src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse4_1_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_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              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_vdw->data;
 128     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_ps();
 171         fiy0             = _mm_setzero_ps();
 172         fiz0             = _mm_setzero_ps();
 173         fix1             = _mm_setzero_ps();
 174         fiy1             = _mm_setzero_ps();
 175         fiz1             = _mm_setzero_ps();
 176         fix2             = _mm_setzero_ps();
 177         fiy2             = _mm_setzero_ps();
 178         fiz2             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 216             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 217             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 218
 219             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 220             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 221             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 222
 223             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 224             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 225             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                               charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             fjx0             = _mm_setzero_ps();
 236             fjy0             = _mm_setzero_ps();
 237             fjz0             = _mm_setzero_ps();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm_mul_ps(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm_mul_ps(iq0,jq0);
 247             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,
 249                                          vdwparam+vdwioffset0+vdwjidx0C,
 250                                          vdwparam+vdwioffset0+vdwjidx0D,
 251                                          &c6_00,&c12_00);
 252
 253             /* Calculate table index by multiplying r with table scale and truncate to integer */
 254             rt               = _mm_mul_ps(r00,vftabscale);
 255             vfitab           = _mm_cvttps_epi32(rt);
 256             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 257             vfitab           = _mm_slli_epi32(vfitab,3);
 258
 259             /* COULOMB ELECTROSTATICS */
 260             velec            = _mm_mul_ps(qq00,rinv00);
 261             felec            = _mm_mul_ps(velec,rinvsq00);
 262
 263             /* CUBIC SPLINE TABLE DISPERSION */
 264             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 265             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 266             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 267             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 268             _MM_TRANSPOSE4_PS(Y,F,G,H);
 269             Heps             = _mm_mul_ps(vfeps,H);
 270             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 271             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 272             vvdw6            = _mm_mul_ps(c6_00,VV);
 273             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 274             fvdw6            = _mm_mul_ps(c6_00,FF);
 275
 276             /* CUBIC SPLINE TABLE REPULSION */
 277             vfitab           = _mm_add_epi32(vfitab,ifour);
 278             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 279             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 280             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 281             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 282             _MM_TRANSPOSE4_PS(Y,F,G,H);
 283             Heps             = _mm_mul_ps(vfeps,H);
 284             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 285             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 286             vvdw12           = _mm_mul_ps(c12_00,VV);
 287             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 288             fvdw12           = _mm_mul_ps(c12_00,FF);
 289             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 290             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velecsum         = _mm_add_ps(velecsum,velec);
 294             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 295
 296             fscal            = _mm_add_ps(felec,fvdw);
 297
 298             /* Calculate temporary vectorial force */
 299             tx               = _mm_mul_ps(fscal,dx00);
 300             ty               = _mm_mul_ps(fscal,dy00);
 301             tz               = _mm_mul_ps(fscal,dz00);
 302
 303             /* Update vectorial force */
 304             fix0             = _mm_add_ps(fix0,tx);
 305             fiy0             = _mm_add_ps(fiy0,ty);
 306             fiz0             = _mm_add_ps(fiz0,tz);
 307
 308             fjx0             = _mm_add_ps(fjx0,tx);
 309             fjy0             = _mm_add_ps(fjy0,ty);
 310             fjz0             = _mm_add_ps(fjz0,tz);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             /* Compute parameters for interactions between i and j atoms */
 317             qq10             = _mm_mul_ps(iq1,jq0);
 318
 319             /* COULOMB ELECTROSTATICS */
 320             velec            = _mm_mul_ps(qq10,rinv10);
 321             felec            = _mm_mul_ps(velec,rinvsq10);
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velecsum         = _mm_add_ps(velecsum,velec);
 325
 326             fscal            = felec;
 327
 328             /* Calculate temporary vectorial force */
 329             tx               = _mm_mul_ps(fscal,dx10);
 330             ty               = _mm_mul_ps(fscal,dy10);
 331             tz               = _mm_mul_ps(fscal,dz10);
 332
 333             /* Update vectorial force */
 334             fix1             = _mm_add_ps(fix1,tx);
 335             fiy1             = _mm_add_ps(fiy1,ty);
 336             fiz1             = _mm_add_ps(fiz1,tz);
 337
 338             fjx0             = _mm_add_ps(fjx0,tx);
 339             fjy0             = _mm_add_ps(fjy0,ty);
 340             fjz0             = _mm_add_ps(fjz0,tz);
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             /* Compute parameters for interactions between i and j atoms */
 347             qq20             = _mm_mul_ps(iq2,jq0);
 348
 349             /* COULOMB ELECTROSTATICS */
 350             velec            = _mm_mul_ps(qq20,rinv20);
 351             felec            = _mm_mul_ps(velec,rinvsq20);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velecsum         = _mm_add_ps(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358             /* Calculate temporary vectorial force */
 359             tx               = _mm_mul_ps(fscal,dx20);
 360             ty               = _mm_mul_ps(fscal,dy20);
 361             tz               = _mm_mul_ps(fscal,dz20);
 362
 363             /* Update vectorial force */
 364             fix2             = _mm_add_ps(fix2,tx);
 365             fiy2             = _mm_add_ps(fiy2,ty);
 366             fiz2             = _mm_add_ps(fiz2,tz);
 367
 368             fjx0             = _mm_add_ps(fjx0,tx);
 369             fjy0             = _mm_add_ps(fjy0,ty);
 370             fjz0             = _mm_add_ps(fjz0,tz);
 371
 372             fjptrA             = f+j_coord_offsetA;
 373             fjptrB             = f+j_coord_offsetB;
 374             fjptrC             = f+j_coord_offsetC;
 375             fjptrD             = f+j_coord_offsetD;
 376
 377             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 378
 379             /* Inner loop uses 119 flops */
 380         }
 381
 382         if(jidx<j_index_end)
 383         {
 384
 385             /* Get j neighbor index, and coordinate index */
 386             jnrlistA         = jjnr[jidx];
 387             jnrlistB         = jjnr[jidx+1];
 388             jnrlistC         = jjnr[jidx+2];
 389             jnrlistD         = jjnr[jidx+3];
 390             /* Sign of each element will be negative for non-real atoms.
 391              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 392              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 393              */
 394             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 395             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 396             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 397             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 398             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 399             j_coord_offsetA  = DIM*jnrA;
 400             j_coord_offsetB  = DIM*jnrB;
 401             j_coord_offsetC  = DIM*jnrC;
 402             j_coord_offsetD  = DIM*jnrD;
 403
 404             /* load j atom coordinates */
 405             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 406                                               x+j_coord_offsetC,x+j_coord_offsetD,
 407                                               &jx0,&jy0,&jz0);
 408
 409             /* Calculate displacement vector */
 410             dx00             = _mm_sub_ps(ix0,jx0);
 411             dy00             = _mm_sub_ps(iy0,jy0);
 412             dz00             = _mm_sub_ps(iz0,jz0);
 413             dx10             = _mm_sub_ps(ix1,jx0);
 414             dy10             = _mm_sub_ps(iy1,jy0);
 415             dz10             = _mm_sub_ps(iz1,jz0);
 416             dx20             = _mm_sub_ps(ix2,jx0);
 417             dy20             = _mm_sub_ps(iy2,jy0);
 418             dz20             = _mm_sub_ps(iz2,jz0);
 419
 420             /* Calculate squared distance and things based on it */
 421             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 422             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 423             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 424
 425             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 426             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 427             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 428
 429             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 430             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 431             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 432
 433             /* Load parameters for j particles */
 434             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 435                                                               charge+jnrC+0,charge+jnrD+0);
 436             vdwjidx0A        = 2*vdwtype[jnrA+0];
 437             vdwjidx0B        = 2*vdwtype[jnrB+0];
 438             vdwjidx0C        = 2*vdwtype[jnrC+0];
 439             vdwjidx0D        = 2*vdwtype[jnrD+0];
 440
 441             fjx0             = _mm_setzero_ps();
 442             fjy0             = _mm_setzero_ps();
 443             fjz0             = _mm_setzero_ps();
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             r00              = _mm_mul_ps(rsq00,rinv00);
 450             r00              = _mm_andnot_ps(dummy_mask,r00);
 451
 452             /* Compute parameters for interactions between i and j atoms */
 453             qq00             = _mm_mul_ps(iq0,jq0);
 454             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 455                                          vdwparam+vdwioffset0+vdwjidx0B,
 456                                          vdwparam+vdwioffset0+vdwjidx0C,
 457                                          vdwparam+vdwioffset0+vdwjidx0D,
 458                                          &c6_00,&c12_00);
 459
 460             /* Calculate table index by multiplying r with table scale and truncate to integer */
 461             rt               = _mm_mul_ps(r00,vftabscale);
 462             vfitab           = _mm_cvttps_epi32(rt);
 463             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 464             vfitab           = _mm_slli_epi32(vfitab,3);
 465
 466             /* COULOMB ELECTROSTATICS */
 467             velec            = _mm_mul_ps(qq00,rinv00);
 468             felec            = _mm_mul_ps(velec,rinvsq00);
 469
 470             /* CUBIC SPLINE TABLE DISPERSION */
 471             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 472             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 473             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 474             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 475             _MM_TRANSPOSE4_PS(Y,F,G,H);
 476             Heps             = _mm_mul_ps(vfeps,H);
 477             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 478             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 479             vvdw6            = _mm_mul_ps(c6_00,VV);
 480             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 481             fvdw6            = _mm_mul_ps(c6_00,FF);
 482
 483             /* CUBIC SPLINE TABLE REPULSION */
 484             vfitab           = _mm_add_epi32(vfitab,ifour);
 485             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 486             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 487             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 488             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 489             _MM_TRANSPOSE4_PS(Y,F,G,H);
 490             Heps             = _mm_mul_ps(vfeps,H);
 491             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 492             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 493             vvdw12           = _mm_mul_ps(c12_00,VV);
 494             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 495             fvdw12           = _mm_mul_ps(c12_00,FF);
 496             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 497             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 498
 499             /* Update potential sum for this i atom from the interaction with this j atom. */
 500             velec            = _mm_andnot_ps(dummy_mask,velec);
 501             velecsum         = _mm_add_ps(velecsum,velec);
 502             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 503             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 504
 505             fscal            = _mm_add_ps(felec,fvdw);
 506
 507             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm_mul_ps(fscal,dx00);
 511             ty               = _mm_mul_ps(fscal,dy00);
 512             tz               = _mm_mul_ps(fscal,dz00);
 513
 514             /* Update vectorial force */
 515             fix0             = _mm_add_ps(fix0,tx);
 516             fiy0             = _mm_add_ps(fiy0,ty);
 517             fiz0             = _mm_add_ps(fiz0,tz);
 518
 519             fjx0             = _mm_add_ps(fjx0,tx);
 520             fjy0             = _mm_add_ps(fjy0,ty);
 521             fjz0             = _mm_add_ps(fjz0,tz);
 522
 523             /**************************
 524              * CALCULATE INTERACTIONS *
 525              **************************/
 526
 527             /* Compute parameters for interactions between i and j atoms */
 528             qq10             = _mm_mul_ps(iq1,jq0);
 529
 530             /* COULOMB ELECTROSTATICS */
 531             velec            = _mm_mul_ps(qq10,rinv10);
 532             felec            = _mm_mul_ps(velec,rinvsq10);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_andnot_ps(dummy_mask,velec);
 536             velecsum         = _mm_add_ps(velecsum,velec);
 537
 538             fscal            = felec;
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_ps(fscal,dx10);
 544             ty               = _mm_mul_ps(fscal,dy10);
 545             tz               = _mm_mul_ps(fscal,dz10);
 546
 547             /* Update vectorial force */
 548             fix1             = _mm_add_ps(fix1,tx);
 549             fiy1             = _mm_add_ps(fiy1,ty);
 550             fiz1             = _mm_add_ps(fiz1,tz);
 551
 552             fjx0             = _mm_add_ps(fjx0,tx);
 553             fjy0             = _mm_add_ps(fjy0,ty);
 554             fjz0             = _mm_add_ps(fjz0,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             qq20             = _mm_mul_ps(iq2,jq0);
 562
 563             /* COULOMB ELECTROSTATICS */
 564             velec            = _mm_mul_ps(qq20,rinv20);
 565             felec            = _mm_mul_ps(velec,rinvsq20);
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _mm_andnot_ps(dummy_mask,velec);
 569             velecsum         = _mm_add_ps(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_ps(fscal,dx20);
 577             ty               = _mm_mul_ps(fscal,dy20);
 578             tz               = _mm_mul_ps(fscal,dz20);
 579
 580             /* Update vectorial force */
 581             fix2             = _mm_add_ps(fix2,tx);
 582             fiy2             = _mm_add_ps(fiy2,ty);
 583             fiz2             = _mm_add_ps(fiz2,tz);
 584
 585             fjx0             = _mm_add_ps(fjx0,tx);
 586             fjy0             = _mm_add_ps(fjy0,ty);
 587             fjz0             = _mm_add_ps(fjz0,tz);
 588
 589             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 590             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 591             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 592             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 593
 594             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 595
 596             /* Inner loop uses 120 flops */
 597         }
 598
 599         /* End of innermost loop */
 600
 601         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 602                                               f+i_coord_offset,fshift+i_shift_offset);
 603
 604         ggid                        = gid[iidx];
 605         /* Update potential energies */
 606         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 607         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 608
 609         /* Increment number of inner iterations */
 610         inneriter                  += j_index_end - j_index_start;
 611
 612         /* Outer loop uses 20 flops */
 613     }
 614
 615     /* Increment number of outer iterations */
 616     outeriter        += nri;
 617
 618     /* Update outer/inner flops */
 619
 620     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 621 }
 622 /*
 623  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_single
 624  * Electrostatics interaction: Coulomb
 625  * VdW interaction:            CubicSplineTable
 626  * Geometry:                   Water3-Particle
 627  * Calculate force/pot:        Force
 628  */
 629 void
 630 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_single
 631                     (t_nblist                    * gmx_restrict       nlist,
 632                      rvec                        * gmx_restrict          xx,
 633                      rvec                        * gmx_restrict          ff,
 634                      t_forcerec                  * gmx_restrict          fr,
 635                      t_mdatoms                   * gmx_restrict     mdatoms,
 636                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 637                      t_nrnb                      * gmx_restrict        nrnb)
 638 {
 639     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 640      * just 0 for non-waters.
 641      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 642      * jnr indices corresponding to data put in the four positions in the SIMD register.
 643      */
 644     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 645     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 646     int              jnrA,jnrB,jnrC,jnrD;
 647     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 648     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 649     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 650     real             rcutoff_scalar;
 651     real             *shiftvec,*fshift,*x,*f;
 652     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 653     real             scratch[4*DIM];
 654     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 655     int              vdwioffset0;
 656     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 657     int              vdwioffset1;
 658     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 659     int              vdwioffset2;
 660     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 661     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 662     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 663     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 664     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 665     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 666     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 667     real             *charge;
 668     int              nvdwtype;
 669     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 670     int              *vdwtype;
 671     real             *vdwparam;
 672     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 673     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 674     __m128i          vfitab;
 675     __m128i          ifour       = _mm_set1_epi32(4);
 676     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 677     real             *vftab;
 678     __m128           dummy_mask,cutoff_mask;
 679     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 680     __m128           one     = _mm_set1_ps(1.0);
 681     __m128           two     = _mm_set1_ps(2.0);
 682     x                = xx[0];
 683     f                = ff[0];
 684
 685     nri              = nlist->nri;
 686     iinr             = nlist->iinr;
 687     jindex           = nlist->jindex;
 688     jjnr             = nlist->jjnr;
 689     shiftidx         = nlist->shift;
 690     gid              = nlist->gid;
 691     shiftvec         = fr->shift_vec[0];
 692     fshift           = fr->fshift[0];
 693     facel            = _mm_set1_ps(fr->epsfac);
 694     charge           = mdatoms->chargeA;
 695     nvdwtype         = fr->ntype;
 696     vdwparam         = fr->nbfp;
 697     vdwtype          = mdatoms->typeA;
 698
 699     vftab            = kernel_data->table_vdw->data;
 700     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 701
 702     /* Setup water-specific parameters */
 703     inr              = nlist->iinr[0];
 704     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 705     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 706     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 707     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 708
 709     /* Avoid stupid compiler warnings */
 710     jnrA = jnrB = jnrC = jnrD = 0;
 711     j_coord_offsetA = 0;
 712     j_coord_offsetB = 0;
 713     j_coord_offsetC = 0;
 714     j_coord_offsetD = 0;
 715
 716     outeriter        = 0;
 717     inneriter        = 0;
 718
 719     for(iidx=0;iidx<4*DIM;iidx++)
 720     {
 721         scratch[iidx] = 0.0;
 722     }
 723
 724     /* Start outer loop over neighborlists */
 725     for(iidx=0; iidx<nri; iidx++)
 726     {
 727         /* Load shift vector for this list */
 728         i_shift_offset   = DIM*shiftidx[iidx];
 729
 730         /* Load limits for loop over neighbors */
 731         j_index_start    = jindex[iidx];
 732         j_index_end      = jindex[iidx+1];
 733
 734         /* Get outer coordinate index */
 735         inr              = iinr[iidx];
 736         i_coord_offset   = DIM*inr;
 737
 738         /* Load i particle coords and add shift vector */
 739         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 740                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 741
 742         fix0             = _mm_setzero_ps();
 743         fiy0             = _mm_setzero_ps();
 744         fiz0             = _mm_setzero_ps();
 745         fix1             = _mm_setzero_ps();
 746         fiy1             = _mm_setzero_ps();
 747         fiz1             = _mm_setzero_ps();
 748         fix2             = _mm_setzero_ps();
 749         fiy2             = _mm_setzero_ps();
 750         fiz2             = _mm_setzero_ps();
 751
 752         /* Start inner kernel loop */
 753         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 754         {
 755
 756             /* Get j neighbor index, and coordinate index */
 757             jnrA             = jjnr[jidx];
 758             jnrB             = jjnr[jidx+1];
 759             jnrC             = jjnr[jidx+2];
 760             jnrD             = jjnr[jidx+3];
 761             j_coord_offsetA  = DIM*jnrA;
 762             j_coord_offsetB  = DIM*jnrB;
 763             j_coord_offsetC  = DIM*jnrC;
 764             j_coord_offsetD  = DIM*jnrD;
 765
 766             /* load j atom coordinates */
 767             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 768                                               x+j_coord_offsetC,x+j_coord_offsetD,
 769                                               &jx0,&jy0,&jz0);
 770
 771             /* Calculate displacement vector */
 772             dx00             = _mm_sub_ps(ix0,jx0);
 773             dy00             = _mm_sub_ps(iy0,jy0);
 774             dz00             = _mm_sub_ps(iz0,jz0);
 775             dx10             = _mm_sub_ps(ix1,jx0);
 776             dy10             = _mm_sub_ps(iy1,jy0);
 777             dz10             = _mm_sub_ps(iz1,jz0);
 778             dx20             = _mm_sub_ps(ix2,jx0);
 779             dy20             = _mm_sub_ps(iy2,jy0);
 780             dz20             = _mm_sub_ps(iz2,jz0);
 781
 782             /* Calculate squared distance and things based on it */
 783             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 784             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 785             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 786
 787             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 788             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 789             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 790
 791             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 792             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 793             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 794
 795             /* Load parameters for j particles */
 796             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 797                                                               charge+jnrC+0,charge+jnrD+0);
 798             vdwjidx0A        = 2*vdwtype[jnrA+0];
 799             vdwjidx0B        = 2*vdwtype[jnrB+0];
 800             vdwjidx0C        = 2*vdwtype[jnrC+0];
 801             vdwjidx0D        = 2*vdwtype[jnrD+0];
 802
 803             fjx0             = _mm_setzero_ps();
 804             fjy0             = _mm_setzero_ps();
 805             fjz0             = _mm_setzero_ps();
 806
 807             /**************************
 808              * CALCULATE INTERACTIONS *
 809              **************************/
 810
 811             r00              = _mm_mul_ps(rsq00,rinv00);
 812
 813             /* Compute parameters for interactions between i and j atoms */
 814             qq00             = _mm_mul_ps(iq0,jq0);
 815             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 816                                          vdwparam+vdwioffset0+vdwjidx0B,
 817                                          vdwparam+vdwioffset0+vdwjidx0C,
 818                                          vdwparam+vdwioffset0+vdwjidx0D,
 819                                          &c6_00,&c12_00);
 820
 821             /* Calculate table index by multiplying r with table scale and truncate to integer */
 822             rt               = _mm_mul_ps(r00,vftabscale);
 823             vfitab           = _mm_cvttps_epi32(rt);
 824             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 825             vfitab           = _mm_slli_epi32(vfitab,3);
 826
 827             /* COULOMB ELECTROSTATICS */
 828             velec            = _mm_mul_ps(qq00,rinv00);
 829             felec            = _mm_mul_ps(velec,rinvsq00);
 830
 831             /* CUBIC SPLINE TABLE DISPERSION */
 832             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 833             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 834             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 835             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 836             _MM_TRANSPOSE4_PS(Y,F,G,H);
 837             Heps             = _mm_mul_ps(vfeps,H);
 838             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 839             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 840             fvdw6            = _mm_mul_ps(c6_00,FF);
 841
 842             /* CUBIC SPLINE TABLE REPULSION */
 843             vfitab           = _mm_add_epi32(vfitab,ifour);
 844             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 845             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 846             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 847             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 848             _MM_TRANSPOSE4_PS(Y,F,G,H);
 849             Heps             = _mm_mul_ps(vfeps,H);
 850             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 851             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 852             fvdw12           = _mm_mul_ps(c12_00,FF);
 853             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 854
 855             fscal            = _mm_add_ps(felec,fvdw);
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm_mul_ps(fscal,dx00);
 859             ty               = _mm_mul_ps(fscal,dy00);
 860             tz               = _mm_mul_ps(fscal,dz00);
 861
 862             /* Update vectorial force */
 863             fix0             = _mm_add_ps(fix0,tx);
 864             fiy0             = _mm_add_ps(fiy0,ty);
 865             fiz0             = _mm_add_ps(fiz0,tz);
 866
 867             fjx0             = _mm_add_ps(fjx0,tx);
 868             fjy0             = _mm_add_ps(fjy0,ty);
 869             fjz0             = _mm_add_ps(fjz0,tz);
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             /* Compute parameters for interactions between i and j atoms */
 876             qq10             = _mm_mul_ps(iq1,jq0);
 877
 878             /* COULOMB ELECTROSTATICS */
 879             velec            = _mm_mul_ps(qq10,rinv10);
 880             felec            = _mm_mul_ps(velec,rinvsq10);
 881
 882             fscal            = felec;
 883
 884             /* Calculate temporary vectorial force */
 885             tx               = _mm_mul_ps(fscal,dx10);
 886             ty               = _mm_mul_ps(fscal,dy10);
 887             tz               = _mm_mul_ps(fscal,dz10);
 888
 889             /* Update vectorial force */
 890             fix1             = _mm_add_ps(fix1,tx);
 891             fiy1             = _mm_add_ps(fiy1,ty);
 892             fiz1             = _mm_add_ps(fiz1,tz);
 893
 894             fjx0             = _mm_add_ps(fjx0,tx);
 895             fjy0             = _mm_add_ps(fjy0,ty);
 896             fjz0             = _mm_add_ps(fjz0,tz);
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq20             = _mm_mul_ps(iq2,jq0);
 904
 905             /* COULOMB ELECTROSTATICS */
 906             velec            = _mm_mul_ps(qq20,rinv20);
 907             felec            = _mm_mul_ps(velec,rinvsq20);
 908
 909             fscal            = felec;
 910
 911             /* Calculate temporary vectorial force */
 912             tx               = _mm_mul_ps(fscal,dx20);
 913             ty               = _mm_mul_ps(fscal,dy20);
 914             tz               = _mm_mul_ps(fscal,dz20);
 915
 916             /* Update vectorial force */
 917             fix2             = _mm_add_ps(fix2,tx);
 918             fiy2             = _mm_add_ps(fiy2,ty);
 919             fiz2             = _mm_add_ps(fiz2,tz);
 920
 921             fjx0             = _mm_add_ps(fjx0,tx);
 922             fjy0             = _mm_add_ps(fjy0,ty);
 923             fjz0             = _mm_add_ps(fjz0,tz);
 924
 925             fjptrA             = f+j_coord_offsetA;
 926             fjptrB             = f+j_coord_offsetB;
 927             fjptrC             = f+j_coord_offsetC;
 928             fjptrD             = f+j_coord_offsetD;
 929
 930             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 931
 932             /* Inner loop uses 108 flops */
 933         }
 934
 935         if(jidx<j_index_end)
 936         {
 937
 938             /* Get j neighbor index, and coordinate index */
 939             jnrlistA         = jjnr[jidx];
 940             jnrlistB         = jjnr[jidx+1];
 941             jnrlistC         = jjnr[jidx+2];
 942             jnrlistD         = jjnr[jidx+3];
 943             /* Sign of each element will be negative for non-real atoms.
 944              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 945              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 946              */
 947             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 948             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 949             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 950             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 951             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 952             j_coord_offsetA  = DIM*jnrA;
 953             j_coord_offsetB  = DIM*jnrB;
 954             j_coord_offsetC  = DIM*jnrC;
 955             j_coord_offsetD  = DIM*jnrD;
 956
 957             /* load j atom coordinates */
 958             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 959                                               x+j_coord_offsetC,x+j_coord_offsetD,
 960                                               &jx0,&jy0,&jz0);
 961
 962             /* Calculate displacement vector */
 963             dx00             = _mm_sub_ps(ix0,jx0);
 964             dy00             = _mm_sub_ps(iy0,jy0);
 965             dz00             = _mm_sub_ps(iz0,jz0);
 966             dx10             = _mm_sub_ps(ix1,jx0);
 967             dy10             = _mm_sub_ps(iy1,jy0);
 968             dz10             = _mm_sub_ps(iz1,jz0);
 969             dx20             = _mm_sub_ps(ix2,jx0);
 970             dy20             = _mm_sub_ps(iy2,jy0);
 971             dz20             = _mm_sub_ps(iz2,jz0);
 972
 973             /* Calculate squared distance and things based on it */
 974             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 975             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 976             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 977
 978             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 979             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 980             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 981
 982             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 983             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 984             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 985
 986             /* Load parameters for j particles */
 987             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 988                                                               charge+jnrC+0,charge+jnrD+0);
 989             vdwjidx0A        = 2*vdwtype[jnrA+0];
 990             vdwjidx0B        = 2*vdwtype[jnrB+0];
 991             vdwjidx0C        = 2*vdwtype[jnrC+0];
 992             vdwjidx0D        = 2*vdwtype[jnrD+0];
 993
 994             fjx0             = _mm_setzero_ps();
 995             fjy0             = _mm_setzero_ps();
 996             fjz0             = _mm_setzero_ps();
 997
 998             /**************************
 999              * CALCULATE INTERACTIONS *
1000              **************************/
1001
1002             r00              = _mm_mul_ps(rsq00,rinv00);
1003             r00              = _mm_andnot_ps(dummy_mask,r00);
1004
1005             /* Compute parameters for interactions between i and j atoms */
1006             qq00             = _mm_mul_ps(iq0,jq0);
1007             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1008                                          vdwparam+vdwioffset0+vdwjidx0B,
1009                                          vdwparam+vdwioffset0+vdwjidx0C,
1010                                          vdwparam+vdwioffset0+vdwjidx0D,
1011                                          &c6_00,&c12_00);
1012
1013             /* Calculate table index by multiplying r with table scale and truncate to integer */
1014             rt               = _mm_mul_ps(r00,vftabscale);
1015             vfitab           = _mm_cvttps_epi32(rt);
1016             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1017             vfitab           = _mm_slli_epi32(vfitab,3);
1018
1019             /* COULOMB ELECTROSTATICS */
1020             velec            = _mm_mul_ps(qq00,rinv00);
1021             felec            = _mm_mul_ps(velec,rinvsq00);
1022
1023             /* CUBIC SPLINE TABLE DISPERSION */
1024             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1025             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1026             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1027             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1028             _MM_TRANSPOSE4_PS(Y,F,G,H);
1029             Heps             = _mm_mul_ps(vfeps,H);
1030             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1031             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1032             fvdw6            = _mm_mul_ps(c6_00,FF);
1033
1034             /* CUBIC SPLINE TABLE REPULSION */
1035             vfitab           = _mm_add_epi32(vfitab,ifour);
1036             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1037             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1038             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1039             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1040             _MM_TRANSPOSE4_PS(Y,F,G,H);
1041             Heps             = _mm_mul_ps(vfeps,H);
1042             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1043             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1044             fvdw12           = _mm_mul_ps(c12_00,FF);
1045             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1046
1047             fscal            = _mm_add_ps(felec,fvdw);
1048
1049             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm_mul_ps(fscal,dx00);
1053             ty               = _mm_mul_ps(fscal,dy00);
1054             tz               = _mm_mul_ps(fscal,dz00);
1055
1056             /* Update vectorial force */
1057             fix0             = _mm_add_ps(fix0,tx);
1058             fiy0             = _mm_add_ps(fiy0,ty);
1059             fiz0             = _mm_add_ps(fiz0,tz);
1060
1061             fjx0             = _mm_add_ps(fjx0,tx);
1062             fjy0             = _mm_add_ps(fjy0,ty);
1063             fjz0             = _mm_add_ps(fjz0,tz);
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq10             = _mm_mul_ps(iq1,jq0);
1071
1072             /* COULOMB ELECTROSTATICS */
1073             velec            = _mm_mul_ps(qq10,rinv10);
1074             felec            = _mm_mul_ps(velec,rinvsq10);
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm_mul_ps(fscal,dx10);
1082             ty               = _mm_mul_ps(fscal,dy10);
1083             tz               = _mm_mul_ps(fscal,dz10);
1084
1085             /* Update vectorial force */
1086             fix1             = _mm_add_ps(fix1,tx);
1087             fiy1             = _mm_add_ps(fiy1,ty);
1088             fiz1             = _mm_add_ps(fiz1,tz);
1089
1090             fjx0             = _mm_add_ps(fjx0,tx);
1091             fjy0             = _mm_add_ps(fjy0,ty);
1092             fjz0             = _mm_add_ps(fjz0,tz);
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq20             = _mm_mul_ps(iq2,jq0);
1100
1101             /* COULOMB ELECTROSTATICS */
1102             velec            = _mm_mul_ps(qq20,rinv20);
1103             felec            = _mm_mul_ps(velec,rinvsq20);
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm_mul_ps(fscal,dx20);
1111             ty               = _mm_mul_ps(fscal,dy20);
1112             tz               = _mm_mul_ps(fscal,dz20);
1113
1114             /* Update vectorial force */
1115             fix2             = _mm_add_ps(fix2,tx);
1116             fiy2             = _mm_add_ps(fiy2,ty);
1117             fiz2             = _mm_add_ps(fiz2,tz);
1118
1119             fjx0             = _mm_add_ps(fjx0,tx);
1120             fjy0             = _mm_add_ps(fjy0,ty);
1121             fjz0             = _mm_add_ps(fjz0,tz);
1122
1123             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1124             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1125             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1126             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1127
1128             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1129
1130             /* Inner loop uses 109 flops */
1131         }
1132
1133         /* End of innermost loop */
1134
1135         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1136                                               f+i_coord_offset,fshift+i_shift_offset);
1137
1138         /* Increment number of inner iterations */
1139         inneriter                  += j_index_end - j_index_start;
1140
1141         /* Outer loop uses 18 flops */
1142     }
1143
1144     /* Increment number of outer iterations */
1145     outeriter        += nri;
1146
1147     /* Update outer/inner flops */
1148
1149     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1150 }