src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_sse2_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse2_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     krf              = _mm_set1_ps(fr->ic->k_rf);
 123     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 124     crf              = _mm_set1_ps(fr->ic->c_rf);
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 140     rcutoff_scalar   = fr->ic->rcoulomb;
 141     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 142     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 176
 177         fix0             = _mm_setzero_ps();
 178         fiy0             = _mm_setzero_ps();
 179         fiz0             = _mm_setzero_ps();
 180         fix1             = _mm_setzero_ps();
 181         fiy1             = _mm_setzero_ps();
 182         fiz1             = _mm_setzero_ps();
 183         fix2             = _mm_setzero_ps();
 184         fiy2             = _mm_setzero_ps();
 185         fiz2             = _mm_setzero_ps();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm_setzero_ps();
 189         vvdwsum          = _mm_setzero_ps();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               x+j_coord_offsetC,x+j_coord_offsetD,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_ps(ix0,jx0);
 212             dy00             = _mm_sub_ps(iy0,jy0);
 213             dz00             = _mm_sub_ps(iz0,jz0);
 214             dx10             = _mm_sub_ps(ix1,jx0);
 215             dy10             = _mm_sub_ps(iy1,jy0);
 216             dz10             = _mm_sub_ps(iz1,jz0);
 217             dx20             = _mm_sub_ps(ix2,jx0);
 218             dy20             = _mm_sub_ps(iy2,jy0);
 219             dz20             = _mm_sub_ps(iz2,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 223             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 224             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 225
 226             rinv00           = sse2_invsqrt_f(rsq00);
 227             rinv10           = sse2_invsqrt_f(rsq10);
 228             rinv20           = sse2_invsqrt_f(rsq20);
 229
 230             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 231             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 232             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 236                                                               charge+jnrC+0,charge+jnrD+0);
 237             vdwjidx0A        = 2*vdwtype[jnrA+0];
 238             vdwjidx0B        = 2*vdwtype[jnrB+0];
 239             vdwjidx0C        = 2*vdwtype[jnrC+0];
 240             vdwjidx0D        = 2*vdwtype[jnrD+0];
 241
 242             fjx0             = _mm_setzero_ps();
 243             fjy0             = _mm_setzero_ps();
 244             fjz0             = _mm_setzero_ps();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             if (gmx_mm_any_lt(rsq00,rcutoff2))
 251             {
 252
 253             r00              = _mm_mul_ps(rsq00,rinv00);
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq00             = _mm_mul_ps(iq0,jq0);
 257             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 258                                          vdwparam+vdwioffset0+vdwjidx0B,
 259                                          vdwparam+vdwioffset0+vdwjidx0C,
 260                                          vdwparam+vdwioffset0+vdwjidx0D,
 261                                          &c6_00,&c12_00);
 262
 263             /* Calculate table index by multiplying r with table scale and truncate to integer */
 264             rt               = _mm_mul_ps(r00,vftabscale);
 265             vfitab           = _mm_cvttps_epi32(rt);
 266             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 267             vfitab           = _mm_slli_epi32(vfitab,3);
 268
 269             /* REACTION-FIELD ELECTROSTATICS */
 270             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 271             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 272
 273             /* CUBIC SPLINE TABLE DISPERSION */
 274             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 276             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 277             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 278             _MM_TRANSPOSE4_PS(Y,F,G,H);
 279             Heps             = _mm_mul_ps(vfeps,H);
 280             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 281             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 282             vvdw6            = _mm_mul_ps(c6_00,VV);
 283             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 284             fvdw6            = _mm_mul_ps(c6_00,FF);
 285
 286             /* CUBIC SPLINE TABLE REPULSION */
 287             vfitab           = _mm_add_epi32(vfitab,ifour);
 288             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 289             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 290             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 291             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 292             _MM_TRANSPOSE4_PS(Y,F,G,H);
 293             Heps             = _mm_mul_ps(vfeps,H);
 294             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 295             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 296             vvdw12           = _mm_mul_ps(c12_00,VV);
 297             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 298             fvdw12           = _mm_mul_ps(c12_00,FF);
 299             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 300             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 301
 302             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm_and_ps(velec,cutoff_mask);
 306             velecsum         = _mm_add_ps(velecsum,velec);
 307             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 308             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 309
 310             fscal            = _mm_add_ps(felec,fvdw);
 311
 312             fscal            = _mm_and_ps(fscal,cutoff_mask);
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm_mul_ps(fscal,dx00);
 316             ty               = _mm_mul_ps(fscal,dy00);
 317             tz               = _mm_mul_ps(fscal,dz00);
 318
 319             /* Update vectorial force */
 320             fix0             = _mm_add_ps(fix0,tx);
 321             fiy0             = _mm_add_ps(fiy0,ty);
 322             fiz0             = _mm_add_ps(fiz0,tz);
 323
 324             fjx0             = _mm_add_ps(fjx0,tx);
 325             fjy0             = _mm_add_ps(fjy0,ty);
 326             fjz0             = _mm_add_ps(fjz0,tz);
 327
 328             }
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             if (gmx_mm_any_lt(rsq10,rcutoff2))
 335             {
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq10             = _mm_mul_ps(iq1,jq0);
 339
 340             /* REACTION-FIELD ELECTROSTATICS */
 341             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 342             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 343
 344             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velec            = _mm_and_ps(velec,cutoff_mask);
 348             velecsum         = _mm_add_ps(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             fscal            = _mm_and_ps(fscal,cutoff_mask);
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm_mul_ps(fscal,dx10);
 356             ty               = _mm_mul_ps(fscal,dy10);
 357             tz               = _mm_mul_ps(fscal,dz10);
 358
 359             /* Update vectorial force */
 360             fix1             = _mm_add_ps(fix1,tx);
 361             fiy1             = _mm_add_ps(fiy1,ty);
 362             fiz1             = _mm_add_ps(fiz1,tz);
 363
 364             fjx0             = _mm_add_ps(fjx0,tx);
 365             fjy0             = _mm_add_ps(fjy0,ty);
 366             fjz0             = _mm_add_ps(fjz0,tz);
 367
 368             }
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             if (gmx_mm_any_lt(rsq20,rcutoff2))
 375             {
 376
 377             /* Compute parameters for interactions between i and j atoms */
 378             qq20             = _mm_mul_ps(iq2,jq0);
 379
 380             /* REACTION-FIELD ELECTROSTATICS */
 381             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 382             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 383
 384             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm_and_ps(velec,cutoff_mask);
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm_and_ps(fscal,cutoff_mask);
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm_mul_ps(fscal,dx20);
 396             ty               = _mm_mul_ps(fscal,dy20);
 397             tz               = _mm_mul_ps(fscal,dz20);
 398
 399             /* Update vectorial force */
 400             fix2             = _mm_add_ps(fix2,tx);
 401             fiy2             = _mm_add_ps(fiy2,ty);
 402             fiz2             = _mm_add_ps(fiz2,tz);
 403
 404             fjx0             = _mm_add_ps(fjx0,tx);
 405             fjy0             = _mm_add_ps(fjy0,ty);
 406             fjz0             = _mm_add_ps(fjz0,tz);
 407
 408             }
 409
 410             fjptrA             = f+j_coord_offsetA;
 411             fjptrB             = f+j_coord_offsetB;
 412             fjptrC             = f+j_coord_offsetC;
 413             fjptrD             = f+j_coord_offsetD;
 414
 415             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 416
 417             /* Inner loop uses 144 flops */
 418         }
 419
 420         if(jidx<j_index_end)
 421         {
 422
 423             /* Get j neighbor index, and coordinate index */
 424             jnrlistA         = jjnr[jidx];
 425             jnrlistB         = jjnr[jidx+1];
 426             jnrlistC         = jjnr[jidx+2];
 427             jnrlistD         = jjnr[jidx+3];
 428             /* Sign of each element will be negative for non-real atoms.
 429              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 430              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 431              */
 432             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 433             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 434             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 435             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 436             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 437             j_coord_offsetA  = DIM*jnrA;
 438             j_coord_offsetB  = DIM*jnrB;
 439             j_coord_offsetC  = DIM*jnrC;
 440             j_coord_offsetD  = DIM*jnrD;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                               x+j_coord_offsetC,x+j_coord_offsetD,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_ps(ix0,jx0);
 449             dy00             = _mm_sub_ps(iy0,jy0);
 450             dz00             = _mm_sub_ps(iz0,jz0);
 451             dx10             = _mm_sub_ps(ix1,jx0);
 452             dy10             = _mm_sub_ps(iy1,jy0);
 453             dz10             = _mm_sub_ps(iz1,jz0);
 454             dx20             = _mm_sub_ps(ix2,jx0);
 455             dy20             = _mm_sub_ps(iy2,jy0);
 456             dz20             = _mm_sub_ps(iz2,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 462
 463             rinv00           = sse2_invsqrt_f(rsq00);
 464             rinv10           = sse2_invsqrt_f(rsq10);
 465             rinv20           = sse2_invsqrt_f(rsq20);
 466
 467             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 468             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 470
 471             /* Load parameters for j particles */
 472             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 473                                                               charge+jnrC+0,charge+jnrD+0);
 474             vdwjidx0A        = 2*vdwtype[jnrA+0];
 475             vdwjidx0B        = 2*vdwtype[jnrB+0];
 476             vdwjidx0C        = 2*vdwtype[jnrC+0];
 477             vdwjidx0D        = 2*vdwtype[jnrD+0];
 478
 479             fjx0             = _mm_setzero_ps();
 480             fjy0             = _mm_setzero_ps();
 481             fjz0             = _mm_setzero_ps();
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             if (gmx_mm_any_lt(rsq00,rcutoff2))
 488             {
 489
 490             r00              = _mm_mul_ps(rsq00,rinv00);
 491             r00              = _mm_andnot_ps(dummy_mask,r00);
 492
 493             /* Compute parameters for interactions between i and j atoms */
 494             qq00             = _mm_mul_ps(iq0,jq0);
 495             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 496                                          vdwparam+vdwioffset0+vdwjidx0B,
 497                                          vdwparam+vdwioffset0+vdwjidx0C,
 498                                          vdwparam+vdwioffset0+vdwjidx0D,
 499                                          &c6_00,&c12_00);
 500
 501             /* Calculate table index by multiplying r with table scale and truncate to integer */
 502             rt               = _mm_mul_ps(r00,vftabscale);
 503             vfitab           = _mm_cvttps_epi32(rt);
 504             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 505             vfitab           = _mm_slli_epi32(vfitab,3);
 506
 507             /* REACTION-FIELD ELECTROSTATICS */
 508             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 509             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 510
 511             /* CUBIC SPLINE TABLE DISPERSION */
 512             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 513             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 514             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 515             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 516             _MM_TRANSPOSE4_PS(Y,F,G,H);
 517             Heps             = _mm_mul_ps(vfeps,H);
 518             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 519             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 520             vvdw6            = _mm_mul_ps(c6_00,VV);
 521             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 522             fvdw6            = _mm_mul_ps(c6_00,FF);
 523
 524             /* CUBIC SPLINE TABLE REPULSION */
 525             vfitab           = _mm_add_epi32(vfitab,ifour);
 526             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 527             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 528             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 529             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 530             _MM_TRANSPOSE4_PS(Y,F,G,H);
 531             Heps             = _mm_mul_ps(vfeps,H);
 532             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 533             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 534             vvdw12           = _mm_mul_ps(c12_00,VV);
 535             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 536             fvdw12           = _mm_mul_ps(c12_00,FF);
 537             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 538             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 539
 540             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 541
 542             /* Update potential sum for this i atom from the interaction with this j atom. */
 543             velec            = _mm_and_ps(velec,cutoff_mask);
 544             velec            = _mm_andnot_ps(dummy_mask,velec);
 545             velecsum         = _mm_add_ps(velecsum,velec);
 546             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 547             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 548             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 549
 550             fscal            = _mm_add_ps(felec,fvdw);
 551
 552             fscal            = _mm_and_ps(fscal,cutoff_mask);
 553
 554             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 555
 556             /* Calculate temporary vectorial force */
 557             tx               = _mm_mul_ps(fscal,dx00);
 558             ty               = _mm_mul_ps(fscal,dy00);
 559             tz               = _mm_mul_ps(fscal,dz00);
 560
 561             /* Update vectorial force */
 562             fix0             = _mm_add_ps(fix0,tx);
 563             fiy0             = _mm_add_ps(fiy0,ty);
 564             fiz0             = _mm_add_ps(fiz0,tz);
 565
 566             fjx0             = _mm_add_ps(fjx0,tx);
 567             fjy0             = _mm_add_ps(fjy0,ty);
 568             fjz0             = _mm_add_ps(fjz0,tz);
 569
 570             }
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             if (gmx_mm_any_lt(rsq10,rcutoff2))
 577             {
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             qq10             = _mm_mul_ps(iq1,jq0);
 581
 582             /* REACTION-FIELD ELECTROSTATICS */
 583             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 584             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 585
 586             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             velec            = _mm_and_ps(velec,cutoff_mask);
 590             velec            = _mm_andnot_ps(dummy_mask,velec);
 591             velecsum         = _mm_add_ps(velecsum,velec);
 592
 593             fscal            = felec;
 594
 595             fscal            = _mm_and_ps(fscal,cutoff_mask);
 596
 597             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm_mul_ps(fscal,dx10);
 601             ty               = _mm_mul_ps(fscal,dy10);
 602             tz               = _mm_mul_ps(fscal,dz10);
 603
 604             /* Update vectorial force */
 605             fix1             = _mm_add_ps(fix1,tx);
 606             fiy1             = _mm_add_ps(fiy1,ty);
 607             fiz1             = _mm_add_ps(fiz1,tz);
 608
 609             fjx0             = _mm_add_ps(fjx0,tx);
 610             fjy0             = _mm_add_ps(fjy0,ty);
 611             fjz0             = _mm_add_ps(fjz0,tz);
 612
 613             }
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             if (gmx_mm_any_lt(rsq20,rcutoff2))
 620             {
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq20             = _mm_mul_ps(iq2,jq0);
 624
 625             /* REACTION-FIELD ELECTROSTATICS */
 626             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 627             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 628
 629             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 630
 631             /* Update potential sum for this i atom from the interaction with this j atom. */
 632             velec            = _mm_and_ps(velec,cutoff_mask);
 633             velec            = _mm_andnot_ps(dummy_mask,velec);
 634             velecsum         = _mm_add_ps(velecsum,velec);
 635
 636             fscal            = felec;
 637
 638             fscal            = _mm_and_ps(fscal,cutoff_mask);
 639
 640             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 641
 642             /* Calculate temporary vectorial force */
 643             tx               = _mm_mul_ps(fscal,dx20);
 644             ty               = _mm_mul_ps(fscal,dy20);
 645             tz               = _mm_mul_ps(fscal,dz20);
 646
 647             /* Update vectorial force */
 648             fix2             = _mm_add_ps(fix2,tx);
 649             fiy2             = _mm_add_ps(fiy2,ty);
 650             fiz2             = _mm_add_ps(fiz2,tz);
 651
 652             fjx0             = _mm_add_ps(fjx0,tx);
 653             fjy0             = _mm_add_ps(fjy0,ty);
 654             fjz0             = _mm_add_ps(fjz0,tz);
 655
 656             }
 657
 658             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 659             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 660             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 661             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 662
 663             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 664
 665             /* Inner loop uses 145 flops */
 666         }
 667
 668         /* End of innermost loop */
 669
 670         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 671                                               f+i_coord_offset,fshift+i_shift_offset);
 672
 673         ggid                        = gid[iidx];
 674         /* Update potential energies */
 675         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 676         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 677
 678         /* Increment number of inner iterations */
 679         inneriter                  += j_index_end - j_index_start;
 680
 681         /* Outer loop uses 20 flops */
 682     }
 683
 684     /* Increment number of outer iterations */
 685     outeriter        += nri;
 686
 687     /* Update outer/inner flops */
 688
 689     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 690 }
 691 /*
 692  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse2_single
 693  * Electrostatics interaction: ReactionField
 694  * VdW interaction:            CubicSplineTable
 695  * Geometry:                   Water3-Particle
 696  * Calculate force/pot:        Force
 697  */
 698 void
 699 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse2_single
 700                     (t_nblist                    * gmx_restrict       nlist,
 701                      rvec                        * gmx_restrict          xx,
 702                      rvec                        * gmx_restrict          ff,
 703                      struct t_forcerec           * gmx_restrict          fr,
 704                      t_mdatoms                   * gmx_restrict     mdatoms,
 705                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 706                      t_nrnb                      * gmx_restrict        nrnb)
 707 {
 708     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 709      * just 0 for non-waters.
 710      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 711      * jnr indices corresponding to data put in the four positions in the SIMD register.
 712      */
 713     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 714     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 715     int              jnrA,jnrB,jnrC,jnrD;
 716     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 717     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 718     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 719     real             rcutoff_scalar;
 720     real             *shiftvec,*fshift,*x,*f;
 721     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 722     real             scratch[4*DIM];
 723     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 724     int              vdwioffset0;
 725     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 726     int              vdwioffset1;
 727     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 728     int              vdwioffset2;
 729     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 730     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 731     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 732     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 733     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 734     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 735     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 736     real             *charge;
 737     int              nvdwtype;
 738     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 739     int              *vdwtype;
 740     real             *vdwparam;
 741     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 742     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 743     __m128i          vfitab;
 744     __m128i          ifour       = _mm_set1_epi32(4);
 745     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 746     real             *vftab;
 747     __m128           dummy_mask,cutoff_mask;
 748     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 749     __m128           one     = _mm_set1_ps(1.0);
 750     __m128           two     = _mm_set1_ps(2.0);
 751     x                = xx[0];
 752     f                = ff[0];
 753
 754     nri              = nlist->nri;
 755     iinr             = nlist->iinr;
 756     jindex           = nlist->jindex;
 757     jjnr             = nlist->jjnr;
 758     shiftidx         = nlist->shift;
 759     gid              = nlist->gid;
 760     shiftvec         = fr->shift_vec[0];
 761     fshift           = fr->fshift[0];
 762     facel            = _mm_set1_ps(fr->ic->epsfac);
 763     charge           = mdatoms->chargeA;
 764     krf              = _mm_set1_ps(fr->ic->k_rf);
 765     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 766     crf              = _mm_set1_ps(fr->ic->c_rf);
 767     nvdwtype         = fr->ntype;
 768     vdwparam         = fr->nbfp;
 769     vdwtype          = mdatoms->typeA;
 770
 771     vftab            = kernel_data->table_vdw->data;
 772     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 773
 774     /* Setup water-specific parameters */
 775     inr              = nlist->iinr[0];
 776     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 777     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 778     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 779     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 780
 781     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 782     rcutoff_scalar   = fr->ic->rcoulomb;
 783     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 784     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 785
 786     /* Avoid stupid compiler warnings */
 787     jnrA = jnrB = jnrC = jnrD = 0;
 788     j_coord_offsetA = 0;
 789     j_coord_offsetB = 0;
 790     j_coord_offsetC = 0;
 791     j_coord_offsetD = 0;
 792
 793     outeriter        = 0;
 794     inneriter        = 0;
 795
 796     for(iidx=0;iidx<4*DIM;iidx++)
 797     {
 798         scratch[iidx] = 0.0;
 799     }
 800
 801     /* Start outer loop over neighborlists */
 802     for(iidx=0; iidx<nri; iidx++)
 803     {
 804         /* Load shift vector for this list */
 805         i_shift_offset   = DIM*shiftidx[iidx];
 806
 807         /* Load limits for loop over neighbors */
 808         j_index_start    = jindex[iidx];
 809         j_index_end      = jindex[iidx+1];
 810
 811         /* Get outer coordinate index */
 812         inr              = iinr[iidx];
 813         i_coord_offset   = DIM*inr;
 814
 815         /* Load i particle coords and add shift vector */
 816         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 817                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 818
 819         fix0             = _mm_setzero_ps();
 820         fiy0             = _mm_setzero_ps();
 821         fiz0             = _mm_setzero_ps();
 822         fix1             = _mm_setzero_ps();
 823         fiy1             = _mm_setzero_ps();
 824         fiz1             = _mm_setzero_ps();
 825         fix2             = _mm_setzero_ps();
 826         fiy2             = _mm_setzero_ps();
 827         fiz2             = _mm_setzero_ps();
 828
 829         /* Start inner kernel loop */
 830         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 831         {
 832
 833             /* Get j neighbor index, and coordinate index */
 834             jnrA             = jjnr[jidx];
 835             jnrB             = jjnr[jidx+1];
 836             jnrC             = jjnr[jidx+2];
 837             jnrD             = jjnr[jidx+3];
 838             j_coord_offsetA  = DIM*jnrA;
 839             j_coord_offsetB  = DIM*jnrB;
 840             j_coord_offsetC  = DIM*jnrC;
 841             j_coord_offsetD  = DIM*jnrD;
 842
 843             /* load j atom coordinates */
 844             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 845                                               x+j_coord_offsetC,x+j_coord_offsetD,
 846                                               &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx00             = _mm_sub_ps(ix0,jx0);
 850             dy00             = _mm_sub_ps(iy0,jy0);
 851             dz00             = _mm_sub_ps(iz0,jz0);
 852             dx10             = _mm_sub_ps(ix1,jx0);
 853             dy10             = _mm_sub_ps(iy1,jy0);
 854             dz10             = _mm_sub_ps(iz1,jz0);
 855             dx20             = _mm_sub_ps(ix2,jx0);
 856             dy20             = _mm_sub_ps(iy2,jy0);
 857             dz20             = _mm_sub_ps(iz2,jz0);
 858
 859             /* Calculate squared distance and things based on it */
 860             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 861             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 862             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 863
 864             rinv00           = sse2_invsqrt_f(rsq00);
 865             rinv10           = sse2_invsqrt_f(rsq10);
 866             rinv20           = sse2_invsqrt_f(rsq20);
 867
 868             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 869             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 870             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 871
 872             /* Load parameters for j particles */
 873             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 874                                                               charge+jnrC+0,charge+jnrD+0);
 875             vdwjidx0A        = 2*vdwtype[jnrA+0];
 876             vdwjidx0B        = 2*vdwtype[jnrB+0];
 877             vdwjidx0C        = 2*vdwtype[jnrC+0];
 878             vdwjidx0D        = 2*vdwtype[jnrD+0];
 879
 880             fjx0             = _mm_setzero_ps();
 881             fjy0             = _mm_setzero_ps();
 882             fjz0             = _mm_setzero_ps();
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             if (gmx_mm_any_lt(rsq00,rcutoff2))
 889             {
 890
 891             r00              = _mm_mul_ps(rsq00,rinv00);
 892
 893             /* Compute parameters for interactions between i and j atoms */
 894             qq00             = _mm_mul_ps(iq0,jq0);
 895             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 896                                          vdwparam+vdwioffset0+vdwjidx0B,
 897                                          vdwparam+vdwioffset0+vdwjidx0C,
 898                                          vdwparam+vdwioffset0+vdwjidx0D,
 899                                          &c6_00,&c12_00);
 900
 901             /* Calculate table index by multiplying r with table scale and truncate to integer */
 902             rt               = _mm_mul_ps(r00,vftabscale);
 903             vfitab           = _mm_cvttps_epi32(rt);
 904             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 905             vfitab           = _mm_slli_epi32(vfitab,3);
 906
 907             /* REACTION-FIELD ELECTROSTATICS */
 908             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 909
 910             /* CUBIC SPLINE TABLE DISPERSION */
 911             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 912             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 913             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 914             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 915             _MM_TRANSPOSE4_PS(Y,F,G,H);
 916             Heps             = _mm_mul_ps(vfeps,H);
 917             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 918             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 919             fvdw6            = _mm_mul_ps(c6_00,FF);
 920
 921             /* CUBIC SPLINE TABLE REPULSION */
 922             vfitab           = _mm_add_epi32(vfitab,ifour);
 923             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 924             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 925             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 926             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 927             _MM_TRANSPOSE4_PS(Y,F,G,H);
 928             Heps             = _mm_mul_ps(vfeps,H);
 929             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 930             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 931             fvdw12           = _mm_mul_ps(c12_00,FF);
 932             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 933
 934             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 935
 936             fscal            = _mm_add_ps(felec,fvdw);
 937
 938             fscal            = _mm_and_ps(fscal,cutoff_mask);
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm_mul_ps(fscal,dx00);
 942             ty               = _mm_mul_ps(fscal,dy00);
 943             tz               = _mm_mul_ps(fscal,dz00);
 944
 945             /* Update vectorial force */
 946             fix0             = _mm_add_ps(fix0,tx);
 947             fiy0             = _mm_add_ps(fiy0,ty);
 948             fiz0             = _mm_add_ps(fiz0,tz);
 949
 950             fjx0             = _mm_add_ps(fjx0,tx);
 951             fjy0             = _mm_add_ps(fjy0,ty);
 952             fjz0             = _mm_add_ps(fjz0,tz);
 953
 954             }
 955
 956             /**************************
 957              * CALCULATE INTERACTIONS *
 958              **************************/
 959
 960             if (gmx_mm_any_lt(rsq10,rcutoff2))
 961             {
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq10             = _mm_mul_ps(iq1,jq0);
 965
 966             /* REACTION-FIELD ELECTROSTATICS */
 967             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 968
 969             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 970
 971             fscal            = felec;
 972
 973             fscal            = _mm_and_ps(fscal,cutoff_mask);
 974
 975             /* Calculate temporary vectorial force */
 976             tx               = _mm_mul_ps(fscal,dx10);
 977             ty               = _mm_mul_ps(fscal,dy10);
 978             tz               = _mm_mul_ps(fscal,dz10);
 979
 980             /* Update vectorial force */
 981             fix1             = _mm_add_ps(fix1,tx);
 982             fiy1             = _mm_add_ps(fiy1,ty);
 983             fiz1             = _mm_add_ps(fiz1,tz);
 984
 985             fjx0             = _mm_add_ps(fjx0,tx);
 986             fjy0             = _mm_add_ps(fjy0,ty);
 987             fjz0             = _mm_add_ps(fjz0,tz);
 988
 989             }
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             if (gmx_mm_any_lt(rsq20,rcutoff2))
 996             {
 997
 998             /* Compute parameters for interactions between i and j atoms */
 999             qq20             = _mm_mul_ps(iq2,jq0);
1000
1001             /* REACTION-FIELD ELECTROSTATICS */
1002             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1003
1004             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1005
1006             fscal            = felec;
1007
1008             fscal            = _mm_and_ps(fscal,cutoff_mask);
1009
1010             /* Calculate temporary vectorial force */
1011             tx               = _mm_mul_ps(fscal,dx20);
1012             ty               = _mm_mul_ps(fscal,dy20);
1013             tz               = _mm_mul_ps(fscal,dz20);
1014
1015             /* Update vectorial force */
1016             fix2             = _mm_add_ps(fix2,tx);
1017             fiy2             = _mm_add_ps(fiy2,ty);
1018             fiz2             = _mm_add_ps(fiz2,tz);
1019
1020             fjx0             = _mm_add_ps(fjx0,tx);
1021             fjy0             = _mm_add_ps(fjy0,ty);
1022             fjz0             = _mm_add_ps(fjz0,tz);
1023
1024             }
1025
1026             fjptrA             = f+j_coord_offsetA;
1027             fjptrB             = f+j_coord_offsetB;
1028             fjptrC             = f+j_coord_offsetC;
1029             fjptrD             = f+j_coord_offsetD;
1030
1031             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1032
1033             /* Inner loop uses 117 flops */
1034         }
1035
1036         if(jidx<j_index_end)
1037         {
1038
1039             /* Get j neighbor index, and coordinate index */
1040             jnrlistA         = jjnr[jidx];
1041             jnrlistB         = jjnr[jidx+1];
1042             jnrlistC         = jjnr[jidx+2];
1043             jnrlistD         = jjnr[jidx+3];
1044             /* Sign of each element will be negative for non-real atoms.
1045              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1046              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1047              */
1048             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1049             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1050             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1051             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1052             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1053             j_coord_offsetA  = DIM*jnrA;
1054             j_coord_offsetB  = DIM*jnrB;
1055             j_coord_offsetC  = DIM*jnrC;
1056             j_coord_offsetD  = DIM*jnrD;
1057
1058             /* load j atom coordinates */
1059             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1060                                               x+j_coord_offsetC,x+j_coord_offsetD,
1061                                               &jx0,&jy0,&jz0);
1062
1063             /* Calculate displacement vector */
1064             dx00             = _mm_sub_ps(ix0,jx0);
1065             dy00             = _mm_sub_ps(iy0,jy0);
1066             dz00             = _mm_sub_ps(iz0,jz0);
1067             dx10             = _mm_sub_ps(ix1,jx0);
1068             dy10             = _mm_sub_ps(iy1,jy0);
1069             dz10             = _mm_sub_ps(iz1,jz0);
1070             dx20             = _mm_sub_ps(ix2,jx0);
1071             dy20             = _mm_sub_ps(iy2,jy0);
1072             dz20             = _mm_sub_ps(iz2,jz0);
1073
1074             /* Calculate squared distance and things based on it */
1075             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1076             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1077             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1078
1079             rinv00           = sse2_invsqrt_f(rsq00);
1080             rinv10           = sse2_invsqrt_f(rsq10);
1081             rinv20           = sse2_invsqrt_f(rsq20);
1082
1083             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1084             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1085             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1086
1087             /* Load parameters for j particles */
1088             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1089                                                               charge+jnrC+0,charge+jnrD+0);
1090             vdwjidx0A        = 2*vdwtype[jnrA+0];
1091             vdwjidx0B        = 2*vdwtype[jnrB+0];
1092             vdwjidx0C        = 2*vdwtype[jnrC+0];
1093             vdwjidx0D        = 2*vdwtype[jnrD+0];
1094
1095             fjx0             = _mm_setzero_ps();
1096             fjy0             = _mm_setzero_ps();
1097             fjz0             = _mm_setzero_ps();
1098
1099             /**************************
1100              * CALCULATE INTERACTIONS *
1101              **************************/
1102
1103             if (gmx_mm_any_lt(rsq00,rcutoff2))
1104             {
1105
1106             r00              = _mm_mul_ps(rsq00,rinv00);
1107             r00              = _mm_andnot_ps(dummy_mask,r00);
1108
1109             /* Compute parameters for interactions between i and j atoms */
1110             qq00             = _mm_mul_ps(iq0,jq0);
1111             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1112                                          vdwparam+vdwioffset0+vdwjidx0B,
1113                                          vdwparam+vdwioffset0+vdwjidx0C,
1114                                          vdwparam+vdwioffset0+vdwjidx0D,
1115                                          &c6_00,&c12_00);
1116
1117             /* Calculate table index by multiplying r with table scale and truncate to integer */
1118             rt               = _mm_mul_ps(r00,vftabscale);
1119             vfitab           = _mm_cvttps_epi32(rt);
1120             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1121             vfitab           = _mm_slli_epi32(vfitab,3);
1122
1123             /* REACTION-FIELD ELECTROSTATICS */
1124             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1125
1126             /* CUBIC SPLINE TABLE DISPERSION */
1127             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1128             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1129             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1130             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1131             _MM_TRANSPOSE4_PS(Y,F,G,H);
1132             Heps             = _mm_mul_ps(vfeps,H);
1133             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1134             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1135             fvdw6            = _mm_mul_ps(c6_00,FF);
1136
1137             /* CUBIC SPLINE TABLE REPULSION */
1138             vfitab           = _mm_add_epi32(vfitab,ifour);
1139             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1140             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1141             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1142             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1143             _MM_TRANSPOSE4_PS(Y,F,G,H);
1144             Heps             = _mm_mul_ps(vfeps,H);
1145             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1146             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1147             fvdw12           = _mm_mul_ps(c12_00,FF);
1148             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1149
1150             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1151
1152             fscal            = _mm_add_ps(felec,fvdw);
1153
1154             fscal            = _mm_and_ps(fscal,cutoff_mask);
1155
1156             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1157
1158             /* Calculate temporary vectorial force */
1159             tx               = _mm_mul_ps(fscal,dx00);
1160             ty               = _mm_mul_ps(fscal,dy00);
1161             tz               = _mm_mul_ps(fscal,dz00);
1162
1163             /* Update vectorial force */
1164             fix0             = _mm_add_ps(fix0,tx);
1165             fiy0             = _mm_add_ps(fiy0,ty);
1166             fiz0             = _mm_add_ps(fiz0,tz);
1167
1168             fjx0             = _mm_add_ps(fjx0,tx);
1169             fjy0             = _mm_add_ps(fjy0,ty);
1170             fjz0             = _mm_add_ps(fjz0,tz);
1171
1172             }
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             if (gmx_mm_any_lt(rsq10,rcutoff2))
1179             {
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq10             = _mm_mul_ps(iq1,jq0);
1183
1184             /* REACTION-FIELD ELECTROSTATICS */
1185             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1186
1187             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1188
1189             fscal            = felec;
1190
1191             fscal            = _mm_and_ps(fscal,cutoff_mask);
1192
1193             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1194
1195             /* Calculate temporary vectorial force */
1196             tx               = _mm_mul_ps(fscal,dx10);
1197             ty               = _mm_mul_ps(fscal,dy10);
1198             tz               = _mm_mul_ps(fscal,dz10);
1199
1200             /* Update vectorial force */
1201             fix1             = _mm_add_ps(fix1,tx);
1202             fiy1             = _mm_add_ps(fiy1,ty);
1203             fiz1             = _mm_add_ps(fiz1,tz);
1204
1205             fjx0             = _mm_add_ps(fjx0,tx);
1206             fjy0             = _mm_add_ps(fjy0,ty);
1207             fjz0             = _mm_add_ps(fjz0,tz);
1208
1209             }
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             if (gmx_mm_any_lt(rsq20,rcutoff2))
1216             {
1217
1218             /* Compute parameters for interactions between i and j atoms */
1219             qq20             = _mm_mul_ps(iq2,jq0);
1220
1221             /* REACTION-FIELD ELECTROSTATICS */
1222             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1223
1224             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1225
1226             fscal            = felec;
1227
1228             fscal            = _mm_and_ps(fscal,cutoff_mask);
1229
1230             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1231
1232             /* Calculate temporary vectorial force */
1233             tx               = _mm_mul_ps(fscal,dx20);
1234             ty               = _mm_mul_ps(fscal,dy20);
1235             tz               = _mm_mul_ps(fscal,dz20);
1236
1237             /* Update vectorial force */
1238             fix2             = _mm_add_ps(fix2,tx);
1239             fiy2             = _mm_add_ps(fiy2,ty);
1240             fiz2             = _mm_add_ps(fiz2,tz);
1241
1242             fjx0             = _mm_add_ps(fjx0,tx);
1243             fjy0             = _mm_add_ps(fjy0,ty);
1244             fjz0             = _mm_add_ps(fjz0,tz);
1245
1246             }
1247
1248             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1249             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1250             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1251             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1252
1253             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1254
1255             /* Inner loop uses 118 flops */
1256         }
1257
1258         /* End of innermost loop */
1259
1260         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1261                                               f+i_coord_offset,fshift+i_shift_offset);
1262
1263         /* Increment number of inner iterations */
1264         inneriter                  += j_index_end - j_index_start;
1265
1266         /* Outer loop uses 18 flops */
1267     }
1268
1269     /* Increment number of outer iterations */
1270     outeriter        += nri;
1271
1272     /* Update outer/inner flops */
1273
1274     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1275 }