src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_sse4_1_single.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_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse4_1_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse4_1_single
  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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128i          vfitab;
  97     __m128i          ifour       = _mm_set1_epi32(4);
  98     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  99     real             *vftab;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(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_ps(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_ps(kernel_data->table_vdw->scale);
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     for(iidx=0;iidx<4*DIM;iidx++)
 135     {
 136         scratch[iidx] = 0.0;
 137     }
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _mm_setzero_ps();
 157         fiy0             = _mm_setzero_ps();
 158         fiz0             = _mm_setzero_ps();
 159
 160         /* Load parameters for i particles */
 161         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 162         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_ps();
 166         vvdwsum          = _mm_setzero_ps();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             jnrC             = jjnr[jidx+2];
 176             jnrD             = jjnr[jidx+3];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179             j_coord_offsetC  = DIM*jnrC;
 180             j_coord_offsetD  = DIM*jnrD;
 181
 182             /* load j atom coordinates */
 183             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 184                                               x+j_coord_offsetC,x+j_coord_offsetD,
 185                                               &jx0,&jy0,&jz0);
 186
 187             /* Calculate displacement vector */
 188             dx00             = _mm_sub_ps(ix0,jx0);
 189             dy00             = _mm_sub_ps(iy0,jy0);
 190             dz00             = _mm_sub_ps(iz0,jz0);
 191
 192             /* Calculate squared distance and things based on it */
 193             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 194
 195             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 196
 197             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 198
 199             /* Load parameters for j particles */
 200             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 201                                                               charge+jnrC+0,charge+jnrD+0);
 202             vdwjidx0A        = 2*vdwtype[jnrA+0];
 203             vdwjidx0B        = 2*vdwtype[jnrB+0];
 204             vdwjidx0C        = 2*vdwtype[jnrC+0];
 205             vdwjidx0D        = 2*vdwtype[jnrD+0];
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             r00              = _mm_mul_ps(rsq00,rinv00);
 212
 213             /* Compute parameters for interactions between i and j atoms */
 214             qq00             = _mm_mul_ps(iq0,jq0);
 215             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 216                                          vdwparam+vdwioffset0+vdwjidx0B,
 217                                          vdwparam+vdwioffset0+vdwjidx0C,
 218                                          vdwparam+vdwioffset0+vdwjidx0D,
 219                                          &c6_00,&c12_00);
 220
 221             /* Calculate table index by multiplying r with table scale and truncate to integer */
 222             rt               = _mm_mul_ps(r00,vftabscale);
 223             vfitab           = _mm_cvttps_epi32(rt);
 224             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 225             vfitab           = _mm_slli_epi32(vfitab,3);
 226
 227             /* COULOMB ELECTROSTATICS */
 228             velec            = _mm_mul_ps(qq00,rinv00);
 229             felec            = _mm_mul_ps(velec,rinvsq00);
 230
 231             /* CUBIC SPLINE TABLE DISPERSION */
 232             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 233             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 234             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 235             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 236             _MM_TRANSPOSE4_PS(Y,F,G,H);
 237             Heps             = _mm_mul_ps(vfeps,H);
 238             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 239             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 240             vvdw6            = _mm_mul_ps(c6_00,VV);
 241             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 242             fvdw6            = _mm_mul_ps(c6_00,FF);
 243
 244             /* CUBIC SPLINE TABLE REPULSION */
 245             vfitab           = _mm_add_epi32(vfitab,ifour);
 246             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 247             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 248             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 249             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 250             _MM_TRANSPOSE4_PS(Y,F,G,H);
 251             Heps             = _mm_mul_ps(vfeps,H);
 252             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 253             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 254             vvdw12           = _mm_mul_ps(c12_00,VV);
 255             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 256             fvdw12           = _mm_mul_ps(c12_00,FF);
 257             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 258             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velecsum         = _mm_add_ps(velecsum,velec);
 262             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 263
 264             fscal            = _mm_add_ps(felec,fvdw);
 265
 266             /* Calculate temporary vectorial force */
 267             tx               = _mm_mul_ps(fscal,dx00);
 268             ty               = _mm_mul_ps(fscal,dy00);
 269             tz               = _mm_mul_ps(fscal,dz00);
 270
 271             /* Update vectorial force */
 272             fix0             = _mm_add_ps(fix0,tx);
 273             fiy0             = _mm_add_ps(fiy0,ty);
 274             fiz0             = _mm_add_ps(fiz0,tz);
 275
 276             fjptrA             = f+j_coord_offsetA;
 277             fjptrB             = f+j_coord_offsetB;
 278             fjptrC             = f+j_coord_offsetC;
 279             fjptrD             = f+j_coord_offsetD;
 280             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 281
 282             /* Inner loop uses 63 flops */
 283         }
 284
 285         if(jidx<j_index_end)
 286         {
 287
 288             /* Get j neighbor index, and coordinate index */
 289             jnrlistA         = jjnr[jidx];
 290             jnrlistB         = jjnr[jidx+1];
 291             jnrlistC         = jjnr[jidx+2];
 292             jnrlistD         = jjnr[jidx+3];
 293             /* Sign of each element will be negative for non-real atoms.
 294              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 295              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 296              */
 297             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 298             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 299             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 300             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 301             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 302             j_coord_offsetA  = DIM*jnrA;
 303             j_coord_offsetB  = DIM*jnrB;
 304             j_coord_offsetC  = DIM*jnrC;
 305             j_coord_offsetD  = DIM*jnrD;
 306
 307             /* load j atom coordinates */
 308             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 309                                               x+j_coord_offsetC,x+j_coord_offsetD,
 310                                               &jx0,&jy0,&jz0);
 311
 312             /* Calculate displacement vector */
 313             dx00             = _mm_sub_ps(ix0,jx0);
 314             dy00             = _mm_sub_ps(iy0,jy0);
 315             dz00             = _mm_sub_ps(iz0,jz0);
 316
 317             /* Calculate squared distance and things based on it */
 318             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 319
 320             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 321
 322             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 323
 324             /* Load parameters for j particles */
 325             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 326                                                               charge+jnrC+0,charge+jnrD+0);
 327             vdwjidx0A        = 2*vdwtype[jnrA+0];
 328             vdwjidx0B        = 2*vdwtype[jnrB+0];
 329             vdwjidx0C        = 2*vdwtype[jnrC+0];
 330             vdwjidx0D        = 2*vdwtype[jnrD+0];
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             r00              = _mm_mul_ps(rsq00,rinv00);
 337             r00              = _mm_andnot_ps(dummy_mask,r00);
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq00             = _mm_mul_ps(iq0,jq0);
 341             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 342                                          vdwparam+vdwioffset0+vdwjidx0B,
 343                                          vdwparam+vdwioffset0+vdwjidx0C,
 344                                          vdwparam+vdwioffset0+vdwjidx0D,
 345                                          &c6_00,&c12_00);
 346
 347             /* Calculate table index by multiplying r with table scale and truncate to integer */
 348             rt               = _mm_mul_ps(r00,vftabscale);
 349             vfitab           = _mm_cvttps_epi32(rt);
 350             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 351             vfitab           = _mm_slli_epi32(vfitab,3);
 352
 353             /* COULOMB ELECTROSTATICS */
 354             velec            = _mm_mul_ps(qq00,rinv00);
 355             felec            = _mm_mul_ps(velec,rinvsq00);
 356
 357             /* CUBIC SPLINE TABLE DISPERSION */
 358             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 359             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 360             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 361             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 362             _MM_TRANSPOSE4_PS(Y,F,G,H);
 363             Heps             = _mm_mul_ps(vfeps,H);
 364             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 365             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 366             vvdw6            = _mm_mul_ps(c6_00,VV);
 367             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 368             fvdw6            = _mm_mul_ps(c6_00,FF);
 369
 370             /* CUBIC SPLINE TABLE REPULSION */
 371             vfitab           = _mm_add_epi32(vfitab,ifour);
 372             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 373             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 374             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 375             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 376             _MM_TRANSPOSE4_PS(Y,F,G,H);
 377             Heps             = _mm_mul_ps(vfeps,H);
 378             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 379             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 380             vvdw12           = _mm_mul_ps(c12_00,VV);
 381             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 382             fvdw12           = _mm_mul_ps(c12_00,FF);
 383             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 384             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm_andnot_ps(dummy_mask,velec);
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 390             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 391
 392             fscal            = _mm_add_ps(felec,fvdw);
 393
 394             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm_mul_ps(fscal,dx00);
 398             ty               = _mm_mul_ps(fscal,dy00);
 399             tz               = _mm_mul_ps(fscal,dz00);
 400
 401             /* Update vectorial force */
 402             fix0             = _mm_add_ps(fix0,tx);
 403             fiy0             = _mm_add_ps(fiy0,ty);
 404             fiz0             = _mm_add_ps(fiz0,tz);
 405
 406             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 407             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 408             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 409             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 410             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 411
 412             /* Inner loop uses 64 flops */
 413         }
 414
 415         /* End of innermost loop */
 416
 417         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 418                                               f+i_coord_offset,fshift+i_shift_offset);
 419
 420         ggid                        = gid[iidx];
 421         /* Update potential energies */
 422         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 423         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 424
 425         /* Increment number of inner iterations */
 426         inneriter                  += j_index_end - j_index_start;
 427
 428         /* Outer loop uses 9 flops */
 429     }
 430
 431     /* Increment number of outer iterations */
 432     outeriter        += nri;
 433
 434     /* Update outer/inner flops */
 435
 436     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*64);
 437 }
 438 /*
 439  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse4_1_single
 440  * Electrostatics interaction: Coulomb
 441  * VdW interaction:            CubicSplineTable
 442  * Geometry:                   Particle-Particle
 443  * Calculate force/pot:        Force
 444  */
 445 void
 446 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse4_1_single
 447                     (t_nblist                    * gmx_restrict       nlist,
 448                      rvec                        * gmx_restrict          xx,
 449                      rvec                        * gmx_restrict          ff,
 450                      t_forcerec                  * gmx_restrict          fr,
 451                      t_mdatoms                   * gmx_restrict     mdatoms,
 452                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 453                      t_nrnb                      * gmx_restrict        nrnb)
 454 {
 455     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 456      * just 0 for non-waters.
 457      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 458      * jnr indices corresponding to data put in the four positions in the SIMD register.
 459      */
 460     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 461     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 462     int              jnrA,jnrB,jnrC,jnrD;
 463     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 464     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 465     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 466     real             rcutoff_scalar;
 467     real             *shiftvec,*fshift,*x,*f;
 468     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 469     real             scratch[4*DIM];
 470     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 471     int              vdwioffset0;
 472     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 473     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 474     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 475     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 476     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 477     real             *charge;
 478     int              nvdwtype;
 479     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 480     int              *vdwtype;
 481     real             *vdwparam;
 482     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 483     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 484     __m128i          vfitab;
 485     __m128i          ifour       = _mm_set1_epi32(4);
 486     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 487     real             *vftab;
 488     __m128           dummy_mask,cutoff_mask;
 489     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 490     __m128           one     = _mm_set1_ps(1.0);
 491     __m128           two     = _mm_set1_ps(2.0);
 492     x                = xx[0];
 493     f                = ff[0];
 494
 495     nri              = nlist->nri;
 496     iinr             = nlist->iinr;
 497     jindex           = nlist->jindex;
 498     jjnr             = nlist->jjnr;
 499     shiftidx         = nlist->shift;
 500     gid              = nlist->gid;
 501     shiftvec         = fr->shift_vec[0];
 502     fshift           = fr->fshift[0];
 503     facel            = _mm_set1_ps(fr->epsfac);
 504     charge           = mdatoms->chargeA;
 505     nvdwtype         = fr->ntype;
 506     vdwparam         = fr->nbfp;
 507     vdwtype          = mdatoms->typeA;
 508
 509     vftab            = kernel_data->table_vdw->data;
 510     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 511
 512     /* Avoid stupid compiler warnings */
 513     jnrA = jnrB = jnrC = jnrD = 0;
 514     j_coord_offsetA = 0;
 515     j_coord_offsetB = 0;
 516     j_coord_offsetC = 0;
 517     j_coord_offsetD = 0;
 518
 519     outeriter        = 0;
 520     inneriter        = 0;
 521
 522     for(iidx=0;iidx<4*DIM;iidx++)
 523     {
 524         scratch[iidx] = 0.0;
 525     }
 526
 527     /* Start outer loop over neighborlists */
 528     for(iidx=0; iidx<nri; iidx++)
 529     {
 530         /* Load shift vector for this list */
 531         i_shift_offset   = DIM*shiftidx[iidx];
 532
 533         /* Load limits for loop over neighbors */
 534         j_index_start    = jindex[iidx];
 535         j_index_end      = jindex[iidx+1];
 536
 537         /* Get outer coordinate index */
 538         inr              = iinr[iidx];
 539         i_coord_offset   = DIM*inr;
 540
 541         /* Load i particle coords and add shift vector */
 542         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 543
 544         fix0             = _mm_setzero_ps();
 545         fiy0             = _mm_setzero_ps();
 546         fiz0             = _mm_setzero_ps();
 547
 548         /* Load parameters for i particles */
 549         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 550         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 551
 552         /* Start inner kernel loop */
 553         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 554         {
 555
 556             /* Get j neighbor index, and coordinate index */
 557             jnrA             = jjnr[jidx];
 558             jnrB             = jjnr[jidx+1];
 559             jnrC             = jjnr[jidx+2];
 560             jnrD             = jjnr[jidx+3];
 561             j_coord_offsetA  = DIM*jnrA;
 562             j_coord_offsetB  = DIM*jnrB;
 563             j_coord_offsetC  = DIM*jnrC;
 564             j_coord_offsetD  = DIM*jnrD;
 565
 566             /* load j atom coordinates */
 567             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 568                                               x+j_coord_offsetC,x+j_coord_offsetD,
 569                                               &jx0,&jy0,&jz0);
 570
 571             /* Calculate displacement vector */
 572             dx00             = _mm_sub_ps(ix0,jx0);
 573             dy00             = _mm_sub_ps(iy0,jy0);
 574             dz00             = _mm_sub_ps(iz0,jz0);
 575
 576             /* Calculate squared distance and things based on it */
 577             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 578
 579             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 580
 581             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 582
 583             /* Load parameters for j particles */
 584             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 585                                                               charge+jnrC+0,charge+jnrD+0);
 586             vdwjidx0A        = 2*vdwtype[jnrA+0];
 587             vdwjidx0B        = 2*vdwtype[jnrB+0];
 588             vdwjidx0C        = 2*vdwtype[jnrC+0];
 589             vdwjidx0D        = 2*vdwtype[jnrD+0];
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             r00              = _mm_mul_ps(rsq00,rinv00);
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             qq00             = _mm_mul_ps(iq0,jq0);
 599             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 600                                          vdwparam+vdwioffset0+vdwjidx0B,
 601                                          vdwparam+vdwioffset0+vdwjidx0C,
 602                                          vdwparam+vdwioffset0+vdwjidx0D,
 603                                          &c6_00,&c12_00);
 604
 605             /* Calculate table index by multiplying r with table scale and truncate to integer */
 606             rt               = _mm_mul_ps(r00,vftabscale);
 607             vfitab           = _mm_cvttps_epi32(rt);
 608             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 609             vfitab           = _mm_slli_epi32(vfitab,3);
 610
 611             /* COULOMB ELECTROSTATICS */
 612             velec            = _mm_mul_ps(qq00,rinv00);
 613             felec            = _mm_mul_ps(velec,rinvsq00);
 614
 615             /* CUBIC SPLINE TABLE DISPERSION */
 616             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 617             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 618             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 619             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 620             _MM_TRANSPOSE4_PS(Y,F,G,H);
 621             Heps             = _mm_mul_ps(vfeps,H);
 622             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 623             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 624             fvdw6            = _mm_mul_ps(c6_00,FF);
 625
 626             /* CUBIC SPLINE TABLE REPULSION */
 627             vfitab           = _mm_add_epi32(vfitab,ifour);
 628             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 629             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 630             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 631             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 632             _MM_TRANSPOSE4_PS(Y,F,G,H);
 633             Heps             = _mm_mul_ps(vfeps,H);
 634             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 635             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 636             fvdw12           = _mm_mul_ps(c12_00,FF);
 637             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 638
 639             fscal            = _mm_add_ps(felec,fvdw);
 640
 641             /* Calculate temporary vectorial force */
 642             tx               = _mm_mul_ps(fscal,dx00);
 643             ty               = _mm_mul_ps(fscal,dy00);
 644             tz               = _mm_mul_ps(fscal,dz00);
 645
 646             /* Update vectorial force */
 647             fix0             = _mm_add_ps(fix0,tx);
 648             fiy0             = _mm_add_ps(fiy0,ty);
 649             fiz0             = _mm_add_ps(fiz0,tz);
 650
 651             fjptrA             = f+j_coord_offsetA;
 652             fjptrB             = f+j_coord_offsetB;
 653             fjptrC             = f+j_coord_offsetC;
 654             fjptrD             = f+j_coord_offsetD;
 655             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 656
 657             /* Inner loop uses 54 flops */
 658         }
 659
 660         if(jidx<j_index_end)
 661         {
 662
 663             /* Get j neighbor index, and coordinate index */
 664             jnrlistA         = jjnr[jidx];
 665             jnrlistB         = jjnr[jidx+1];
 666             jnrlistC         = jjnr[jidx+2];
 667             jnrlistD         = jjnr[jidx+3];
 668             /* Sign of each element will be negative for non-real atoms.
 669              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 670              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 671              */
 672             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 673             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 674             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 675             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 676             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 677             j_coord_offsetA  = DIM*jnrA;
 678             j_coord_offsetB  = DIM*jnrB;
 679             j_coord_offsetC  = DIM*jnrC;
 680             j_coord_offsetD  = DIM*jnrD;
 681
 682             /* load j atom coordinates */
 683             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 684                                               x+j_coord_offsetC,x+j_coord_offsetD,
 685                                               &jx0,&jy0,&jz0);
 686
 687             /* Calculate displacement vector */
 688             dx00             = _mm_sub_ps(ix0,jx0);
 689             dy00             = _mm_sub_ps(iy0,jy0);
 690             dz00             = _mm_sub_ps(iz0,jz0);
 691
 692             /* Calculate squared distance and things based on it */
 693             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 694
 695             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 696
 697             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 698
 699             /* Load parameters for j particles */
 700             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 701                                                               charge+jnrC+0,charge+jnrD+0);
 702             vdwjidx0A        = 2*vdwtype[jnrA+0];
 703             vdwjidx0B        = 2*vdwtype[jnrB+0];
 704             vdwjidx0C        = 2*vdwtype[jnrC+0];
 705             vdwjidx0D        = 2*vdwtype[jnrD+0];
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             r00              = _mm_mul_ps(rsq00,rinv00);
 712             r00              = _mm_andnot_ps(dummy_mask,r00);
 713
 714             /* Compute parameters for interactions between i and j atoms */
 715             qq00             = _mm_mul_ps(iq0,jq0);
 716             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 717                                          vdwparam+vdwioffset0+vdwjidx0B,
 718                                          vdwparam+vdwioffset0+vdwjidx0C,
 719                                          vdwparam+vdwioffset0+vdwjidx0D,
 720                                          &c6_00,&c12_00);
 721
 722             /* Calculate table index by multiplying r with table scale and truncate to integer */
 723             rt               = _mm_mul_ps(r00,vftabscale);
 724             vfitab           = _mm_cvttps_epi32(rt);
 725             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 726             vfitab           = _mm_slli_epi32(vfitab,3);
 727
 728             /* COULOMB ELECTROSTATICS */
 729             velec            = _mm_mul_ps(qq00,rinv00);
 730             felec            = _mm_mul_ps(velec,rinvsq00);
 731
 732             /* CUBIC SPLINE TABLE DISPERSION */
 733             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 734             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 735             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 736             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 737             _MM_TRANSPOSE4_PS(Y,F,G,H);
 738             Heps             = _mm_mul_ps(vfeps,H);
 739             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 740             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 741             fvdw6            = _mm_mul_ps(c6_00,FF);
 742
 743             /* CUBIC SPLINE TABLE REPULSION */
 744             vfitab           = _mm_add_epi32(vfitab,ifour);
 745             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 746             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 747             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 748             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 749             _MM_TRANSPOSE4_PS(Y,F,G,H);
 750             Heps             = _mm_mul_ps(vfeps,H);
 751             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 752             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 753             fvdw12           = _mm_mul_ps(c12_00,FF);
 754             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 755
 756             fscal            = _mm_add_ps(felec,fvdw);
 757
 758             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 759
 760             /* Calculate temporary vectorial force */
 761             tx               = _mm_mul_ps(fscal,dx00);
 762             ty               = _mm_mul_ps(fscal,dy00);
 763             tz               = _mm_mul_ps(fscal,dz00);
 764
 765             /* Update vectorial force */
 766             fix0             = _mm_add_ps(fix0,tx);
 767             fiy0             = _mm_add_ps(fiy0,ty);
 768             fiz0             = _mm_add_ps(fiz0,tz);
 769
 770             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 771             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 772             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 773             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 774             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 775
 776             /* Inner loop uses 55 flops */
 777         }
 778
 779         /* End of innermost loop */
 780
 781         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 782                                               f+i_coord_offset,fshift+i_shift_offset);
 783
 784         /* Increment number of inner iterations */
 785         inneriter                  += j_index_end - j_index_start;
 786
 787         /* Outer loop uses 7 flops */
 788     }
 789
 790     /* Increment number of outer iterations */
 791     outeriter        += nri;
 792
 793     /* Update outer/inner flops */
 794
 795     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);
 796 }