src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_sse2_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 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_elec->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     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_ps();
 184         vvdwsum          = _mm_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               x+j_coord_offsetC,x+j_coord_offsetD,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_ps(ix0,jx0);
 207             dy00             = _mm_sub_ps(iy0,jy0);
 208             dz00             = _mm_sub_ps(iz0,jz0);
 209             dx10             = _mm_sub_ps(ix1,jx0);
 210             dy10             = _mm_sub_ps(iy1,jy0);
 211             dz10             = _mm_sub_ps(iz1,jz0);
 212             dx20             = _mm_sub_ps(ix2,jx0);
 213             dy20             = _mm_sub_ps(iy2,jy0);
 214             dz20             = _mm_sub_ps(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 220
 221             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224
 225             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 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_cvtepi32_ps(vfitab));
 257             vfitab           = _mm_slli_epi32(vfitab,2);
 258
 259             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 260             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 261             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 262             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 263             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 264             _MM_TRANSPOSE4_PS(Y,F,G,H);
 265             Heps             = _mm_mul_ps(vfeps,H);
 266             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 267             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 268             velec            = _mm_mul_ps(qq00,VV);
 269             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 270             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 271
 272             /* LENNARD-JONES DISPERSION/REPULSION */
 273
 274             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 275             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 276             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 277             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 278             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm_add_ps(velecsum,velec);
 282             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 283
 284             fscal            = _mm_add_ps(felec,fvdw);
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_ps(fscal,dx00);
 288             ty               = _mm_mul_ps(fscal,dy00);
 289             tz               = _mm_mul_ps(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_add_ps(fix0,tx);
 293             fiy0             = _mm_add_ps(fiy0,ty);
 294             fiz0             = _mm_add_ps(fiz0,tz);
 295
 296             fjx0             = _mm_add_ps(fjx0,tx);
 297             fjy0             = _mm_add_ps(fjy0,ty);
 298             fjz0             = _mm_add_ps(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r10              = _mm_mul_ps(rsq10,rinv10);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq10             = _mm_mul_ps(iq1,jq0);
 308
 309             /* Calculate table index by multiplying r with table scale and truncate to integer */
 310             rt               = _mm_mul_ps(r10,vftabscale);
 311             vfitab           = _mm_cvttps_epi32(rt);
 312             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 313             vfitab           = _mm_slli_epi32(vfitab,2);
 314
 315             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 316             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 317             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 318             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 319             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 320             _MM_TRANSPOSE4_PS(Y,F,G,H);
 321             Heps             = _mm_mul_ps(vfeps,H);
 322             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 323             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 324             velec            = _mm_mul_ps(qq10,VV);
 325             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 326             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_ps(fscal,dx10);
 335             ty               = _mm_mul_ps(fscal,dy10);
 336             tz               = _mm_mul_ps(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm_add_ps(fix1,tx);
 340             fiy1             = _mm_add_ps(fiy1,ty);
 341             fiz1             = _mm_add_ps(fiz1,tz);
 342
 343             fjx0             = _mm_add_ps(fjx0,tx);
 344             fjy0             = _mm_add_ps(fjy0,ty);
 345             fjz0             = _mm_add_ps(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r20              = _mm_mul_ps(rsq20,rinv20);
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_ps(iq2,jq0);
 355
 356             /* Calculate table index by multiplying r with table scale and truncate to integer */
 357             rt               = _mm_mul_ps(r20,vftabscale);
 358             vfitab           = _mm_cvttps_epi32(rt);
 359             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 360             vfitab           = _mm_slli_epi32(vfitab,2);
 361
 362             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 363             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 364             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 365             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 366             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 367             _MM_TRANSPOSE4_PS(Y,F,G,H);
 368             Heps             = _mm_mul_ps(vfeps,H);
 369             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 370             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 371             velec            = _mm_mul_ps(qq20,VV);
 372             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 373             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velecsum         = _mm_add_ps(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm_mul_ps(fscal,dx20);
 382             ty               = _mm_mul_ps(fscal,dy20);
 383             tz               = _mm_mul_ps(fscal,dz20);
 384
 385             /* Update vectorial force */
 386             fix2             = _mm_add_ps(fix2,tx);
 387             fiy2             = _mm_add_ps(fiy2,ty);
 388             fiz2             = _mm_add_ps(fiz2,tz);
 389
 390             fjx0             = _mm_add_ps(fjx0,tx);
 391             fjy0             = _mm_add_ps(fjy0,ty);
 392             fjz0             = _mm_add_ps(fjz0,tz);
 393
 394             fjptrA             = f+j_coord_offsetA;
 395             fjptrB             = f+j_coord_offsetB;
 396             fjptrC             = f+j_coord_offsetC;
 397             fjptrD             = f+j_coord_offsetD;
 398
 399             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 400
 401             /* Inner loop uses 142 flops */
 402         }
 403
 404         if(jidx<j_index_end)
 405         {
 406
 407             /* Get j neighbor index, and coordinate index */
 408             jnrlistA         = jjnr[jidx];
 409             jnrlistB         = jjnr[jidx+1];
 410             jnrlistC         = jjnr[jidx+2];
 411             jnrlistD         = jjnr[jidx+3];
 412             /* Sign of each element will be negative for non-real atoms.
 413              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 414              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 415              */
 416             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 417             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 418             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 419             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 420             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 421             j_coord_offsetA  = DIM*jnrA;
 422             j_coord_offsetB  = DIM*jnrB;
 423             j_coord_offsetC  = DIM*jnrC;
 424             j_coord_offsetD  = DIM*jnrD;
 425
 426             /* load j atom coordinates */
 427             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 428                                               x+j_coord_offsetC,x+j_coord_offsetD,
 429                                               &jx0,&jy0,&jz0);
 430
 431             /* Calculate displacement vector */
 432             dx00             = _mm_sub_ps(ix0,jx0);
 433             dy00             = _mm_sub_ps(iy0,jy0);
 434             dz00             = _mm_sub_ps(iz0,jz0);
 435             dx10             = _mm_sub_ps(ix1,jx0);
 436             dy10             = _mm_sub_ps(iy1,jy0);
 437             dz10             = _mm_sub_ps(iz1,jz0);
 438             dx20             = _mm_sub_ps(ix2,jx0);
 439             dy20             = _mm_sub_ps(iy2,jy0);
 440             dz20             = _mm_sub_ps(iz2,jz0);
 441
 442             /* Calculate squared distance and things based on it */
 443             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 444             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 445             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 446
 447             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 448             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 449             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 450
 451             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 452
 453             /* Load parameters for j particles */
 454             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 455                                                               charge+jnrC+0,charge+jnrD+0);
 456             vdwjidx0A        = 2*vdwtype[jnrA+0];
 457             vdwjidx0B        = 2*vdwtype[jnrB+0];
 458             vdwjidx0C        = 2*vdwtype[jnrC+0];
 459             vdwjidx0D        = 2*vdwtype[jnrD+0];
 460
 461             fjx0             = _mm_setzero_ps();
 462             fjy0             = _mm_setzero_ps();
 463             fjz0             = _mm_setzero_ps();
 464
 465             /**************************
 466              * CALCULATE INTERACTIONS *
 467              **************************/
 468
 469             r00              = _mm_mul_ps(rsq00,rinv00);
 470             r00              = _mm_andnot_ps(dummy_mask,r00);
 471
 472             /* Compute parameters for interactions between i and j atoms */
 473             qq00             = _mm_mul_ps(iq0,jq0);
 474             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 475                                          vdwparam+vdwioffset0+vdwjidx0B,
 476                                          vdwparam+vdwioffset0+vdwjidx0C,
 477                                          vdwparam+vdwioffset0+vdwjidx0D,
 478                                          &c6_00,&c12_00);
 479
 480             /* Calculate table index by multiplying r with table scale and truncate to integer */
 481             rt               = _mm_mul_ps(r00,vftabscale);
 482             vfitab           = _mm_cvttps_epi32(rt);
 483             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 484             vfitab           = _mm_slli_epi32(vfitab,2);
 485
 486             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 487             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 488             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 489             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 490             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 491             _MM_TRANSPOSE4_PS(Y,F,G,H);
 492             Heps             = _mm_mul_ps(vfeps,H);
 493             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 494             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 495             velec            = _mm_mul_ps(qq00,VV);
 496             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 497             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 498
 499             /* LENNARD-JONES DISPERSION/REPULSION */
 500
 501             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 502             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 503             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 504             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 505             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 506
 507             /* Update potential sum for this i atom from the interaction with this j atom. */
 508             velec            = _mm_andnot_ps(dummy_mask,velec);
 509             velecsum         = _mm_add_ps(velecsum,velec);
 510             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 511             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 512
 513             fscal            = _mm_add_ps(felec,fvdw);
 514
 515             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 516
 517             /* Calculate temporary vectorial force */
 518             tx               = _mm_mul_ps(fscal,dx00);
 519             ty               = _mm_mul_ps(fscal,dy00);
 520             tz               = _mm_mul_ps(fscal,dz00);
 521
 522             /* Update vectorial force */
 523             fix0             = _mm_add_ps(fix0,tx);
 524             fiy0             = _mm_add_ps(fiy0,ty);
 525             fiz0             = _mm_add_ps(fiz0,tz);
 526
 527             fjx0             = _mm_add_ps(fjx0,tx);
 528             fjy0             = _mm_add_ps(fjy0,ty);
 529             fjz0             = _mm_add_ps(fjz0,tz);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r10              = _mm_mul_ps(rsq10,rinv10);
 536             r10              = _mm_andnot_ps(dummy_mask,r10);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq10             = _mm_mul_ps(iq1,jq0);
 540
 541             /* Calculate table index by multiplying r with table scale and truncate to integer */
 542             rt               = _mm_mul_ps(r10,vftabscale);
 543             vfitab           = _mm_cvttps_epi32(rt);
 544             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 545             vfitab           = _mm_slli_epi32(vfitab,2);
 546
 547             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 548             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 549             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 550             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 551             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 552             _MM_TRANSPOSE4_PS(Y,F,G,H);
 553             Heps             = _mm_mul_ps(vfeps,H);
 554             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 555             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 556             velec            = _mm_mul_ps(qq10,VV);
 557             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 558             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm_andnot_ps(dummy_mask,velec);
 562             velecsum         = _mm_add_ps(velecsum,velec);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm_mul_ps(fscal,dx10);
 570             ty               = _mm_mul_ps(fscal,dy10);
 571             tz               = _mm_mul_ps(fscal,dz10);
 572
 573             /* Update vectorial force */
 574             fix1             = _mm_add_ps(fix1,tx);
 575             fiy1             = _mm_add_ps(fiy1,ty);
 576             fiz1             = _mm_add_ps(fiz1,tz);
 577
 578             fjx0             = _mm_add_ps(fjx0,tx);
 579             fjy0             = _mm_add_ps(fjy0,ty);
 580             fjz0             = _mm_add_ps(fjz0,tz);
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             r20              = _mm_mul_ps(rsq20,rinv20);
 587             r20              = _mm_andnot_ps(dummy_mask,r20);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq20             = _mm_mul_ps(iq2,jq0);
 591
 592             /* Calculate table index by multiplying r with table scale and truncate to integer */
 593             rt               = _mm_mul_ps(r20,vftabscale);
 594             vfitab           = _mm_cvttps_epi32(rt);
 595             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 596             vfitab           = _mm_slli_epi32(vfitab,2);
 597
 598             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 599             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 600             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 601             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 602             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 603             _MM_TRANSPOSE4_PS(Y,F,G,H);
 604             Heps             = _mm_mul_ps(vfeps,H);
 605             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 606             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 607             velec            = _mm_mul_ps(qq20,VV);
 608             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 609             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 610
 611             /* Update potential sum for this i atom from the interaction with this j atom. */
 612             velec            = _mm_andnot_ps(dummy_mask,velec);
 613             velecsum         = _mm_add_ps(velecsum,velec);
 614
 615             fscal            = felec;
 616
 617             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_ps(fscal,dx20);
 621             ty               = _mm_mul_ps(fscal,dy20);
 622             tz               = _mm_mul_ps(fscal,dz20);
 623
 624             /* Update vectorial force */
 625             fix2             = _mm_add_ps(fix2,tx);
 626             fiy2             = _mm_add_ps(fiy2,ty);
 627             fiz2             = _mm_add_ps(fiz2,tz);
 628
 629             fjx0             = _mm_add_ps(fjx0,tx);
 630             fjy0             = _mm_add_ps(fjy0,ty);
 631             fjz0             = _mm_add_ps(fjz0,tz);
 632
 633             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 634             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 635             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 636             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 637
 638             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 639
 640             /* Inner loop uses 145 flops */
 641         }
 642
 643         /* End of innermost loop */
 644
 645         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 646                                               f+i_coord_offset,fshift+i_shift_offset);
 647
 648         ggid                        = gid[iidx];
 649         /* Update potential energies */
 650         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 651         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 652
 653         /* Increment number of inner iterations */
 654         inneriter                  += j_index_end - j_index_start;
 655
 656         /* Outer loop uses 20 flops */
 657     }
 658
 659     /* Increment number of outer iterations */
 660     outeriter        += nri;
 661
 662     /* Update outer/inner flops */
 663
 664     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 665 }
 666 /*
 667  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
 668  * Electrostatics interaction: CubicSplineTable
 669  * VdW interaction:            LennardJones
 670  * Geometry:                   Water3-Particle
 671  * Calculate force/pot:        Force
 672  */
 673 void
 674 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
 675                     (t_nblist                    * gmx_restrict       nlist,
 676                      rvec                        * gmx_restrict          xx,
 677                      rvec                        * gmx_restrict          ff,
 678                      t_forcerec                  * gmx_restrict          fr,
 679                      t_mdatoms                   * gmx_restrict     mdatoms,
 680                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 681                      t_nrnb                      * gmx_restrict        nrnb)
 682 {
 683     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 684      * just 0 for non-waters.
 685      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 686      * jnr indices corresponding to data put in the four positions in the SIMD register.
 687      */
 688     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 689     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 690     int              jnrA,jnrB,jnrC,jnrD;
 691     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 692     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 693     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 694     real             rcutoff_scalar;
 695     real             *shiftvec,*fshift,*x,*f;
 696     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 697     real             scratch[4*DIM];
 698     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 699     int              vdwioffset0;
 700     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 701     int              vdwioffset1;
 702     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 703     int              vdwioffset2;
 704     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 705     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 706     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 707     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 708     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 709     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 710     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 711     real             *charge;
 712     int              nvdwtype;
 713     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 714     int              *vdwtype;
 715     real             *vdwparam;
 716     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 717     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 718     __m128i          vfitab;
 719     __m128i          ifour       = _mm_set1_epi32(4);
 720     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 721     real             *vftab;
 722     __m128           dummy_mask,cutoff_mask;
 723     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 724     __m128           one     = _mm_set1_ps(1.0);
 725     __m128           two     = _mm_set1_ps(2.0);
 726     x                = xx[0];
 727     f                = ff[0];
 728
 729     nri              = nlist->nri;
 730     iinr             = nlist->iinr;
 731     jindex           = nlist->jindex;
 732     jjnr             = nlist->jjnr;
 733     shiftidx         = nlist->shift;
 734     gid              = nlist->gid;
 735     shiftvec         = fr->shift_vec[0];
 736     fshift           = fr->fshift[0];
 737     facel            = _mm_set1_ps(fr->epsfac);
 738     charge           = mdatoms->chargeA;
 739     nvdwtype         = fr->ntype;
 740     vdwparam         = fr->nbfp;
 741     vdwtype          = mdatoms->typeA;
 742
 743     vftab            = kernel_data->table_elec->data;
 744     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 745
 746     /* Setup water-specific parameters */
 747     inr              = nlist->iinr[0];
 748     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 749     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 750     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 751     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 752
 753     /* Avoid stupid compiler warnings */
 754     jnrA = jnrB = jnrC = jnrD = 0;
 755     j_coord_offsetA = 0;
 756     j_coord_offsetB = 0;
 757     j_coord_offsetC = 0;
 758     j_coord_offsetD = 0;
 759
 760     outeriter        = 0;
 761     inneriter        = 0;
 762
 763     for(iidx=0;iidx<4*DIM;iidx++)
 764     {
 765         scratch[iidx] = 0.0;
 766     }
 767
 768     /* Start outer loop over neighborlists */
 769     for(iidx=0; iidx<nri; iidx++)
 770     {
 771         /* Load shift vector for this list */
 772         i_shift_offset   = DIM*shiftidx[iidx];
 773
 774         /* Load limits for loop over neighbors */
 775         j_index_start    = jindex[iidx];
 776         j_index_end      = jindex[iidx+1];
 777
 778         /* Get outer coordinate index */
 779         inr              = iinr[iidx];
 780         i_coord_offset   = DIM*inr;
 781
 782         /* Load i particle coords and add shift vector */
 783         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 784                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 785
 786         fix0             = _mm_setzero_ps();
 787         fiy0             = _mm_setzero_ps();
 788         fiz0             = _mm_setzero_ps();
 789         fix1             = _mm_setzero_ps();
 790         fiy1             = _mm_setzero_ps();
 791         fiz1             = _mm_setzero_ps();
 792         fix2             = _mm_setzero_ps();
 793         fiy2             = _mm_setzero_ps();
 794         fiz2             = _mm_setzero_ps();
 795
 796         /* Start inner kernel loop */
 797         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 798         {
 799
 800             /* Get j neighbor index, and coordinate index */
 801             jnrA             = jjnr[jidx];
 802             jnrB             = jjnr[jidx+1];
 803             jnrC             = jjnr[jidx+2];
 804             jnrD             = jjnr[jidx+3];
 805             j_coord_offsetA  = DIM*jnrA;
 806             j_coord_offsetB  = DIM*jnrB;
 807             j_coord_offsetC  = DIM*jnrC;
 808             j_coord_offsetD  = DIM*jnrD;
 809
 810             /* load j atom coordinates */
 811             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 812                                               x+j_coord_offsetC,x+j_coord_offsetD,
 813                                               &jx0,&jy0,&jz0);
 814
 815             /* Calculate displacement vector */
 816             dx00             = _mm_sub_ps(ix0,jx0);
 817             dy00             = _mm_sub_ps(iy0,jy0);
 818             dz00             = _mm_sub_ps(iz0,jz0);
 819             dx10             = _mm_sub_ps(ix1,jx0);
 820             dy10             = _mm_sub_ps(iy1,jy0);
 821             dz10             = _mm_sub_ps(iz1,jz0);
 822             dx20             = _mm_sub_ps(ix2,jx0);
 823             dy20             = _mm_sub_ps(iy2,jy0);
 824             dz20             = _mm_sub_ps(iz2,jz0);
 825
 826             /* Calculate squared distance and things based on it */
 827             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 828             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 829             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 830
 831             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 832             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 833             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 834
 835             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 836
 837             /* Load parameters for j particles */
 838             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 839                                                               charge+jnrC+0,charge+jnrD+0);
 840             vdwjidx0A        = 2*vdwtype[jnrA+0];
 841             vdwjidx0B        = 2*vdwtype[jnrB+0];
 842             vdwjidx0C        = 2*vdwtype[jnrC+0];
 843             vdwjidx0D        = 2*vdwtype[jnrD+0];
 844
 845             fjx0             = _mm_setzero_ps();
 846             fjy0             = _mm_setzero_ps();
 847             fjz0             = _mm_setzero_ps();
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             r00              = _mm_mul_ps(rsq00,rinv00);
 854
 855             /* Compute parameters for interactions between i and j atoms */
 856             qq00             = _mm_mul_ps(iq0,jq0);
 857             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 858                                          vdwparam+vdwioffset0+vdwjidx0B,
 859                                          vdwparam+vdwioffset0+vdwjidx0C,
 860                                          vdwparam+vdwioffset0+vdwjidx0D,
 861                                          &c6_00,&c12_00);
 862
 863             /* Calculate table index by multiplying r with table scale and truncate to integer */
 864             rt               = _mm_mul_ps(r00,vftabscale);
 865             vfitab           = _mm_cvttps_epi32(rt);
 866             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 867             vfitab           = _mm_slli_epi32(vfitab,2);
 868
 869             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 870             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 871             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 872             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 873             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 874             _MM_TRANSPOSE4_PS(Y,F,G,H);
 875             Heps             = _mm_mul_ps(vfeps,H);
 876             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 877             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 878             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 879
 880             /* LENNARD-JONES DISPERSION/REPULSION */
 881
 882             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 883             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 884
 885             fscal            = _mm_add_ps(felec,fvdw);
 886
 887             /* Calculate temporary vectorial force */
 888             tx               = _mm_mul_ps(fscal,dx00);
 889             ty               = _mm_mul_ps(fscal,dy00);
 890             tz               = _mm_mul_ps(fscal,dz00);
 891
 892             /* Update vectorial force */
 893             fix0             = _mm_add_ps(fix0,tx);
 894             fiy0             = _mm_add_ps(fiy0,ty);
 895             fiz0             = _mm_add_ps(fiz0,tz);
 896
 897             fjx0             = _mm_add_ps(fjx0,tx);
 898             fjy0             = _mm_add_ps(fjy0,ty);
 899             fjz0             = _mm_add_ps(fjz0,tz);
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             r10              = _mm_mul_ps(rsq10,rinv10);
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq10             = _mm_mul_ps(iq1,jq0);
 909
 910             /* Calculate table index by multiplying r with table scale and truncate to integer */
 911             rt               = _mm_mul_ps(r10,vftabscale);
 912             vfitab           = _mm_cvttps_epi32(rt);
 913             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 914             vfitab           = _mm_slli_epi32(vfitab,2);
 915
 916             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 917             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 918             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 919             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 920             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 921             _MM_TRANSPOSE4_PS(Y,F,G,H);
 922             Heps             = _mm_mul_ps(vfeps,H);
 923             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 924             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 925             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 926
 927             fscal            = felec;
 928
 929             /* Calculate temporary vectorial force */
 930             tx               = _mm_mul_ps(fscal,dx10);
 931             ty               = _mm_mul_ps(fscal,dy10);
 932             tz               = _mm_mul_ps(fscal,dz10);
 933
 934             /* Update vectorial force */
 935             fix1             = _mm_add_ps(fix1,tx);
 936             fiy1             = _mm_add_ps(fiy1,ty);
 937             fiz1             = _mm_add_ps(fiz1,tz);
 938
 939             fjx0             = _mm_add_ps(fjx0,tx);
 940             fjy0             = _mm_add_ps(fjy0,ty);
 941             fjz0             = _mm_add_ps(fjz0,tz);
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             r20              = _mm_mul_ps(rsq20,rinv20);
 948
 949             /* Compute parameters for interactions between i and j atoms */
 950             qq20             = _mm_mul_ps(iq2,jq0);
 951
 952             /* Calculate table index by multiplying r with table scale and truncate to integer */
 953             rt               = _mm_mul_ps(r20,vftabscale);
 954             vfitab           = _mm_cvttps_epi32(rt);
 955             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 956             vfitab           = _mm_slli_epi32(vfitab,2);
 957
 958             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 959             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 960             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 961             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 962             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 963             _MM_TRANSPOSE4_PS(Y,F,G,H);
 964             Heps             = _mm_mul_ps(vfeps,H);
 965             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 966             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 967             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 968
 969             fscal            = felec;
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm_mul_ps(fscal,dx20);
 973             ty               = _mm_mul_ps(fscal,dy20);
 974             tz               = _mm_mul_ps(fscal,dz20);
 975
 976             /* Update vectorial force */
 977             fix2             = _mm_add_ps(fix2,tx);
 978             fiy2             = _mm_add_ps(fiy2,ty);
 979             fiz2             = _mm_add_ps(fiz2,tz);
 980
 981             fjx0             = _mm_add_ps(fjx0,tx);
 982             fjy0             = _mm_add_ps(fjy0,ty);
 983             fjz0             = _mm_add_ps(fjz0,tz);
 984
 985             fjptrA             = f+j_coord_offsetA;
 986             fjptrB             = f+j_coord_offsetB;
 987             fjptrC             = f+j_coord_offsetC;
 988             fjptrD             = f+j_coord_offsetD;
 989
 990             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 991
 992             /* Inner loop uses 125 flops */
 993         }
 994
 995         if(jidx<j_index_end)
 996         {
 997
 998             /* Get j neighbor index, and coordinate index */
 999             jnrlistA         = jjnr[jidx];
1000             jnrlistB         = jjnr[jidx+1];
1001             jnrlistC         = jjnr[jidx+2];
1002             jnrlistD         = jjnr[jidx+3];
1003             /* Sign of each element will be negative for non-real atoms.
1004              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1005              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1006              */
1007             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1008             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1009             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1010             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1011             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1012             j_coord_offsetA  = DIM*jnrA;
1013             j_coord_offsetB  = DIM*jnrB;
1014             j_coord_offsetC  = DIM*jnrC;
1015             j_coord_offsetD  = DIM*jnrD;
1016
1017             /* load j atom coordinates */
1018             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1019                                               x+j_coord_offsetC,x+j_coord_offsetD,
1020                                               &jx0,&jy0,&jz0);
1021
1022             /* Calculate displacement vector */
1023             dx00             = _mm_sub_ps(ix0,jx0);
1024             dy00             = _mm_sub_ps(iy0,jy0);
1025             dz00             = _mm_sub_ps(iz0,jz0);
1026             dx10             = _mm_sub_ps(ix1,jx0);
1027             dy10             = _mm_sub_ps(iy1,jy0);
1028             dz10             = _mm_sub_ps(iz1,jz0);
1029             dx20             = _mm_sub_ps(ix2,jx0);
1030             dy20             = _mm_sub_ps(iy2,jy0);
1031             dz20             = _mm_sub_ps(iz2,jz0);
1032
1033             /* Calculate squared distance and things based on it */
1034             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1035             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1036             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1037
1038             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1039             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1040             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1041
1042             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1043
1044             /* Load parameters for j particles */
1045             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1046                                                               charge+jnrC+0,charge+jnrD+0);
1047             vdwjidx0A        = 2*vdwtype[jnrA+0];
1048             vdwjidx0B        = 2*vdwtype[jnrB+0];
1049             vdwjidx0C        = 2*vdwtype[jnrC+0];
1050             vdwjidx0D        = 2*vdwtype[jnrD+0];
1051
1052             fjx0             = _mm_setzero_ps();
1053             fjy0             = _mm_setzero_ps();
1054             fjz0             = _mm_setzero_ps();
1055
1056             /**************************
1057              * CALCULATE INTERACTIONS *
1058              **************************/
1059
1060             r00              = _mm_mul_ps(rsq00,rinv00);
1061             r00              = _mm_andnot_ps(dummy_mask,r00);
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             qq00             = _mm_mul_ps(iq0,jq0);
1065             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1066                                          vdwparam+vdwioffset0+vdwjidx0B,
1067                                          vdwparam+vdwioffset0+vdwjidx0C,
1068                                          vdwparam+vdwioffset0+vdwjidx0D,
1069                                          &c6_00,&c12_00);
1070
1071             /* Calculate table index by multiplying r with table scale and truncate to integer */
1072             rt               = _mm_mul_ps(r00,vftabscale);
1073             vfitab           = _mm_cvttps_epi32(rt);
1074             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1075             vfitab           = _mm_slli_epi32(vfitab,2);
1076
1077             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1078             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1079             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1080             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1081             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1082             _MM_TRANSPOSE4_PS(Y,F,G,H);
1083             Heps             = _mm_mul_ps(vfeps,H);
1084             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1085             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1086             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1087
1088             /* LENNARD-JONES DISPERSION/REPULSION */
1089
1090             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1091             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1092
1093             fscal            = _mm_add_ps(felec,fvdw);
1094
1095             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1096
1097             /* Calculate temporary vectorial force */
1098             tx               = _mm_mul_ps(fscal,dx00);
1099             ty               = _mm_mul_ps(fscal,dy00);
1100             tz               = _mm_mul_ps(fscal,dz00);
1101
1102             /* Update vectorial force */
1103             fix0             = _mm_add_ps(fix0,tx);
1104             fiy0             = _mm_add_ps(fiy0,ty);
1105             fiz0             = _mm_add_ps(fiz0,tz);
1106
1107             fjx0             = _mm_add_ps(fjx0,tx);
1108             fjy0             = _mm_add_ps(fjy0,ty);
1109             fjz0             = _mm_add_ps(fjz0,tz);
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             r10              = _mm_mul_ps(rsq10,rinv10);
1116             r10              = _mm_andnot_ps(dummy_mask,r10);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq10             = _mm_mul_ps(iq1,jq0);
1120
1121             /* Calculate table index by multiplying r with table scale and truncate to integer */
1122             rt               = _mm_mul_ps(r10,vftabscale);
1123             vfitab           = _mm_cvttps_epi32(rt);
1124             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1125             vfitab           = _mm_slli_epi32(vfitab,2);
1126
1127             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1128             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1129             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1130             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1131             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1132             _MM_TRANSPOSE4_PS(Y,F,G,H);
1133             Heps             = _mm_mul_ps(vfeps,H);
1134             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1135             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1136             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1137
1138             fscal            = felec;
1139
1140             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1141
1142             /* Calculate temporary vectorial force */
1143             tx               = _mm_mul_ps(fscal,dx10);
1144             ty               = _mm_mul_ps(fscal,dy10);
1145             tz               = _mm_mul_ps(fscal,dz10);
1146
1147             /* Update vectorial force */
1148             fix1             = _mm_add_ps(fix1,tx);
1149             fiy1             = _mm_add_ps(fiy1,ty);
1150             fiz1             = _mm_add_ps(fiz1,tz);
1151
1152             fjx0             = _mm_add_ps(fjx0,tx);
1153             fjy0             = _mm_add_ps(fjy0,ty);
1154             fjz0             = _mm_add_ps(fjz0,tz);
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             r20              = _mm_mul_ps(rsq20,rinv20);
1161             r20              = _mm_andnot_ps(dummy_mask,r20);
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             qq20             = _mm_mul_ps(iq2,jq0);
1165
1166             /* Calculate table index by multiplying r with table scale and truncate to integer */
1167             rt               = _mm_mul_ps(r20,vftabscale);
1168             vfitab           = _mm_cvttps_epi32(rt);
1169             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1170             vfitab           = _mm_slli_epi32(vfitab,2);
1171
1172             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1173             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1174             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1175             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1176             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1177             _MM_TRANSPOSE4_PS(Y,F,G,H);
1178             Heps             = _mm_mul_ps(vfeps,H);
1179             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1180             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1181             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1182
1183             fscal            = felec;
1184
1185             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1186
1187             /* Calculate temporary vectorial force */
1188             tx               = _mm_mul_ps(fscal,dx20);
1189             ty               = _mm_mul_ps(fscal,dy20);
1190             tz               = _mm_mul_ps(fscal,dz20);
1191
1192             /* Update vectorial force */
1193             fix2             = _mm_add_ps(fix2,tx);
1194             fiy2             = _mm_add_ps(fiy2,ty);
1195             fiz2             = _mm_add_ps(fiz2,tz);
1196
1197             fjx0             = _mm_add_ps(fjx0,tx);
1198             fjy0             = _mm_add_ps(fjy0,ty);
1199             fjz0             = _mm_add_ps(fjz0,tz);
1200
1201             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1202             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1203             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1204             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1205
1206             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1207
1208             /* Inner loop uses 128 flops */
1209         }
1210
1211         /* End of innermost loop */
1212
1213         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1214                                               f+i_coord_offset,fshift+i_shift_offset);
1215
1216         /* Increment number of inner iterations */
1217         inneriter                  += j_index_end - j_index_start;
1218
1219         /* Outer loop uses 18 flops */
1220     }
1221
1222     /* Increment number of outer iterations */
1223     outeriter        += nri;
1224
1225     /* Update outer/inner flops */
1226
1227     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1228 }