src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_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_ElecCSTab_VdwCSTab_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_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_elec_vdw->data;
 121     vftabscale       = _mm_set1_ps(kernel_data->table_elec_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             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 198                                                               charge+jnrC+0,charge+jnrD+0);
 199             vdwjidx0A        = 2*vdwtype[jnrA+0];
 200             vdwjidx0B        = 2*vdwtype[jnrB+0];
 201             vdwjidx0C        = 2*vdwtype[jnrC+0];
 202             vdwjidx0D        = 2*vdwtype[jnrD+0];
 203
 204             /**************************
 205              * CALCULATE INTERACTIONS *
 206              **************************/
 207
 208             r00              = _mm_mul_ps(rsq00,rinv00);
 209
 210             /* Compute parameters for interactions between i and j atoms */
 211             qq00             = _mm_mul_ps(iq0,jq0);
 212             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 213                                          vdwparam+vdwioffset0+vdwjidx0B,
 214                                          vdwparam+vdwioffset0+vdwjidx0C,
 215                                          vdwparam+vdwioffset0+vdwjidx0D,
 216                                          &c6_00,&c12_00);
 217
 218             /* Calculate table index by multiplying r with table scale and truncate to integer */
 219             rt               = _mm_mul_ps(r00,vftabscale);
 220             vfitab           = _mm_cvttps_epi32(rt);
 221             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 222             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 223
 224             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 225             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 226             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 227             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 228             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 229             _MM_TRANSPOSE4_PS(Y,F,G,H);
 230             Heps             = _mm_mul_ps(vfeps,H);
 231             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 232             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 233             velec            = _mm_mul_ps(qq00,VV);
 234             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 235             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 236
 237             /* CUBIC SPLINE TABLE DISPERSION */
 238             vfitab           = _mm_add_epi32(vfitab,ifour);
 239             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 240             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 241             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 242             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 243             _MM_TRANSPOSE4_PS(Y,F,G,H);
 244             Heps             = _mm_mul_ps(vfeps,H);
 245             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 246             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 247             vvdw6            = _mm_mul_ps(c6_00,VV);
 248             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 249             fvdw6            = _mm_mul_ps(c6_00,FF);
 250
 251             /* CUBIC SPLINE TABLE REPULSION */
 252             vfitab           = _mm_add_epi32(vfitab,ifour);
 253             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 254             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 255             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 256             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 257             _MM_TRANSPOSE4_PS(Y,F,G,H);
 258             Heps             = _mm_mul_ps(vfeps,H);
 259             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 260             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 261             vvdw12           = _mm_mul_ps(c12_00,VV);
 262             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 263             fvdw12           = _mm_mul_ps(c12_00,FF);
 264             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 265             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velecsum         = _mm_add_ps(velecsum,velec);
 269             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 270
 271             fscal            = _mm_add_ps(felec,fvdw);
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm_mul_ps(fscal,dx00);
 275             ty               = _mm_mul_ps(fscal,dy00);
 276             tz               = _mm_mul_ps(fscal,dz00);
 277
 278             /* Update vectorial force */
 279             fix0             = _mm_add_ps(fix0,tx);
 280             fiy0             = _mm_add_ps(fiy0,ty);
 281             fiz0             = _mm_add_ps(fiz0,tz);
 282
 283             fjptrA             = f+j_coord_offsetA;
 284             fjptrB             = f+j_coord_offsetB;
 285             fjptrC             = f+j_coord_offsetC;
 286             fjptrD             = f+j_coord_offsetD;
 287             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 288
 289             /* Inner loop uses 73 flops */
 290         }
 291
 292         if(jidx<j_index_end)
 293         {
 294
 295             /* Get j neighbor index, and coordinate index */
 296             jnrlistA         = jjnr[jidx];
 297             jnrlistB         = jjnr[jidx+1];
 298             jnrlistC         = jjnr[jidx+2];
 299             jnrlistD         = jjnr[jidx+3];
 300             /* Sign of each element will be negative for non-real atoms.
 301              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 302              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 303              */
 304             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 305             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 306             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 307             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 308             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 309             j_coord_offsetA  = DIM*jnrA;
 310             j_coord_offsetB  = DIM*jnrB;
 311             j_coord_offsetC  = DIM*jnrC;
 312             j_coord_offsetD  = DIM*jnrD;
 313
 314             /* load j atom coordinates */
 315             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 316                                               x+j_coord_offsetC,x+j_coord_offsetD,
 317                                               &jx0,&jy0,&jz0);
 318
 319             /* Calculate displacement vector */
 320             dx00             = _mm_sub_ps(ix0,jx0);
 321             dy00             = _mm_sub_ps(iy0,jy0);
 322             dz00             = _mm_sub_ps(iz0,jz0);
 323
 324             /* Calculate squared distance and things based on it */
 325             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 326
 327             rinv00           = sse2_invsqrt_f(rsq00);
 328
 329             /* Load parameters for j particles */
 330             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 331                                                               charge+jnrC+0,charge+jnrD+0);
 332             vdwjidx0A        = 2*vdwtype[jnrA+0];
 333             vdwjidx0B        = 2*vdwtype[jnrB+0];
 334             vdwjidx0C        = 2*vdwtype[jnrC+0];
 335             vdwjidx0D        = 2*vdwtype[jnrD+0];
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             r00              = _mm_mul_ps(rsq00,rinv00);
 342             r00              = _mm_andnot_ps(dummy_mask,r00);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq00             = _mm_mul_ps(iq0,jq0);
 346             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 347                                          vdwparam+vdwioffset0+vdwjidx0B,
 348                                          vdwparam+vdwioffset0+vdwjidx0C,
 349                                          vdwparam+vdwioffset0+vdwjidx0D,
 350                                          &c6_00,&c12_00);
 351
 352             /* Calculate table index by multiplying r with table scale and truncate to integer */
 353             rt               = _mm_mul_ps(r00,vftabscale);
 354             vfitab           = _mm_cvttps_epi32(rt);
 355             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 356             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 357
 358             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 359             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 360             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 361             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 362             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 363             _MM_TRANSPOSE4_PS(Y,F,G,H);
 364             Heps             = _mm_mul_ps(vfeps,H);
 365             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 366             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 367             velec            = _mm_mul_ps(qq00,VV);
 368             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 369             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 370
 371             /* CUBIC SPLINE TABLE DISPERSION */
 372             vfitab           = _mm_add_epi32(vfitab,ifour);
 373             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 374             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 375             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 376             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 377             _MM_TRANSPOSE4_PS(Y,F,G,H);
 378             Heps             = _mm_mul_ps(vfeps,H);
 379             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 380             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 381             vvdw6            = _mm_mul_ps(c6_00,VV);
 382             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 383             fvdw6            = _mm_mul_ps(c6_00,FF);
 384
 385             /* CUBIC SPLINE TABLE REPULSION */
 386             vfitab           = _mm_add_epi32(vfitab,ifour);
 387             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 388             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 389             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 390             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 391             _MM_TRANSPOSE4_PS(Y,F,G,H);
 392             Heps             = _mm_mul_ps(vfeps,H);
 393             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 394             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 395             vvdw12           = _mm_mul_ps(c12_00,VV);
 396             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 397             fvdw12           = _mm_mul_ps(c12_00,FF);
 398             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 399             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velec            = _mm_andnot_ps(dummy_mask,velec);
 403             velecsum         = _mm_add_ps(velecsum,velec);
 404             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 405             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 406
 407             fscal            = _mm_add_ps(felec,fvdw);
 408
 409             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm_mul_ps(fscal,dx00);
 413             ty               = _mm_mul_ps(fscal,dy00);
 414             tz               = _mm_mul_ps(fscal,dz00);
 415
 416             /* Update vectorial force */
 417             fix0             = _mm_add_ps(fix0,tx);
 418             fiy0             = _mm_add_ps(fiy0,ty);
 419             fiz0             = _mm_add_ps(fiz0,tz);
 420
 421             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 422             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 423             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 424             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 425             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 426
 427             /* Inner loop uses 74 flops */
 428         }
 429
 430         /* End of innermost loop */
 431
 432         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 433                                               f+i_coord_offset,fshift+i_shift_offset);
 434
 435         ggid                        = gid[iidx];
 436         /* Update potential energies */
 437         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 438         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 439
 440         /* Increment number of inner iterations */
 441         inneriter                  += j_index_end - j_index_start;
 442
 443         /* Outer loop uses 9 flops */
 444     }
 445
 446     /* Increment number of outer iterations */
 447     outeriter        += nri;
 448
 449     /* Update outer/inner flops */
 450
 451     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*74);
 452 }
 453 /*
 454  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_single
 455  * Electrostatics interaction: CubicSplineTable
 456  * VdW interaction:            CubicSplineTable
 457  * Geometry:                   Particle-Particle
 458  * Calculate force/pot:        Force
 459  */
 460 void
 461 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_sse2_single
 462                     (t_nblist                    * gmx_restrict       nlist,
 463                      rvec                        * gmx_restrict          xx,
 464                      rvec                        * gmx_restrict          ff,
 465                      struct t_forcerec           * gmx_restrict          fr,
 466                      t_mdatoms                   * gmx_restrict     mdatoms,
 467                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 468                      t_nrnb                      * gmx_restrict        nrnb)
 469 {
 470     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 471      * just 0 for non-waters.
 472      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 473      * jnr indices corresponding to data put in the four positions in the SIMD register.
 474      */
 475     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 476     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 477     int              jnrA,jnrB,jnrC,jnrD;
 478     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 479     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 480     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 481     real             rcutoff_scalar;
 482     real             *shiftvec,*fshift,*x,*f;
 483     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 484     real             scratch[4*DIM];
 485     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 486     int              vdwioffset0;
 487     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 488     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 489     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 490     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 491     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 492     real             *charge;
 493     int              nvdwtype;
 494     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 495     int              *vdwtype;
 496     real             *vdwparam;
 497     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 498     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 499     __m128i          vfitab;
 500     __m128i          ifour       = _mm_set1_epi32(4);
 501     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 502     real             *vftab;
 503     __m128           dummy_mask,cutoff_mask;
 504     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 505     __m128           one     = _mm_set1_ps(1.0);
 506     __m128           two     = _mm_set1_ps(2.0);
 507     x                = xx[0];
 508     f                = ff[0];
 509
 510     nri              = nlist->nri;
 511     iinr             = nlist->iinr;
 512     jindex           = nlist->jindex;
 513     jjnr             = nlist->jjnr;
 514     shiftidx         = nlist->shift;
 515     gid              = nlist->gid;
 516     shiftvec         = fr->shift_vec[0];
 517     fshift           = fr->fshift[0];
 518     facel            = _mm_set1_ps(fr->ic->epsfac);
 519     charge           = mdatoms->chargeA;
 520     nvdwtype         = fr->ntype;
 521     vdwparam         = fr->nbfp;
 522     vdwtype          = mdatoms->typeA;
 523
 524     vftab            = kernel_data->table_elec_vdw->data;
 525     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 526
 527     /* Avoid stupid compiler warnings */
 528     jnrA = jnrB = jnrC = jnrD = 0;
 529     j_coord_offsetA = 0;
 530     j_coord_offsetB = 0;
 531     j_coord_offsetC = 0;
 532     j_coord_offsetD = 0;
 533
 534     outeriter        = 0;
 535     inneriter        = 0;
 536
 537     for(iidx=0;iidx<4*DIM;iidx++)
 538     {
 539         scratch[iidx] = 0.0;
 540     }
 541
 542     /* Start outer loop over neighborlists */
 543     for(iidx=0; iidx<nri; iidx++)
 544     {
 545         /* Load shift vector for this list */
 546         i_shift_offset   = DIM*shiftidx[iidx];
 547
 548         /* Load limits for loop over neighbors */
 549         j_index_start    = jindex[iidx];
 550         j_index_end      = jindex[iidx+1];
 551
 552         /* Get outer coordinate index */
 553         inr              = iinr[iidx];
 554         i_coord_offset   = DIM*inr;
 555
 556         /* Load i particle coords and add shift vector */
 557         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 558
 559         fix0             = _mm_setzero_ps();
 560         fiy0             = _mm_setzero_ps();
 561         fiz0             = _mm_setzero_ps();
 562
 563         /* Load parameters for i particles */
 564         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 565         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 566
 567         /* Start inner kernel loop */
 568         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 569         {
 570
 571             /* Get j neighbor index, and coordinate index */
 572             jnrA             = jjnr[jidx];
 573             jnrB             = jjnr[jidx+1];
 574             jnrC             = jjnr[jidx+2];
 575             jnrD             = jjnr[jidx+3];
 576             j_coord_offsetA  = DIM*jnrA;
 577             j_coord_offsetB  = DIM*jnrB;
 578             j_coord_offsetC  = DIM*jnrC;
 579             j_coord_offsetD  = DIM*jnrD;
 580
 581             /* load j atom coordinates */
 582             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 583                                               x+j_coord_offsetC,x+j_coord_offsetD,
 584                                               &jx0,&jy0,&jz0);
 585
 586             /* Calculate displacement vector */
 587             dx00             = _mm_sub_ps(ix0,jx0);
 588             dy00             = _mm_sub_ps(iy0,jy0);
 589             dz00             = _mm_sub_ps(iz0,jz0);
 590
 591             /* Calculate squared distance and things based on it */
 592             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 593
 594             rinv00           = sse2_invsqrt_f(rsq00);
 595
 596             /* Load parameters for j particles */
 597             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 598                                                               charge+jnrC+0,charge+jnrD+0);
 599             vdwjidx0A        = 2*vdwtype[jnrA+0];
 600             vdwjidx0B        = 2*vdwtype[jnrB+0];
 601             vdwjidx0C        = 2*vdwtype[jnrC+0];
 602             vdwjidx0D        = 2*vdwtype[jnrD+0];
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             r00              = _mm_mul_ps(rsq00,rinv00);
 609
 610             /* Compute parameters for interactions between i and j atoms */
 611             qq00             = _mm_mul_ps(iq0,jq0);
 612             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 613                                          vdwparam+vdwioffset0+vdwjidx0B,
 614                                          vdwparam+vdwioffset0+vdwjidx0C,
 615                                          vdwparam+vdwioffset0+vdwjidx0D,
 616                                          &c6_00,&c12_00);
 617
 618             /* Calculate table index by multiplying r with table scale and truncate to integer */
 619             rt               = _mm_mul_ps(r00,vftabscale);
 620             vfitab           = _mm_cvttps_epi32(rt);
 621             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 622             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 623
 624             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 625             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 626             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 627             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 628             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 629             _MM_TRANSPOSE4_PS(Y,F,G,H);
 630             Heps             = _mm_mul_ps(vfeps,H);
 631             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 632             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 633             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 634
 635             /* CUBIC SPLINE TABLE DISPERSION */
 636             vfitab           = _mm_add_epi32(vfitab,ifour);
 637             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 638             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 639             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 640             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 641             _MM_TRANSPOSE4_PS(Y,F,G,H);
 642             Heps             = _mm_mul_ps(vfeps,H);
 643             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 644             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 645             fvdw6            = _mm_mul_ps(c6_00,FF);
 646
 647             /* CUBIC SPLINE TABLE REPULSION */
 648             vfitab           = _mm_add_epi32(vfitab,ifour);
 649             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 650             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 651             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 652             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 653             _MM_TRANSPOSE4_PS(Y,F,G,H);
 654             Heps             = _mm_mul_ps(vfeps,H);
 655             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 656             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 657             fvdw12           = _mm_mul_ps(c12_00,FF);
 658             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 659
 660             fscal            = _mm_add_ps(felec,fvdw);
 661
 662             /* Calculate temporary vectorial force */
 663             tx               = _mm_mul_ps(fscal,dx00);
 664             ty               = _mm_mul_ps(fscal,dy00);
 665             tz               = _mm_mul_ps(fscal,dz00);
 666
 667             /* Update vectorial force */
 668             fix0             = _mm_add_ps(fix0,tx);
 669             fiy0             = _mm_add_ps(fiy0,ty);
 670             fiz0             = _mm_add_ps(fiz0,tz);
 671
 672             fjptrA             = f+j_coord_offsetA;
 673             fjptrB             = f+j_coord_offsetB;
 674             fjptrC             = f+j_coord_offsetC;
 675             fjptrD             = f+j_coord_offsetD;
 676             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 677
 678             /* Inner loop uses 61 flops */
 679         }
 680
 681         if(jidx<j_index_end)
 682         {
 683
 684             /* Get j neighbor index, and coordinate index */
 685             jnrlistA         = jjnr[jidx];
 686             jnrlistB         = jjnr[jidx+1];
 687             jnrlistC         = jjnr[jidx+2];
 688             jnrlistD         = jjnr[jidx+3];
 689             /* Sign of each element will be negative for non-real atoms.
 690              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 691              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 692              */
 693             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 694             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 695             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 696             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 697             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 698             j_coord_offsetA  = DIM*jnrA;
 699             j_coord_offsetB  = DIM*jnrB;
 700             j_coord_offsetC  = DIM*jnrC;
 701             j_coord_offsetD  = DIM*jnrD;
 702
 703             /* load j atom coordinates */
 704             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 705                                               x+j_coord_offsetC,x+j_coord_offsetD,
 706                                               &jx0,&jy0,&jz0);
 707
 708             /* Calculate displacement vector */
 709             dx00             = _mm_sub_ps(ix0,jx0);
 710             dy00             = _mm_sub_ps(iy0,jy0);
 711             dz00             = _mm_sub_ps(iz0,jz0);
 712
 713             /* Calculate squared distance and things based on it */
 714             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 715
 716             rinv00           = sse2_invsqrt_f(rsq00);
 717
 718             /* Load parameters for j particles */
 719             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 720                                                               charge+jnrC+0,charge+jnrD+0);
 721             vdwjidx0A        = 2*vdwtype[jnrA+0];
 722             vdwjidx0B        = 2*vdwtype[jnrB+0];
 723             vdwjidx0C        = 2*vdwtype[jnrC+0];
 724             vdwjidx0D        = 2*vdwtype[jnrD+0];
 725
 726             /**************************
 727              * CALCULATE INTERACTIONS *
 728              **************************/
 729
 730             r00              = _mm_mul_ps(rsq00,rinv00);
 731             r00              = _mm_andnot_ps(dummy_mask,r00);
 732
 733             /* Compute parameters for interactions between i and j atoms */
 734             qq00             = _mm_mul_ps(iq0,jq0);
 735             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 736                                          vdwparam+vdwioffset0+vdwjidx0B,
 737                                          vdwparam+vdwioffset0+vdwjidx0C,
 738                                          vdwparam+vdwioffset0+vdwjidx0D,
 739                                          &c6_00,&c12_00);
 740
 741             /* Calculate table index by multiplying r with table scale and truncate to integer */
 742             rt               = _mm_mul_ps(r00,vftabscale);
 743             vfitab           = _mm_cvttps_epi32(rt);
 744             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 745             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 746
 747             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 748             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 749             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 750             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 751             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 752             _MM_TRANSPOSE4_PS(Y,F,G,H);
 753             Heps             = _mm_mul_ps(vfeps,H);
 754             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 755             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 756             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 757
 758             /* CUBIC SPLINE TABLE DISPERSION */
 759             vfitab           = _mm_add_epi32(vfitab,ifour);
 760             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 761             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 762             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 763             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 764             _MM_TRANSPOSE4_PS(Y,F,G,H);
 765             Heps             = _mm_mul_ps(vfeps,H);
 766             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 767             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 768             fvdw6            = _mm_mul_ps(c6_00,FF);
 769
 770             /* CUBIC SPLINE TABLE REPULSION */
 771             vfitab           = _mm_add_epi32(vfitab,ifour);
 772             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 773             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 774             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 775             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 776             _MM_TRANSPOSE4_PS(Y,F,G,H);
 777             Heps             = _mm_mul_ps(vfeps,H);
 778             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 779             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 780             fvdw12           = _mm_mul_ps(c12_00,FF);
 781             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 782
 783             fscal            = _mm_add_ps(felec,fvdw);
 784
 785             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 786
 787             /* Calculate temporary vectorial force */
 788             tx               = _mm_mul_ps(fscal,dx00);
 789             ty               = _mm_mul_ps(fscal,dy00);
 790             tz               = _mm_mul_ps(fscal,dz00);
 791
 792             /* Update vectorial force */
 793             fix0             = _mm_add_ps(fix0,tx);
 794             fiy0             = _mm_add_ps(fiy0,ty);
 795             fiz0             = _mm_add_ps(fiz0,tz);
 796
 797             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 798             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 799             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 800             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 801             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 802
 803             /* Inner loop uses 62 flops */
 804         }
 805
 806         /* End of innermost loop */
 807
 808         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 809                                               f+i_coord_offset,fshift+i_shift_offset);
 810
 811         /* Increment number of inner iterations */
 812         inneriter                  += j_index_end - j_index_start;
 813
 814         /* Outer loop uses 7 flops */
 815     }
 816
 817     /* Increment number of outer iterations */
 818     outeriter        += nri;
 819
 820     /* Update outer/inner flops */
 821
 822     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 823 }