src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse4_1_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_elec->data;
 128     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_ps();
 171         fiy0             = _mm_setzero_ps();
 172         fiz0             = _mm_setzero_ps();
 173         fix1             = _mm_setzero_ps();
 174         fiy1             = _mm_setzero_ps();
 175         fiz1             = _mm_setzero_ps();
 176         fix2             = _mm_setzero_ps();
 177         fiy2             = _mm_setzero_ps();
 178         fiz2             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 216             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 217             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 218
 219             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 220             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 221             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 222
 223             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             fjx0             = _mm_setzero_ps();
 234             fjy0             = _mm_setzero_ps();
 235             fjz0             = _mm_setzero_ps();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             r00              = _mm_mul_ps(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm_mul_ps(iq0,jq0);
 245             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 246                                          vdwparam+vdwioffset0+vdwjidx0B,
 247                                          vdwparam+vdwioffset0+vdwjidx0C,
 248                                          vdwparam+vdwioffset0+vdwjidx0D,
 249                                          &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm_mul_ps(r00,vftabscale);
 253             vfitab           = _mm_cvttps_epi32(rt);
 254             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 255             vfitab           = _mm_slli_epi32(vfitab,2);
 256
 257             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 258             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 259             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 260             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 261             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 262             _MM_TRANSPOSE4_PS(Y,F,G,H);
 263             Heps             = _mm_mul_ps(vfeps,H);
 264             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 265             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 266             velec            = _mm_mul_ps(qq00,VV);
 267             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 268             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 269
 270             /* LENNARD-JONES DISPERSION/REPULSION */
 271
 272             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 273             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 274             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 275             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 276             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velecsum         = _mm_add_ps(velecsum,velec);
 280             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 281
 282             fscal            = _mm_add_ps(felec,fvdw);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_ps(fscal,dx00);
 286             ty               = _mm_mul_ps(fscal,dy00);
 287             tz               = _mm_mul_ps(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_add_ps(fix0,tx);
 291             fiy0             = _mm_add_ps(fiy0,ty);
 292             fiz0             = _mm_add_ps(fiz0,tz);
 293
 294             fjx0             = _mm_add_ps(fjx0,tx);
 295             fjy0             = _mm_add_ps(fjy0,ty);
 296             fjz0             = _mm_add_ps(fjz0,tz);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r10              = _mm_mul_ps(rsq10,rinv10);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_ps(iq1,jq0);
 306
 307             /* Calculate table index by multiplying r with table scale and truncate to integer */
 308             rt               = _mm_mul_ps(r10,vftabscale);
 309             vfitab           = _mm_cvttps_epi32(rt);
 310             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 311             vfitab           = _mm_slli_epi32(vfitab,2);
 312
 313             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 314             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 315             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 316             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 317             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 318             _MM_TRANSPOSE4_PS(Y,F,G,H);
 319             Heps             = _mm_mul_ps(vfeps,H);
 320             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 321             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 322             velec            = _mm_mul_ps(qq10,VV);
 323             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 324             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm_mul_ps(fscal,dx10);
 333             ty               = _mm_mul_ps(fscal,dy10);
 334             tz               = _mm_mul_ps(fscal,dz10);
 335
 336             /* Update vectorial force */
 337             fix1             = _mm_add_ps(fix1,tx);
 338             fiy1             = _mm_add_ps(fiy1,ty);
 339             fiz1             = _mm_add_ps(fiz1,tz);
 340
 341             fjx0             = _mm_add_ps(fjx0,tx);
 342             fjy0             = _mm_add_ps(fjy0,ty);
 343             fjz0             = _mm_add_ps(fjz0,tz);
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             r20              = _mm_mul_ps(rsq20,rinv20);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm_mul_ps(iq2,jq0);
 353
 354             /* Calculate table index by multiplying r with table scale and truncate to integer */
 355             rt               = _mm_mul_ps(r20,vftabscale);
 356             vfitab           = _mm_cvttps_epi32(rt);
 357             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 358             vfitab           = _mm_slli_epi32(vfitab,2);
 359
 360             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 361             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 362             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 363             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 364             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 365             _MM_TRANSPOSE4_PS(Y,F,G,H);
 366             Heps             = _mm_mul_ps(vfeps,H);
 367             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 368             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 369             velec            = _mm_mul_ps(qq20,VV);
 370             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 371             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velecsum         = _mm_add_ps(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm_mul_ps(fscal,dx20);
 380             ty               = _mm_mul_ps(fscal,dy20);
 381             tz               = _mm_mul_ps(fscal,dz20);
 382
 383             /* Update vectorial force */
 384             fix2             = _mm_add_ps(fix2,tx);
 385             fiy2             = _mm_add_ps(fiy2,ty);
 386             fiz2             = _mm_add_ps(fiz2,tz);
 387
 388             fjx0             = _mm_add_ps(fjx0,tx);
 389             fjy0             = _mm_add_ps(fjy0,ty);
 390             fjz0             = _mm_add_ps(fjz0,tz);
 391
 392             fjptrA             = f+j_coord_offsetA;
 393             fjptrB             = f+j_coord_offsetB;
 394             fjptrC             = f+j_coord_offsetC;
 395             fjptrD             = f+j_coord_offsetD;
 396
 397             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 398
 399             /* Inner loop uses 142 flops */
 400         }
 401
 402         if(jidx<j_index_end)
 403         {
 404
 405             /* Get j neighbor index, and coordinate index */
 406             jnrlistA         = jjnr[jidx];
 407             jnrlistB         = jjnr[jidx+1];
 408             jnrlistC         = jjnr[jidx+2];
 409             jnrlistD         = jjnr[jidx+3];
 410             /* Sign of each element will be negative for non-real atoms.
 411              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 412              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 413              */
 414             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 415             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 416             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 417             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 418             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 419             j_coord_offsetA  = DIM*jnrA;
 420             j_coord_offsetB  = DIM*jnrB;
 421             j_coord_offsetC  = DIM*jnrC;
 422             j_coord_offsetD  = DIM*jnrD;
 423
 424             /* load j atom coordinates */
 425             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 426                                               x+j_coord_offsetC,x+j_coord_offsetD,
 427                                               &jx0,&jy0,&jz0);
 428
 429             /* Calculate displacement vector */
 430             dx00             = _mm_sub_ps(ix0,jx0);
 431             dy00             = _mm_sub_ps(iy0,jy0);
 432             dz00             = _mm_sub_ps(iz0,jz0);
 433             dx10             = _mm_sub_ps(ix1,jx0);
 434             dy10             = _mm_sub_ps(iy1,jy0);
 435             dz10             = _mm_sub_ps(iz1,jz0);
 436             dx20             = _mm_sub_ps(ix2,jx0);
 437             dy20             = _mm_sub_ps(iy2,jy0);
 438             dz20             = _mm_sub_ps(iz2,jz0);
 439
 440             /* Calculate squared distance and things based on it */
 441             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 442             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 443             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 444
 445             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 446             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 447             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 448
 449             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 450
 451             /* Load parameters for j particles */
 452             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 453                                                               charge+jnrC+0,charge+jnrD+0);
 454             vdwjidx0A        = 2*vdwtype[jnrA+0];
 455             vdwjidx0B        = 2*vdwtype[jnrB+0];
 456             vdwjidx0C        = 2*vdwtype[jnrC+0];
 457             vdwjidx0D        = 2*vdwtype[jnrD+0];
 458
 459             fjx0             = _mm_setzero_ps();
 460             fjy0             = _mm_setzero_ps();
 461             fjz0             = _mm_setzero_ps();
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             r00              = _mm_mul_ps(rsq00,rinv00);
 468             r00              = _mm_andnot_ps(dummy_mask,r00);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq00             = _mm_mul_ps(iq0,jq0);
 472             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 473                                          vdwparam+vdwioffset0+vdwjidx0B,
 474                                          vdwparam+vdwioffset0+vdwjidx0C,
 475                                          vdwparam+vdwioffset0+vdwjidx0D,
 476                                          &c6_00,&c12_00);
 477
 478             /* Calculate table index by multiplying r with table scale and truncate to integer */
 479             rt               = _mm_mul_ps(r00,vftabscale);
 480             vfitab           = _mm_cvttps_epi32(rt);
 481             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 482             vfitab           = _mm_slli_epi32(vfitab,2);
 483
 484             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 485             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 486             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 487             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 488             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 489             _MM_TRANSPOSE4_PS(Y,F,G,H);
 490             Heps             = _mm_mul_ps(vfeps,H);
 491             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 492             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 493             velec            = _mm_mul_ps(qq00,VV);
 494             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 495             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 496
 497             /* LENNARD-JONES DISPERSION/REPULSION */
 498
 499             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 500             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 501             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 502             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 503             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm_andnot_ps(dummy_mask,velec);
 507             velecsum         = _mm_add_ps(velecsum,velec);
 508             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 509             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 510
 511             fscal            = _mm_add_ps(felec,fvdw);
 512
 513             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 514
 515             /* Calculate temporary vectorial force */
 516             tx               = _mm_mul_ps(fscal,dx00);
 517             ty               = _mm_mul_ps(fscal,dy00);
 518             tz               = _mm_mul_ps(fscal,dz00);
 519
 520             /* Update vectorial force */
 521             fix0             = _mm_add_ps(fix0,tx);
 522             fiy0             = _mm_add_ps(fiy0,ty);
 523             fiz0             = _mm_add_ps(fiz0,tz);
 524
 525             fjx0             = _mm_add_ps(fjx0,tx);
 526             fjy0             = _mm_add_ps(fjy0,ty);
 527             fjz0             = _mm_add_ps(fjz0,tz);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r10              = _mm_mul_ps(rsq10,rinv10);
 534             r10              = _mm_andnot_ps(dummy_mask,r10);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq10             = _mm_mul_ps(iq1,jq0);
 538
 539             /* Calculate table index by multiplying r with table scale and truncate to integer */
 540             rt               = _mm_mul_ps(r10,vftabscale);
 541             vfitab           = _mm_cvttps_epi32(rt);
 542             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 543             vfitab           = _mm_slli_epi32(vfitab,2);
 544
 545             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 546             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 547             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 548             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 549             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 550             _MM_TRANSPOSE4_PS(Y,F,G,H);
 551             Heps             = _mm_mul_ps(vfeps,H);
 552             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 553             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 554             velec            = _mm_mul_ps(qq10,VV);
 555             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 556             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             velec            = _mm_andnot_ps(dummy_mask,velec);
 560             velecsum         = _mm_add_ps(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_ps(fscal,dx10);
 568             ty               = _mm_mul_ps(fscal,dy10);
 569             tz               = _mm_mul_ps(fscal,dz10);
 570
 571             /* Update vectorial force */
 572             fix1             = _mm_add_ps(fix1,tx);
 573             fiy1             = _mm_add_ps(fiy1,ty);
 574             fiz1             = _mm_add_ps(fiz1,tz);
 575
 576             fjx0             = _mm_add_ps(fjx0,tx);
 577             fjy0             = _mm_add_ps(fjy0,ty);
 578             fjz0             = _mm_add_ps(fjz0,tz);
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             r20              = _mm_mul_ps(rsq20,rinv20);
 585             r20              = _mm_andnot_ps(dummy_mask,r20);
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             qq20             = _mm_mul_ps(iq2,jq0);
 589
 590             /* Calculate table index by multiplying r with table scale and truncate to integer */
 591             rt               = _mm_mul_ps(r20,vftabscale);
 592             vfitab           = _mm_cvttps_epi32(rt);
 593             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 594             vfitab           = _mm_slli_epi32(vfitab,2);
 595
 596             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 597             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 598             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 599             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 600             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 601             _MM_TRANSPOSE4_PS(Y,F,G,H);
 602             Heps             = _mm_mul_ps(vfeps,H);
 603             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 604             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 605             velec            = _mm_mul_ps(qq20,VV);
 606             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 607             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 608
 609             /* Update potential sum for this i atom from the interaction with this j atom. */
 610             velec            = _mm_andnot_ps(dummy_mask,velec);
 611             velecsum         = _mm_add_ps(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm_mul_ps(fscal,dx20);
 619             ty               = _mm_mul_ps(fscal,dy20);
 620             tz               = _mm_mul_ps(fscal,dz20);
 621
 622             /* Update vectorial force */
 623             fix2             = _mm_add_ps(fix2,tx);
 624             fiy2             = _mm_add_ps(fiy2,ty);
 625             fiz2             = _mm_add_ps(fiz2,tz);
 626
 627             fjx0             = _mm_add_ps(fjx0,tx);
 628             fjy0             = _mm_add_ps(fjy0,ty);
 629             fjz0             = _mm_add_ps(fjz0,tz);
 630
 631             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 632             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 633             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 634             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 635
 636             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 637
 638             /* Inner loop uses 145 flops */
 639         }
 640
 641         /* End of innermost loop */
 642
 643         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 644                                               f+i_coord_offset,fshift+i_shift_offset);
 645
 646         ggid                        = gid[iidx];
 647         /* Update potential energies */
 648         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 649         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 650
 651         /* Increment number of inner iterations */
 652         inneriter                  += j_index_end - j_index_start;
 653
 654         /* Outer loop uses 20 flops */
 655     }
 656
 657     /* Increment number of outer iterations */
 658     outeriter        += nri;
 659
 660     /* Update outer/inner flops */
 661
 662     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 663 }
 664 /*
 665  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse4_1_single
 666  * Electrostatics interaction: CubicSplineTable
 667  * VdW interaction:            LennardJones
 668  * Geometry:                   Water3-Particle
 669  * Calculate force/pot:        Force
 670  */
 671 void
 672 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse4_1_single
 673                     (t_nblist                    * gmx_restrict       nlist,
 674                      rvec                        * gmx_restrict          xx,
 675                      rvec                        * gmx_restrict          ff,
 676                      t_forcerec                  * gmx_restrict          fr,
 677                      t_mdatoms                   * gmx_restrict     mdatoms,
 678                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 679                      t_nrnb                      * gmx_restrict        nrnb)
 680 {
 681     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 682      * just 0 for non-waters.
 683      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 684      * jnr indices corresponding to data put in the four positions in the SIMD register.
 685      */
 686     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 687     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 688     int              jnrA,jnrB,jnrC,jnrD;
 689     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 690     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 691     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 692     real             rcutoff_scalar;
 693     real             *shiftvec,*fshift,*x,*f;
 694     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 695     real             scratch[4*DIM];
 696     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 697     int              vdwioffset0;
 698     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 699     int              vdwioffset1;
 700     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 701     int              vdwioffset2;
 702     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 703     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 704     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 705     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 706     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 707     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 708     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 709     real             *charge;
 710     int              nvdwtype;
 711     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 712     int              *vdwtype;
 713     real             *vdwparam;
 714     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 715     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 716     __m128i          vfitab;
 717     __m128i          ifour       = _mm_set1_epi32(4);
 718     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 719     real             *vftab;
 720     __m128           dummy_mask,cutoff_mask;
 721     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 722     __m128           one     = _mm_set1_ps(1.0);
 723     __m128           two     = _mm_set1_ps(2.0);
 724     x                = xx[0];
 725     f                = ff[0];
 726
 727     nri              = nlist->nri;
 728     iinr             = nlist->iinr;
 729     jindex           = nlist->jindex;
 730     jjnr             = nlist->jjnr;
 731     shiftidx         = nlist->shift;
 732     gid              = nlist->gid;
 733     shiftvec         = fr->shift_vec[0];
 734     fshift           = fr->fshift[0];
 735     facel            = _mm_set1_ps(fr->epsfac);
 736     charge           = mdatoms->chargeA;
 737     nvdwtype         = fr->ntype;
 738     vdwparam         = fr->nbfp;
 739     vdwtype          = mdatoms->typeA;
 740
 741     vftab            = kernel_data->table_elec->data;
 742     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 743
 744     /* Setup water-specific parameters */
 745     inr              = nlist->iinr[0];
 746     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 747     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 748     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 749     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 750
 751     /* Avoid stupid compiler warnings */
 752     jnrA = jnrB = jnrC = jnrD = 0;
 753     j_coord_offsetA = 0;
 754     j_coord_offsetB = 0;
 755     j_coord_offsetC = 0;
 756     j_coord_offsetD = 0;
 757
 758     outeriter        = 0;
 759     inneriter        = 0;
 760
 761     for(iidx=0;iidx<4*DIM;iidx++)
 762     {
 763         scratch[iidx] = 0.0;
 764     }
 765
 766     /* Start outer loop over neighborlists */
 767     for(iidx=0; iidx<nri; iidx++)
 768     {
 769         /* Load shift vector for this list */
 770         i_shift_offset   = DIM*shiftidx[iidx];
 771
 772         /* Load limits for loop over neighbors */
 773         j_index_start    = jindex[iidx];
 774         j_index_end      = jindex[iidx+1];
 775
 776         /* Get outer coordinate index */
 777         inr              = iinr[iidx];
 778         i_coord_offset   = DIM*inr;
 779
 780         /* Load i particle coords and add shift vector */
 781         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 782                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 783
 784         fix0             = _mm_setzero_ps();
 785         fiy0             = _mm_setzero_ps();
 786         fiz0             = _mm_setzero_ps();
 787         fix1             = _mm_setzero_ps();
 788         fiy1             = _mm_setzero_ps();
 789         fiz1             = _mm_setzero_ps();
 790         fix2             = _mm_setzero_ps();
 791         fiy2             = _mm_setzero_ps();
 792         fiz2             = _mm_setzero_ps();
 793
 794         /* Start inner kernel loop */
 795         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 796         {
 797
 798             /* Get j neighbor index, and coordinate index */
 799             jnrA             = jjnr[jidx];
 800             jnrB             = jjnr[jidx+1];
 801             jnrC             = jjnr[jidx+2];
 802             jnrD             = jjnr[jidx+3];
 803             j_coord_offsetA  = DIM*jnrA;
 804             j_coord_offsetB  = DIM*jnrB;
 805             j_coord_offsetC  = DIM*jnrC;
 806             j_coord_offsetD  = DIM*jnrD;
 807
 808             /* load j atom coordinates */
 809             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 810                                               x+j_coord_offsetC,x+j_coord_offsetD,
 811                                               &jx0,&jy0,&jz0);
 812
 813             /* Calculate displacement vector */
 814             dx00             = _mm_sub_ps(ix0,jx0);
 815             dy00             = _mm_sub_ps(iy0,jy0);
 816             dz00             = _mm_sub_ps(iz0,jz0);
 817             dx10             = _mm_sub_ps(ix1,jx0);
 818             dy10             = _mm_sub_ps(iy1,jy0);
 819             dz10             = _mm_sub_ps(iz1,jz0);
 820             dx20             = _mm_sub_ps(ix2,jx0);
 821             dy20             = _mm_sub_ps(iy2,jy0);
 822             dz20             = _mm_sub_ps(iz2,jz0);
 823
 824             /* Calculate squared distance and things based on it */
 825             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 826             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 827             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 828
 829             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 830             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 831             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 832
 833             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 834
 835             /* Load parameters for j particles */
 836             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 837                                                               charge+jnrC+0,charge+jnrD+0);
 838             vdwjidx0A        = 2*vdwtype[jnrA+0];
 839             vdwjidx0B        = 2*vdwtype[jnrB+0];
 840             vdwjidx0C        = 2*vdwtype[jnrC+0];
 841             vdwjidx0D        = 2*vdwtype[jnrD+0];
 842
 843             fjx0             = _mm_setzero_ps();
 844             fjy0             = _mm_setzero_ps();
 845             fjz0             = _mm_setzero_ps();
 846
 847             /**************************
 848              * CALCULATE INTERACTIONS *
 849              **************************/
 850
 851             r00              = _mm_mul_ps(rsq00,rinv00);
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             qq00             = _mm_mul_ps(iq0,jq0);
 855             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 856                                          vdwparam+vdwioffset0+vdwjidx0B,
 857                                          vdwparam+vdwioffset0+vdwjidx0C,
 858                                          vdwparam+vdwioffset0+vdwjidx0D,
 859                                          &c6_00,&c12_00);
 860
 861             /* Calculate table index by multiplying r with table scale and truncate to integer */
 862             rt               = _mm_mul_ps(r00,vftabscale);
 863             vfitab           = _mm_cvttps_epi32(rt);
 864             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 865             vfitab           = _mm_slli_epi32(vfitab,2);
 866
 867             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 868             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 869             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 870             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 871             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 872             _MM_TRANSPOSE4_PS(Y,F,G,H);
 873             Heps             = _mm_mul_ps(vfeps,H);
 874             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 875             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 876             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 877
 878             /* LENNARD-JONES DISPERSION/REPULSION */
 879
 880             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 881             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 882
 883             fscal            = _mm_add_ps(felec,fvdw);
 884
 885             /* Calculate temporary vectorial force */
 886             tx               = _mm_mul_ps(fscal,dx00);
 887             ty               = _mm_mul_ps(fscal,dy00);
 888             tz               = _mm_mul_ps(fscal,dz00);
 889
 890             /* Update vectorial force */
 891             fix0             = _mm_add_ps(fix0,tx);
 892             fiy0             = _mm_add_ps(fiy0,ty);
 893             fiz0             = _mm_add_ps(fiz0,tz);
 894
 895             fjx0             = _mm_add_ps(fjx0,tx);
 896             fjy0             = _mm_add_ps(fjy0,ty);
 897             fjz0             = _mm_add_ps(fjz0,tz);
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             r10              = _mm_mul_ps(rsq10,rinv10);
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             qq10             = _mm_mul_ps(iq1,jq0);
 907
 908             /* Calculate table index by multiplying r with table scale and truncate to integer */
 909             rt               = _mm_mul_ps(r10,vftabscale);
 910             vfitab           = _mm_cvttps_epi32(rt);
 911             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 912             vfitab           = _mm_slli_epi32(vfitab,2);
 913
 914             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 915             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 916             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 917             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 918             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 919             _MM_TRANSPOSE4_PS(Y,F,G,H);
 920             Heps             = _mm_mul_ps(vfeps,H);
 921             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 922             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 923             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 924
 925             fscal            = felec;
 926
 927             /* Calculate temporary vectorial force */
 928             tx               = _mm_mul_ps(fscal,dx10);
 929             ty               = _mm_mul_ps(fscal,dy10);
 930             tz               = _mm_mul_ps(fscal,dz10);
 931
 932             /* Update vectorial force */
 933             fix1             = _mm_add_ps(fix1,tx);
 934             fiy1             = _mm_add_ps(fiy1,ty);
 935             fiz1             = _mm_add_ps(fiz1,tz);
 936
 937             fjx0             = _mm_add_ps(fjx0,tx);
 938             fjy0             = _mm_add_ps(fjy0,ty);
 939             fjz0             = _mm_add_ps(fjz0,tz);
 940
 941             /**************************
 942              * CALCULATE INTERACTIONS *
 943              **************************/
 944
 945             r20              = _mm_mul_ps(rsq20,rinv20);
 946
 947             /* Compute parameters for interactions between i and j atoms */
 948             qq20             = _mm_mul_ps(iq2,jq0);
 949
 950             /* Calculate table index by multiplying r with table scale and truncate to integer */
 951             rt               = _mm_mul_ps(r20,vftabscale);
 952             vfitab           = _mm_cvttps_epi32(rt);
 953             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 954             vfitab           = _mm_slli_epi32(vfitab,2);
 955
 956             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 957             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 958             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 959             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 960             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 961             _MM_TRANSPOSE4_PS(Y,F,G,H);
 962             Heps             = _mm_mul_ps(vfeps,H);
 963             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 964             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 965             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 966
 967             fscal            = felec;
 968
 969             /* Calculate temporary vectorial force */
 970             tx               = _mm_mul_ps(fscal,dx20);
 971             ty               = _mm_mul_ps(fscal,dy20);
 972             tz               = _mm_mul_ps(fscal,dz20);
 973
 974             /* Update vectorial force */
 975             fix2             = _mm_add_ps(fix2,tx);
 976             fiy2             = _mm_add_ps(fiy2,ty);
 977             fiz2             = _mm_add_ps(fiz2,tz);
 978
 979             fjx0             = _mm_add_ps(fjx0,tx);
 980             fjy0             = _mm_add_ps(fjy0,ty);
 981             fjz0             = _mm_add_ps(fjz0,tz);
 982
 983             fjptrA             = f+j_coord_offsetA;
 984             fjptrB             = f+j_coord_offsetB;
 985             fjptrC             = f+j_coord_offsetC;
 986             fjptrD             = f+j_coord_offsetD;
 987
 988             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 989
 990             /* Inner loop uses 125 flops */
 991         }
 992
 993         if(jidx<j_index_end)
 994         {
 995
 996             /* Get j neighbor index, and coordinate index */
 997             jnrlistA         = jjnr[jidx];
 998             jnrlistB         = jjnr[jidx+1];
 999             jnrlistC         = jjnr[jidx+2];
1000             jnrlistD         = jjnr[jidx+3];
1001             /* Sign of each element will be negative for non-real atoms.
1002              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1003              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1004              */
1005             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1006             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1007             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1008             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1009             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1010             j_coord_offsetA  = DIM*jnrA;
1011             j_coord_offsetB  = DIM*jnrB;
1012             j_coord_offsetC  = DIM*jnrC;
1013             j_coord_offsetD  = DIM*jnrD;
1014
1015             /* load j atom coordinates */
1016             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1017                                               x+j_coord_offsetC,x+j_coord_offsetD,
1018                                               &jx0,&jy0,&jz0);
1019
1020             /* Calculate displacement vector */
1021             dx00             = _mm_sub_ps(ix0,jx0);
1022             dy00             = _mm_sub_ps(iy0,jy0);
1023             dz00             = _mm_sub_ps(iz0,jz0);
1024             dx10             = _mm_sub_ps(ix1,jx0);
1025             dy10             = _mm_sub_ps(iy1,jy0);
1026             dz10             = _mm_sub_ps(iz1,jz0);
1027             dx20             = _mm_sub_ps(ix2,jx0);
1028             dy20             = _mm_sub_ps(iy2,jy0);
1029             dz20             = _mm_sub_ps(iz2,jz0);
1030
1031             /* Calculate squared distance and things based on it */
1032             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1033             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1034             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1035
1036             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1037             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1038             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1039
1040             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1041
1042             /* Load parameters for j particles */
1043             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1044                                                               charge+jnrC+0,charge+jnrD+0);
1045             vdwjidx0A        = 2*vdwtype[jnrA+0];
1046             vdwjidx0B        = 2*vdwtype[jnrB+0];
1047             vdwjidx0C        = 2*vdwtype[jnrC+0];
1048             vdwjidx0D        = 2*vdwtype[jnrD+0];
1049
1050             fjx0             = _mm_setzero_ps();
1051             fjy0             = _mm_setzero_ps();
1052             fjz0             = _mm_setzero_ps();
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             r00              = _mm_mul_ps(rsq00,rinv00);
1059             r00              = _mm_andnot_ps(dummy_mask,r00);
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq00             = _mm_mul_ps(iq0,jq0);
1063             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1064                                          vdwparam+vdwioffset0+vdwjidx0B,
1065                                          vdwparam+vdwioffset0+vdwjidx0C,
1066                                          vdwparam+vdwioffset0+vdwjidx0D,
1067                                          &c6_00,&c12_00);
1068
1069             /* Calculate table index by multiplying r with table scale and truncate to integer */
1070             rt               = _mm_mul_ps(r00,vftabscale);
1071             vfitab           = _mm_cvttps_epi32(rt);
1072             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1073             vfitab           = _mm_slli_epi32(vfitab,2);
1074
1075             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1076             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1077             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1078             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1079             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1080             _MM_TRANSPOSE4_PS(Y,F,G,H);
1081             Heps             = _mm_mul_ps(vfeps,H);
1082             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1083             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1084             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1085
1086             /* LENNARD-JONES DISPERSION/REPULSION */
1087
1088             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1089             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1090
1091             fscal            = _mm_add_ps(felec,fvdw);
1092
1093             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm_mul_ps(fscal,dx00);
1097             ty               = _mm_mul_ps(fscal,dy00);
1098             tz               = _mm_mul_ps(fscal,dz00);
1099
1100             /* Update vectorial force */
1101             fix0             = _mm_add_ps(fix0,tx);
1102             fiy0             = _mm_add_ps(fiy0,ty);
1103             fiz0             = _mm_add_ps(fiz0,tz);
1104
1105             fjx0             = _mm_add_ps(fjx0,tx);
1106             fjy0             = _mm_add_ps(fjy0,ty);
1107             fjz0             = _mm_add_ps(fjz0,tz);
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             r10              = _mm_mul_ps(rsq10,rinv10);
1114             r10              = _mm_andnot_ps(dummy_mask,r10);
1115
1116             /* Compute parameters for interactions between i and j atoms */
1117             qq10             = _mm_mul_ps(iq1,jq0);
1118
1119             /* Calculate table index by multiplying r with table scale and truncate to integer */
1120             rt               = _mm_mul_ps(r10,vftabscale);
1121             vfitab           = _mm_cvttps_epi32(rt);
1122             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1123             vfitab           = _mm_slli_epi32(vfitab,2);
1124
1125             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1126             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1127             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1128             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1129             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1130             _MM_TRANSPOSE4_PS(Y,F,G,H);
1131             Heps             = _mm_mul_ps(vfeps,H);
1132             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1133             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1134             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1135
1136             fscal            = felec;
1137
1138             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1139
1140             /* Calculate temporary vectorial force */
1141             tx               = _mm_mul_ps(fscal,dx10);
1142             ty               = _mm_mul_ps(fscal,dy10);
1143             tz               = _mm_mul_ps(fscal,dz10);
1144
1145             /* Update vectorial force */
1146             fix1             = _mm_add_ps(fix1,tx);
1147             fiy1             = _mm_add_ps(fiy1,ty);
1148             fiz1             = _mm_add_ps(fiz1,tz);
1149
1150             fjx0             = _mm_add_ps(fjx0,tx);
1151             fjy0             = _mm_add_ps(fjy0,ty);
1152             fjz0             = _mm_add_ps(fjz0,tz);
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             r20              = _mm_mul_ps(rsq20,rinv20);
1159             r20              = _mm_andnot_ps(dummy_mask,r20);
1160
1161             /* Compute parameters for interactions between i and j atoms */
1162             qq20             = _mm_mul_ps(iq2,jq0);
1163
1164             /* Calculate table index by multiplying r with table scale and truncate to integer */
1165             rt               = _mm_mul_ps(r20,vftabscale);
1166             vfitab           = _mm_cvttps_epi32(rt);
1167             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1168             vfitab           = _mm_slli_epi32(vfitab,2);
1169
1170             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1171             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1172             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1173             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1174             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1175             _MM_TRANSPOSE4_PS(Y,F,G,H);
1176             Heps             = _mm_mul_ps(vfeps,H);
1177             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1178             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1179             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1180
1181             fscal            = felec;
1182
1183             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1184
1185             /* Calculate temporary vectorial force */
1186             tx               = _mm_mul_ps(fscal,dx20);
1187             ty               = _mm_mul_ps(fscal,dy20);
1188             tz               = _mm_mul_ps(fscal,dz20);
1189
1190             /* Update vectorial force */
1191             fix2             = _mm_add_ps(fix2,tx);
1192             fiy2             = _mm_add_ps(fiy2,ty);
1193             fiz2             = _mm_add_ps(fiz2,tz);
1194
1195             fjx0             = _mm_add_ps(fjx0,tx);
1196             fjy0             = _mm_add_ps(fjy0,ty);
1197             fjz0             = _mm_add_ps(fjz0,tz);
1198
1199             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1200             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1201             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1202             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1203
1204             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1205
1206             /* Inner loop uses 128 flops */
1207         }
1208
1209         /* End of innermost loop */
1210
1211         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1212                                               f+i_coord_offset,fshift+i_shift_offset);
1213
1214         /* Increment number of inner iterations */
1215         inneriter                  += j_index_end - j_index_start;
1216
1217         /* Outer loop uses 18 flops */
1218     }
1219
1220     /* Increment number of outer iterations */
1221     outeriter        += nri;
1222
1223     /* Update outer/inner flops */
1224
1225     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1226 }