrA             = f+j_coord_offsetA;
 651             fjptrB             = f+j_coord_offsetB;
 652             fjptrC             = f+j_coord_offsetC;
 653             fjptrD             = f+j_coord_offsetD;
 654             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 655
 656             /* Inner loop uses 54 flops */
 657         }
 658
 659         if(jidx<j_index_end)
 660         {
 661
 662             /* Get j neighbor index, and coordinate index */
 663             jnrlistA         = jjnr[jidx];
 664             jnrlistB         = jjnr[jidx+1];
 665             jnrlistC         = jjnr[jidx+2];
 666             jnrlistD         = jjnr[jidx+3];
 667             /* Sign of each element will be negative for non-real atoms.
 668              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 669              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 670              */
 671             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 672             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 673             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 674             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 675             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 676             j_coord_offsetA  = DIM*jnrA;
 677             j_coord_offsetB  = DIM*jnrB;
 678             j_coord_offsetC  = DIM*jnrC;
 679             j_coord_offsetD  = DIM*jnrD;
 680
 681             /* load j atom coordinates */
 682             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 683                                               x+j_coord_offsetC,x+j_coord_offsetD,
 684                                               &jx0,&jy0,&jz0);
 685
 686             /* Calculate displacement vector */
 687             dx00             = _mm_sub_ps(ix0,jx0);
 688             dy00             = _mm_sub_ps(iy0,jy0);
 689             dz00             = _mm_sub_ps(iz0,jz0);
 690
 691             /* Calculate squared distance and things based on it */
 692             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 693
 694             rinv00           = sse2_invsqrt_f(rsq00);
 695
 696             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 697
 698             /* Load parameters for j particles */
 699             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 700                                                               charge+jnrC+0,charge+jnrD+0);
 701             vdwjidx0A        = 2*vdwtype[jnrA+0];
 702             vdwjidx0B        = 2*vdwtype[jnrB+0];
 703             vdwjidx0C        = 2*vdwtype[jnrC+0];
 704             vdwjidx0D        = 2*vdwtype[jnrD+0];
 705
 706             /**************************
 707              * CALCULATE INTERACTIONS *
 708              **************************/
 709
 710             r00              = _mm_mul_ps(rsq00,rinv00);
 711             r00              = _mm_andnot_ps(dummy_mask,r00);
 712
 713             /* Compute parameters for interactions between i and j atoms */
 714             qq00             = _mm_mul_ps(iq0,jq0);
 715             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 716                                          vdwparam+vdwioffset0+vdwjidx0B,
 717                                          vdwparam+vdwioffset0+vdwjidx0C,
 718                                          vdwparam+vdwioffset0+vdwjidx0D,
 719                                          &c6_00,&c12_00);
 720
 721             /* Calculate table index by multiplying r with table scale and truncate to integer */
 722             rt               = _mm_mul_ps(r00,vftabscale);
 723             vfitab           = _mm_cvttps_epi32(rt);
 724             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 725             vfitab           = _mm_slli_epi32(vfitab,3);
 726
 727             /* COULOMB ELECTROSTATICS */
 728             velec            = _mm_mul_ps(qq00,rinv00);
 729             felec            = _mm_mul_ps(velec,rinvsq00);
 730
 731             /* CUBIC SPLINE TABLE DISPERSION */
 732             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 733             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 734             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 735             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 736             _MM_TRANSPOSE4_PS(Y,F,G,H);
 737             Heps             = _mm_mul_ps(vfeps,H);
 738             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 739             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 740             fvdw6            = _mm_mul_ps(c6_00,FF);
 741
 742             /* CUBIC SPLINE TABLE REPULSION */
 743             vfitab           = _mm_add_epi32(vfitab,ifour);
 744             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 745             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 746             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 747             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 748             _MM_TRANSPOSE4_PS(Y,F,G,H);
 749             Heps             = _mm_mul_ps(vfeps,H);
 750             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 751             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 752             fvdw12           = _mm_mul_ps(c12_00,FF);
 753             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 754
 755             fscal            = _mm_add_ps(felec,fvdw);
 756
 757             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 758
 759             /* Calculate temporary vectorial force */
 760             tx               = _mm_mul_ps(fscal,dx00);
 761             ty               = _mm_mul_ps(fscal,dy00);
 762             tz               = _mm_mul_ps(fscal,dz00);
 763
 764             /* Update vectorial force */
 765             fix0             = _mm_add_ps(fix0,tx);
 766             fiy0             = _mm_add_ps(fiy0,ty);
 767             fiz0             = _mm_add_ps(fiz0,tz);
 768
 769             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 770             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 771             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 772             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 773             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 774
 775             /* Inner loop uses 55 flops */
 776         }
 777
 778         /* End of innermost loop */
 779
 780         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 781                                               f+i_coord_offset,fshift+i_shift_offset);
 782
 783         /* Increment number of inner iterations */
 784         inneriter                  += j_index_end - j_index_start;
 785
 786         /* Outer loop uses 7 flops */
 787     }
 788
 789     /* Increment number of outer iterations */
 790     outeriter        += nri;
 791
 792     /* Update outer/inner flops */
 793
 794     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);
 795 }