src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_sse4_1_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 sse4_1_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_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse4_1_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse4_1_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 117     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 118     crf              = _mm_set1_ps(fr->ic->c_rf);
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 157
 158         fix0             = _mm_setzero_ps();
 159         fiy0             = _mm_setzero_ps();
 160         fiz0             = _mm_setzero_ps();
 161
 162         /* Load parameters for i particles */
 163         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 164         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168         vvdwsum          = _mm_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181             j_coord_offsetC  = DIM*jnrC;
 182             j_coord_offsetD  = DIM*jnrD;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               x+j_coord_offsetC,x+j_coord_offsetD,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_ps(ix0,jx0);
 191             dy00             = _mm_sub_ps(iy0,jy0);
 192             dz00             = _mm_sub_ps(iz0,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 196
 197             rinv00           = sse41_invsqrt_f(rsq00);
 198
 199             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 200
 201             /* Load parameters for j particles */
 202             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 203                                                               charge+jnrC+0,charge+jnrD+0);
 204             vdwjidx0A        = 2*vdwtype[jnrA+0];
 205             vdwjidx0B        = 2*vdwtype[jnrB+0];
 206             vdwjidx0C        = 2*vdwtype[jnrC+0];
 207             vdwjidx0D        = 2*vdwtype[jnrD+0];
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             r00              = _mm_mul_ps(rsq00,rinv00);
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             qq00             = _mm_mul_ps(iq0,jq0);
 217             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 218                                          vdwparam+vdwioffset0+vdwjidx0B,
 219                                          vdwparam+vdwioffset0+vdwjidx0C,
 220                                          vdwparam+vdwioffset0+vdwjidx0D,
 221                                          &c6_00,&c12_00);
 222
 223             /* Calculate table index by multiplying r with table scale and truncate to integer */
 224             rt               = _mm_mul_ps(r00,vftabscale);
 225             vfitab           = _mm_cvttps_epi32(rt);
 226             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 227             vfitab           = _mm_slli_epi32(vfitab,3);
 228
 229             /* REACTION-FIELD ELECTROSTATICS */
 230             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 231             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 232
 233             /* CUBIC SPLINE TABLE DISPERSION */
 234             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 235             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 236             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 237             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 238             _MM_TRANSPOSE4_PS(Y,F,G,H);
 239             Heps             = _mm_mul_ps(vfeps,H);
 240             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 241             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 242             vvdw6            = _mm_mul_ps(c6_00,VV);
 243             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 244             fvdw6            = _mm_mul_ps(c6_00,FF);
 245
 246             /* CUBIC SPLINE TABLE REPULSION */
 247             vfitab           = _mm_add_epi32(vfitab,ifour);
 248             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 249             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 250             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 251             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 252             _MM_TRANSPOSE4_PS(Y,F,G,H);
 253             Heps             = _mm_mul_ps(vfeps,H);
 254             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 255             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 256             vvdw12           = _mm_mul_ps(c12_00,VV);
 257             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 258             fvdw12           = _mm_mul_ps(c12_00,FF);
 259             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 260             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velecsum         = _mm_add_ps(velecsum,velec);
 264             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 265
 266             fscal            = _mm_add_ps(felec,fvdw);
 267
 268             /* Calculate temporary vectorial force */
 269             tx               = _mm_mul_ps(fscal,dx00);
 270             ty               = _mm_mul_ps(fscal,dy00);
 271             tz               = _mm_mul_ps(fscal,dz00);
 272
 273             /* Update vectorial force */
 274             fix0             = _mm_add_ps(fix0,tx);
 275             fiy0             = _mm_add_ps(fiy0,ty);
 276             fiz0             = _mm_add_ps(fiz0,tz);
 277
 278             fjptrA             = f+j_coord_offsetA;
 279             fjptrB             = f+j_coord_offsetB;
 280             fjptrC             = f+j_coord_offsetC;
 281             fjptrD             = f+j_coord_offsetD;
 282             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 283
 284             /* Inner loop uses 67 flops */
 285         }
 286
 287         if(jidx<j_index_end)
 288         {
 289
 290             /* Get j neighbor index, and coordinate index */
 291             jnrlistA         = jjnr[jidx];
 292             jnrlistB         = jjnr[jidx+1];
 293             jnrlistC         = jjnr[jidx+2];
 294             jnrlistD         = jjnr[jidx+3];
 295             /* Sign of each element will be negative for non-real atoms.
 296              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 297              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 298              */
 299             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 300             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 301             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 302             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 303             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 304             j_coord_offsetA  = DIM*jnrA;
 305             j_coord_offsetB  = DIM*jnrB;
 306             j_coord_offsetC  = DIM*jnrC;
 307             j_coord_offsetD  = DIM*jnrD;
 308
 309             /* load j atom coordinates */
 310             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 311                                               x+j_coord_offsetC,x+j_coord_offsetD,
 312                                               &jx0,&jy0,&jz0);
 313
 314             /* Calculate displacement vector */
 315             dx00             = _mm_sub_ps(ix0,jx0);
 316             dy00             = _mm_sub_ps(iy0,jy0);
 317             dz00             = _mm_sub_ps(iz0,jz0);
 318
 319             /* Calculate squared distance and things based on it */
 320             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 321
 322             rinv00           = sse41_invsqrt_f(rsq00);
 323
 324             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 325
 326             /* Load parameters for j particles */
 327             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 328                                                               charge+jnrC+0,charge+jnrD+0);
 329             vdwjidx0A        = 2*vdwtype[jnrA+0];
 330             vdwjidx0B        = 2*vdwtype[jnrB+0];
 331             vdwjidx0C        = 2*vdwtype[jnrC+0];
 332             vdwjidx0D        = 2*vdwtype[jnrD+0];
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             r00              = _mm_mul_ps(rsq00,rinv00);
 339             r00              = _mm_andnot_ps(dummy_mask,r00);
 340
 341             /* Compute parameters for interactions between i and j atoms */
 342             qq00             = _mm_mul_ps(iq0,jq0);
 343             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 344                                          vdwparam+vdwioffset0+vdwjidx0B,
 345                                          vdwparam+vdwioffset0+vdwjidx0C,
 346                                          vdwparam+vdwioffset0+vdwjidx0D,
 347                                          &c6_00,&c12_00);
 348
 349             /* Calculate table index by multiplying r with table scale and truncate to integer */
 350             rt               = _mm_mul_ps(r00,vftabscale);
 351             vfitab           = _mm_cvttps_epi32(rt);
 352             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 353             vfitab           = _mm_slli_epi32(vfitab,3);
 354
 355             /* REACTION-FIELD ELECTROSTATICS */
 356             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 357             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 358
 359             /* CUBIC SPLINE TABLE DISPERSION */
 360             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 361             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 362             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 363             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 364             _MM_TRANSPOSE4_PS(Y,F,G,H);
 365             Heps             = _mm_mul_ps(vfeps,H);
 366             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 367             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 368             vvdw6            = _mm_mul_ps(c6_00,VV);
 369             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 370             fvdw6            = _mm_mul_ps(c6_00,FF);
 371
 372             /* CUBIC SPLINE TABLE REPULSION */
 373             vfitab           = _mm_add_epi32(vfitab,ifour);
 374             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 375             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 376             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 377             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 378             _MM_TRANSPOSE4_PS(Y,F,G,H);
 379             Heps             = _mm_mul_ps(vfeps,H);
 380             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 381             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 382             vvdw12           = _mm_mul_ps(c12_00,VV);
 383             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 384             fvdw12           = _mm_mul_ps(c12_00,FF);
 385             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 386             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velec            = _mm_andnot_ps(dummy_mask,velec);
 390             velecsum         = _mm_add_ps(velecsum,velec);
 391             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 392             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 393
 394             fscal            = _mm_add_ps(felec,fvdw);
 395
 396             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx00);
 400             ty               = _mm_mul_ps(fscal,dy00);
 401             tz               = _mm_mul_ps(fscal,dz00);
 402
 403             /* Update vectorial force */
 404             fix0             = _mm_add_ps(fix0,tx);
 405             fiy0             = _mm_add_ps(fiy0,ty);
 406             fiz0             = _mm_add_ps(fiz0,tz);
 407
 408             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 409             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 410             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 411             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 412             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 413
 414             /* Inner loop uses 68 flops */
 415         }
 416
 417         /* End of innermost loop */
 418
 419         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 420                                               f+i_coord_offset,fshift+i_shift_offset);
 421
 422         ggid                        = gid[iidx];
 423         /* Update potential energies */
 424         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 425         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 426
 427         /* Increment number of inner iterations */
 428         inneriter                  += j_index_end - j_index_start;
 429
 430         /* Outer loop uses 9 flops */
 431     }
 432
 433     /* Increment number of outer iterations */
 434     outeriter        += nri;
 435
 436     /* Update outer/inner flops */
 437
 438     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 439 }
 440 /*
 441  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse4_1_single
 442  * Electrostatics interaction: ReactionField
 443  * VdW interaction:            CubicSplineTable
 444  * Geometry:                   Particle-Particle
 445  * Calculate force/pot:        Force
 446  */
 447 void
 448 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse4_1_single
 449                     (t_nblist                    * gmx_restrict       nlist,
 450                      rvec                        * gmx_restrict          xx,
 451                      rvec                        * gmx_restrict          ff,
 452                      struct t_forcerec           * gmx_restrict          fr,
 453                      t_mdatoms                   * gmx_restrict     mdatoms,
 454                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 455                      t_nrnb                      * gmx_restrict        nrnb)
 456 {
 457     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 458      * just 0 for non-waters.
 459      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 460      * jnr indices corresponding to data put in the four positions in the SIMD register.
 461      */
 462     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 463     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 464     int              jnrA,jnrB,jnrC,jnrD;
 465     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 466     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 467     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 468     real             rcutoff_scalar;
 469     real             *shiftvec,*fshift,*x,*f;
 470     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 471     real             scratch[4*DIM];
 472     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 473     int              vdwioffset0;
 474     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 475     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 476     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 477     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 478     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 479     real             *charge;
 480     int              nvdwtype;
 481     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 482     int              *vdwtype;
 483     real             *vdwparam;
 484     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 485     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 486     __m128i          vfitab;
 487     __m128i          ifour       = _mm_set1_epi32(4);
 488     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 489     real             *vftab;
 490     __m128           dummy_mask,cutoff_mask;
 491     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 492     __m128           one     = _mm_set1_ps(1.0);
 493     __m128           two     = _mm_set1_ps(2.0);
 494     x                = xx[0];
 495     f                = ff[0];
 496
 497     nri              = nlist->nri;
 498     iinr             = nlist->iinr;
 499     jindex           = nlist->jindex;
 500     jjnr             = nlist->jjnr;
 501     shiftidx         = nlist->shift;
 502     gid              = nlist->gid;
 503     shiftvec         = fr->shift_vec[0];
 504     fshift           = fr->fshift[0];
 505     facel            = _mm_set1_ps(fr->ic->epsfac);
 506     charge           = mdatoms->chargeA;
 507     krf              = _mm_set1_ps(fr->ic->k_rf);
 508     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 509     crf              = _mm_set1_ps(fr->ic->c_rf);
 510     nvdwtype         = fr->ntype;
 511     vdwparam         = fr->nbfp;
 512     vdwtype          = mdatoms->typeA;
 513
 514     vftab            = kernel_data->table_vdw->data;
 515     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 516
 517     /* Avoid stupid compiler warnings */
 518     jnrA = jnrB = jnrC = jnrD = 0;
 519     j_coord_offsetA = 0;
 520     j_coord_offsetB = 0;
 521     j_coord_offsetC = 0;
 522     j_coord_offsetD = 0;
 523
 524     outeriter        = 0;
 525     inneriter        = 0;
 526
 527     for(iidx=0;iidx<4*DIM;iidx++)
 528     {
 529         scratch[iidx] = 0.0;
 530     }
 531
 532     /* Start outer loop over neighborlists */
 533     for(iidx=0; iidx<nri; iidx++)
 534     {
 535         /* Load shift vector for this list */
 536         i_shift_offset   = DIM*shiftidx[iidx];
 537
 538         /* Load limits for loop over neighbors */
 539         j_index_start    = jindex[iidx];
 540         j_index_end      = jindex[iidx+1];
 541
 542         /* Get outer coordinate index */
 543         inr              = iinr[iidx];
 544         i_coord_offset   = DIM*inr;
 545
 546         /* Load i particle coords and add shift vector */
 547         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 548
 549         fix0             = _mm_setzero_ps();
 550         fiy0             = _mm_setzero_ps();
 551         fiz0             = _mm_setzero_ps();
 552
 553         /* Load parameters for i particles */
 554         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 555         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 556
 557         /* Start inner kernel loop */
 558         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 559         {
 560
 561             /* Get j neighbor index, and coordinate index */
 562             jnrA             = jjnr[jidx];
 563             jnrB             = jjnr[jidx+1];
 564             jnrC             = jjnr[jidx+2];
 565             jnrD             = jjnr[jidx+3];
 566             j_coord_offsetA  = DIM*jnrA;
 567             j_coord_offsetB  = DIM*jnrB;
 568             j_coord_offsetC  = DIM*jnrC;
 569             j_coord_offsetD  = DIM*jnrD;
 570
 571             /* load j atom coordinates */
 572             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 573                                               x+j_coord_offsetC,x+j_coord_offsetD,
 574                                               &jx0,&jy0,&jz0);
 575
 576             /* Calculate displacement vector */
 577             dx00             = _mm_sub_ps(ix0,jx0);
 578             dy00             = _mm_sub_ps(iy0,jy0);
 579             dz00             = _mm_sub_ps(iz0,jz0);
 580
 581             /* Calculate squared distance and things based on it */
 582             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 583
 584             rinv00           = sse41_invsqrt_f(rsq00);
 585
 586             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 587
 588             /* Load parameters for j particles */
 589             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 590                                                               charge+jnrC+0,charge+jnrD+0);
 591             vdwjidx0A        = 2*vdwtype[jnrA+0];
 592             vdwjidx0B        = 2*vdwtype[jnrB+0];
 593             vdwjidx0C        = 2*vdwtype[jnrC+0];
 594             vdwjidx0D        = 2*vdwtype[jnrD+0];
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             r00              = _mm_mul_ps(rsq00,rinv00);
 601
 602             /* Compute parameters for interactions between i and j atoms */
 603             qq00             = _mm_mul_ps(iq0,jq0);
 604             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 605                                          vdwparam+vdwioffset0+vdwjidx0B,
 606                                          vdwparam+vdwioffset0+vdwjidx0C,
 607                                          vdwparam+vdwioffset0+vdwjidx0D,
 608                                          &c6_00,&c12_00);
 609
 610             /* Calculate table index by multiplying r with table scale and truncate to integer */
 611             rt               = _mm_mul_ps(r00,vftabscale);
 612             vfitab           = _mm_cvttps_epi32(rt);
 613             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 614             vfitab           = _mm_slli_epi32(vfitab,3);
 615
 616             /* REACTION-FIELD ELECTROSTATICS */
 617             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 618
 619             /* CUBIC SPLINE TABLE DISPERSION */
 620             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 621             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 622             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 623             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 624             _MM_TRANSPOSE4_PS(Y,F,G,H);
 625             Heps             = _mm_mul_ps(vfeps,H);
 626             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 627             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 628             fvdw6            = _mm_mul_ps(c6_00,FF);
 629
 630             /* CUBIC SPLINE TABLE REPULSION */
 631             vfitab           = _mm_add_epi32(vfitab,ifour);
 632             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 633             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 634             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 635             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 636             _MM_TRANSPOSE4_PS(Y,F,G,H);
 637             Heps             = _mm_mul_ps(vfeps,H);
 638             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 639             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 640             fvdw12           = _mm_mul_ps(c12_00,FF);
 641             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 642
 643             fscal            = _mm_add_ps(felec,fvdw);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm_mul_ps(fscal,dx00);
 647             ty               = _mm_mul_ps(fscal,dy00);
 648             tz               = _mm_mul_ps(fscal,dz00);
 649
 650             /* Update vectorial force */
 651             fix0             = _mm_add_ps(fix0,tx);
 652             fiy0             = _mm_add_ps(fiy0,ty);
 653             fiz0             = _mm_add_ps(fiz0,tz);
 654
 655             fjptrA             = f+j_coord_offsetA;
 656             fjptrB             = f+j_coord_offsetB;
 657             fjptrC             = f+j_coord_offsetC;
 658             fjptrD             = f+j_coord_offsetD;
 659             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 660
 661             /* Inner loop uses 54 flops */
 662         }
 663
 664         if(jidx<j_index_end)
 665         {
 666
 667             /* Get j neighbor index, and coordinate index */
 668             jnrlistA         = jjnr[jidx];
 669             jnrlistB         = jjnr[jidx+1];
 670             jnrlistC         = jjnr[jidx+2];
 671             jnrlistD         = jjnr[jidx+3];
 672             /* Sign of each element will be negative for non-real atoms.
 673              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 674              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 675              */
 676             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 677             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 678             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 679             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 680             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 681             j_coord_offsetA  = DIM*jnrA;
 682             j_coord_offsetB  = DIM*jnrB;
 683             j_coord_offsetC  = DIM*jnrC;
 684             j_coord_offsetD  = DIM*jnrD;
 685
 686             /* load j atom coordinates */
 687             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 688                                               x+j_coord_offsetC,x+j_coord_offsetD,
 689                                               &jx0,&jy0,&jz0);
 690
 691             /* Calculate displacement vector */
 692             dx00             = _mm_sub_ps(ix0,jx0);
 693             dy00             = _mm_sub_ps(iy0,jy0);
 694             dz00             = _mm_sub_ps(iz0,jz0);
 695
 696             /* Calculate squared distance and things based on it */
 697             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 698
 699             rinv00           = sse41_invsqrt_f(rsq00);
 700
 701             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 702
 703             /* Load parameters for j particles */
 704             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 705                                                               charge+jnrC+0,charge+jnrD+0);
 706             vdwjidx0A        = 2*vdwtype[jnrA+0];
 707             vdwjidx0B        = 2*vdwtype[jnrB+0];
 708             vdwjidx0C        = 2*vdwtype[jnrC+0];
 709             vdwjidx0D        = 2*vdwtype[jnrD+0];
 710
 711             /**************************
 712              * CALCULATE INTERACTIONS *
 713              **************************/
 714
 715             r00              = _mm_mul_ps(rsq00,rinv00);
 716             r00              = _mm_andnot_ps(dummy_mask,r00);
 717
 718             /* Compute parameters for interactions between i and j atoms */
 719             qq00             = _mm_mul_ps(iq0,jq0);
 720             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 721                                          vdwparam+vdwioffset0+vdwjidx0B,
 722                                          vdwparam+vdwioffset0+vdwjidx0C,
 723                                          vdwparam+vdwioffset0+vdwjidx0D,
 724                                          &c6_00,&c12_00);
 725
 726             /* Calculate table index by multiplying r with table scale and truncate to integer */
 727             rt               = _mm_mul_ps(r00,vftabscale);
 728             vfitab           = _mm_cvttps_epi32(rt);
 729             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 730             vfitab           = _mm_slli_epi32(vfitab,3);
 731
 732             /* REACTION-FIELD ELECTROSTATICS */
 733             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 734
 735             /* CUBIC SPLINE TABLE DISPERSION */
 736             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 737             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 738             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 739             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 740             _MM_TRANSPOSE4_PS(Y,F,G,H);
 741             Heps             = _mm_mul_ps(vfeps,H);
 742             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 743             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 744             fvdw6            = _mm_mul_ps(c6_00,FF);
 745
 746             /* CUBIC SPLINE TABLE REPULSION */
 747             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             fvdw12           = _mm_mul_ps(c12_00,FF);
 757             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 758
 759             fscal            = _mm_add_ps(felec,fvdw);
 760
 761             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 762
 763             /* Calculate temporary vectorial force */
 764             tx               = _mm_mul_ps(fscal,dx00);
 765             ty               = _mm_mul_ps(fscal,dy00);
 766             tz               = _mm_mul_ps(fscal,dz00);
 767
 768             /* Update vectorial force */
 769             fix0             = _mm_add_ps(fix0,tx);
 770             fiy0             = _mm_add_ps(fiy0,ty);
 771             fiz0             = _mm_add_ps(fiz0,tz);
 772
 773             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 774             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 775             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 776             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 777             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 778
 779             /* Inner loop uses 55 flops */
 780         }
 781
 782         /* End of innermost loop */
 783
 784         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 785                                               f+i_coord_offset,fshift+i_shift_offset);
 786
 787         /* Increment number of inner iterations */
 788         inneriter                  += j_index_end - j_index_start;
 789
 790         /* Outer loop uses 7 flops */
 791     }
 792
 793     /* Increment number of outer iterations */
 794     outeriter        += nri;
 795
 796     /* Update outer/inner flops */
 797
 798     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);
 799 }