src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwCSTab_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse2_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse2_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 124     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_ps(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 136     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 137     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176         fix1             = _mm_setzero_ps();
 177         fiy1             = _mm_setzero_ps();
 178         fiz1             = _mm_setzero_ps();
 179         fix2             = _mm_setzero_ps();
 180         fiy2             = _mm_setzero_ps();
 181         fiz2             = _mm_setzero_ps();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm_setzero_ps();
 185         vvdwsum          = _mm_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198             j_coord_offsetC  = DIM*jnrC;
 199             j_coord_offsetD  = DIM*jnrD;
 200
 201             /* load j atom coordinates */
 202             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                               x+j_coord_offsetC,x+j_coord_offsetD,
 204                                               &jx0,&jy0,&jz0);
 205
 206             /* Calculate displacement vector */
 207             dx00             = _mm_sub_ps(ix0,jx0);
 208             dy00             = _mm_sub_ps(iy0,jy0);
 209             dz00             = _mm_sub_ps(iz0,jz0);
 210             dx10             = _mm_sub_ps(ix1,jx0);
 211             dy10             = _mm_sub_ps(iy1,jy0);
 212             dz10             = _mm_sub_ps(iz1,jz0);
 213             dx20             = _mm_sub_ps(ix2,jx0);
 214             dy20             = _mm_sub_ps(iy2,jy0);
 215             dz20             = _mm_sub_ps(iz2,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 221
 222             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 223             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 224             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 225
 226             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 227             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 232                                                               charge+jnrC+0,charge+jnrD+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235             vdwjidx0C        = 2*vdwtype[jnrC+0];
 236             vdwjidx0D        = 2*vdwtype[jnrD+0];
 237
 238             fjx0             = _mm_setzero_ps();
 239             fjy0             = _mm_setzero_ps();
 240             fjz0             = _mm_setzero_ps();
 241
 242             /**************************
 243              * CALCULATE INTERACTIONS *
 244              **************************/
 245
 246             r00              = _mm_mul_ps(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             qq00             = _mm_mul_ps(iq0,jq0);
 250             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 251                                          vdwparam+vdwioffset0+vdwjidx0B,
 252                                          vdwparam+vdwioffset0+vdwjidx0C,
 253                                          vdwparam+vdwioffset0+vdwjidx0D,
 254                                          &c6_00,&c12_00);
 255
 256             /* Calculate table index by multiplying r with table scale and truncate to integer */
 257             rt               = _mm_mul_ps(r00,vftabscale);
 258             vfitab           = _mm_cvttps_epi32(rt);
 259             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 260             vfitab           = _mm_slli_epi32(vfitab,3);
 261
 262             /* REACTION-FIELD ELECTROSTATICS */
 263             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 264             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 265
 266             /* CUBIC SPLINE TABLE DISPERSION */
 267             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 269             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 270             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 271             _MM_TRANSPOSE4_PS(Y,F,G,H);
 272             Heps             = _mm_mul_ps(vfeps,H);
 273             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 274             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 275             vvdw6            = _mm_mul_ps(c6_00,VV);
 276             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 277             fvdw6            = _mm_mul_ps(c6_00,FF);
 278
 279             /* CUBIC SPLINE TABLE REPULSION */
 280             vfitab           = _mm_add_epi32(vfitab,ifour);
 281             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 282             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 283             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 284             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 285             _MM_TRANSPOSE4_PS(Y,F,G,H);
 286             Heps             = _mm_mul_ps(vfeps,H);
 287             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 288             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 289             vvdw12           = _mm_mul_ps(c12_00,VV);
 290             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 291             fvdw12           = _mm_mul_ps(c12_00,FF);
 292             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 293             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_ps(velecsum,velec);
 297             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 298
 299             fscal            = _mm_add_ps(felec,fvdw);
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm_mul_ps(fscal,dx00);
 303             ty               = _mm_mul_ps(fscal,dy00);
 304             tz               = _mm_mul_ps(fscal,dz00);
 305
 306             /* Update vectorial force */
 307             fix0             = _mm_add_ps(fix0,tx);
 308             fiy0             = _mm_add_ps(fiy0,ty);
 309             fiz0             = _mm_add_ps(fiz0,tz);
 310
 311             fjx0             = _mm_add_ps(fjx0,tx);
 312             fjy0             = _mm_add_ps(fjy0,ty);
 313             fjz0             = _mm_add_ps(fjz0,tz);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_ps(iq1,jq0);
 321
 322             /* REACTION-FIELD ELECTROSTATICS */
 323             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 324             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 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             /* Compute parameters for interactions between i and j atoms */
 350             qq20             = _mm_mul_ps(iq2,jq0);
 351
 352             /* REACTION-FIELD ELECTROSTATICS */
 353             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 354             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm_add_ps(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm_mul_ps(fscal,dx20);
 363             ty               = _mm_mul_ps(fscal,dy20);
 364             tz               = _mm_mul_ps(fscal,dz20);
 365
 366             /* Update vectorial force */
 367             fix2             = _mm_add_ps(fix2,tx);
 368             fiy2             = _mm_add_ps(fiy2,ty);
 369             fiz2             = _mm_add_ps(fiz2,tz);
 370
 371             fjx0             = _mm_add_ps(fjx0,tx);
 372             fjy0             = _mm_add_ps(fjy0,ty);
 373             fjz0             = _mm_add_ps(fjz0,tz);
 374
 375             fjptrA             = f+j_coord_offsetA;
 376             fjptrB             = f+j_coord_offsetB;
 377             fjptrC             = f+j_coord_offsetC;
 378             fjptrD             = f+j_coord_offsetD;
 379
 380             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 381
 382             /* Inner loop uses 131 flops */
 383         }
 384
 385         if(jidx<j_index_end)
 386         {
 387
 388             /* Get j neighbor index, and coordinate index */
 389             jnrlistA         = jjnr[jidx];
 390             jnrlistB         = jjnr[jidx+1];
 391             jnrlistC         = jjnr[jidx+2];
 392             jnrlistD         = jjnr[jidx+3];
 393             /* Sign of each element will be negative for non-real atoms.
 394              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 395              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 396              */
 397             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 398             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 399             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 400             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 401             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 402             j_coord_offsetA  = DIM*jnrA;
 403             j_coord_offsetB  = DIM*jnrB;
 404             j_coord_offsetC  = DIM*jnrC;
 405             j_coord_offsetD  = DIM*jnrD;
 406
 407             /* load j atom coordinates */
 408             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 409                                               x+j_coord_offsetC,x+j_coord_offsetD,
 410                                               &jx0,&jy0,&jz0);
 411
 412             /* Calculate displacement vector */
 413             dx00             = _mm_sub_ps(ix0,jx0);
 414             dy00             = _mm_sub_ps(iy0,jy0);
 415             dz00             = _mm_sub_ps(iz0,jz0);
 416             dx10             = _mm_sub_ps(ix1,jx0);
 417             dy10             = _mm_sub_ps(iy1,jy0);
 418             dz10             = _mm_sub_ps(iz1,jz0);
 419             dx20             = _mm_sub_ps(ix2,jx0);
 420             dy20             = _mm_sub_ps(iy2,jy0);
 421             dz20             = _mm_sub_ps(iz2,jz0);
 422
 423             /* Calculate squared distance and things based on it */
 424             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 425             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 426             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 427
 428             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 429             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 430             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 431
 432             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 433             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 434             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 435
 436             /* Load parameters for j particles */
 437             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 438                                                               charge+jnrC+0,charge+jnrD+0);
 439             vdwjidx0A        = 2*vdwtype[jnrA+0];
 440             vdwjidx0B        = 2*vdwtype[jnrB+0];
 441             vdwjidx0C        = 2*vdwtype[jnrC+0];
 442             vdwjidx0D        = 2*vdwtype[jnrD+0];
 443
 444             fjx0             = _mm_setzero_ps();
 445             fjy0             = _mm_setzero_ps();
 446             fjz0             = _mm_setzero_ps();
 447
 448             /**************************
 449              * CALCULATE INTERACTIONS *
 450              **************************/
 451
 452             r00              = _mm_mul_ps(rsq00,rinv00);
 453             r00              = _mm_andnot_ps(dummy_mask,r00);
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             qq00             = _mm_mul_ps(iq0,jq0);
 457             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 458                                          vdwparam+vdwioffset0+vdwjidx0B,
 459                                          vdwparam+vdwioffset0+vdwjidx0C,
 460                                          vdwparam+vdwioffset0+vdwjidx0D,
 461                                          &c6_00,&c12_00);
 462
 463             /* Calculate table index by multiplying r with table scale and truncate to integer */
 464             rt               = _mm_mul_ps(r00,vftabscale);
 465             vfitab           = _mm_cvttps_epi32(rt);
 466             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 467             vfitab           = _mm_slli_epi32(vfitab,3);
 468
 469             /* REACTION-FIELD ELECTROSTATICS */
 470             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 471             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 472
 473             /* CUBIC SPLINE TABLE DISPERSION */
 474             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 475             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 476             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 477             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 478             _MM_TRANSPOSE4_PS(Y,F,G,H);
 479             Heps             = _mm_mul_ps(vfeps,H);
 480             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 481             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 482             vvdw6            = _mm_mul_ps(c6_00,VV);
 483             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 484             fvdw6            = _mm_mul_ps(c6_00,FF);
 485
 486             /* CUBIC SPLINE TABLE REPULSION */
 487             vfitab           = _mm_add_epi32(vfitab,ifour);
 488             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 489             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 490             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 491             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 492             _MM_TRANSPOSE4_PS(Y,F,G,H);
 493             Heps             = _mm_mul_ps(vfeps,H);
 494             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 495             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 496             vvdw12           = _mm_mul_ps(c12_00,VV);
 497             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 498             fvdw12           = _mm_mul_ps(c12_00,FF);
 499             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 500             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 501
 502             /* Update potential sum for this i atom from the interaction with this j atom. */
 503             velec            = _mm_andnot_ps(dummy_mask,velec);
 504             velecsum         = _mm_add_ps(velecsum,velec);
 505             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 506             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 507
 508             fscal            = _mm_add_ps(felec,fvdw);
 509
 510             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 511
 512             /* Calculate temporary vectorial force */
 513             tx               = _mm_mul_ps(fscal,dx00);
 514             ty               = _mm_mul_ps(fscal,dy00);
 515             tz               = _mm_mul_ps(fscal,dz00);
 516
 517             /* Update vectorial force */
 518             fix0             = _mm_add_ps(fix0,tx);
 519             fiy0             = _mm_add_ps(fiy0,ty);
 520             fiz0             = _mm_add_ps(fiz0,tz);
 521
 522             fjx0             = _mm_add_ps(fjx0,tx);
 523             fjy0             = _mm_add_ps(fjy0,ty);
 524             fjz0             = _mm_add_ps(fjz0,tz);
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq10             = _mm_mul_ps(iq1,jq0);
 532
 533             /* REACTION-FIELD ELECTROSTATICS */
 534             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 535             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm_andnot_ps(dummy_mask,velec);
 539             velecsum         = _mm_add_ps(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm_mul_ps(fscal,dx10);
 547             ty               = _mm_mul_ps(fscal,dy10);
 548             tz               = _mm_mul_ps(fscal,dz10);
 549
 550             /* Update vectorial force */
 551             fix1             = _mm_add_ps(fix1,tx);
 552             fiy1             = _mm_add_ps(fiy1,ty);
 553             fiz1             = _mm_add_ps(fiz1,tz);
 554
 555             fjx0             = _mm_add_ps(fjx0,tx);
 556             fjy0             = _mm_add_ps(fjy0,ty);
 557             fjz0             = _mm_add_ps(fjz0,tz);
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             /* Compute parameters for interactions between i and j atoms */
 564             qq20             = _mm_mul_ps(iq2,jq0);
 565
 566             /* REACTION-FIELD ELECTROSTATICS */
 567             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 568             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velec            = _mm_andnot_ps(dummy_mask,velec);
 572             velecsum         = _mm_add_ps(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 577
 578             /* Calculate temporary vectorial force */
 579             tx               = _mm_mul_ps(fscal,dx20);
 580             ty               = _mm_mul_ps(fscal,dy20);
 581             tz               = _mm_mul_ps(fscal,dz20);
 582
 583             /* Update vectorial force */
 584             fix2             = _mm_add_ps(fix2,tx);
 585             fiy2             = _mm_add_ps(fiy2,ty);
 586             fiz2             = _mm_add_ps(fiz2,tz);
 587
 588             fjx0             = _mm_add_ps(fjx0,tx);
 589             fjy0             = _mm_add_ps(fjy0,ty);
 590             fjz0             = _mm_add_ps(fjz0,tz);
 591
 592             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 593             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 594             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 595             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 596
 597             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 598
 599             /* Inner loop uses 132 flops */
 600         }
 601
 602         /* End of innermost loop */
 603
 604         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 605                                               f+i_coord_offset,fshift+i_shift_offset);
 606
 607         ggid                        = gid[iidx];
 608         /* Update potential energies */
 609         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 610         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 611
 612         /* Increment number of inner iterations */
 613         inneriter                  += j_index_end - j_index_start;
 614
 615         /* Outer loop uses 20 flops */
 616     }
 617
 618     /* Increment number of outer iterations */
 619     outeriter        += nri;
 620
 621     /* Update outer/inner flops */
 622
 623     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*132);
 624 }
 625 /*
 626  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_single
 627  * Electrostatics interaction: ReactionField
 628  * VdW interaction:            CubicSplineTable
 629  * Geometry:                   Water3-Particle
 630  * Calculate force/pot:        Force
 631  */
 632 void
 633 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_single
 634                     (t_nblist                    * gmx_restrict       nlist,
 635                      rvec                        * gmx_restrict          xx,
 636                      rvec                        * gmx_restrict          ff,
 637                      t_forcerec                  * gmx_restrict          fr,
 638                      t_mdatoms                   * gmx_restrict     mdatoms,
 639                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 640                      t_nrnb                      * gmx_restrict        nrnb)
 641 {
 642     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 643      * just 0 for non-waters.
 644      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 645      * jnr indices corresponding to data put in the four positions in the SIMD register.
 646      */
 647     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 648     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 649     int              jnrA,jnrB,jnrC,jnrD;
 650     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 651     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 652     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 653     real             rcutoff_scalar;
 654     real             *shiftvec,*fshift,*x,*f;
 655     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 656     real             scratch[4*DIM];
 657     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 658     int              vdwioffset0;
 659     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 660     int              vdwioffset1;
 661     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 662     int              vdwioffset2;
 663     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 664     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 665     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 666     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 667     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 668     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 669     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 670     real             *charge;
 671     int              nvdwtype;
 672     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 673     int              *vdwtype;
 674     real             *vdwparam;
 675     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 676     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 677     __m128i          vfitab;
 678     __m128i          ifour       = _mm_set1_epi32(4);
 679     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 680     real             *vftab;
 681     __m128           dummy_mask,cutoff_mask;
 682     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 683     __m128           one     = _mm_set1_ps(1.0);
 684     __m128           two     = _mm_set1_ps(2.0);
 685     x                = xx[0];
 686     f                = ff[0];
 687
 688     nri              = nlist->nri;
 689     iinr             = nlist->iinr;
 690     jindex           = nlist->jindex;
 691     jjnr             = nlist->jjnr;
 692     shiftidx         = nlist->shift;
 693     gid              = nlist->gid;
 694     shiftvec         = fr->shift_vec[0];
 695     fshift           = fr->fshift[0];
 696     facel            = _mm_set1_ps(fr->epsfac);
 697     charge           = mdatoms->chargeA;
 698     krf              = _mm_set1_ps(fr->ic->k_rf);
 699     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 700     crf              = _mm_set1_ps(fr->ic->c_rf);
 701     nvdwtype         = fr->ntype;
 702     vdwparam         = fr->nbfp;
 703     vdwtype          = mdatoms->typeA;
 704
 705     vftab            = kernel_data->table_vdw->data;
 706     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 707
 708     /* Setup water-specific parameters */
 709     inr              = nlist->iinr[0];
 710     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 711     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 712     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 713     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 714
 715     /* Avoid stupid compiler warnings */
 716     jnrA = jnrB = jnrC = jnrD = 0;
 717     j_coord_offsetA = 0;
 718     j_coord_offsetB = 0;
 719     j_coord_offsetC = 0;
 720     j_coord_offsetD = 0;
 721
 722     outeriter        = 0;
 723     inneriter        = 0;
 724
 725     for(iidx=0;iidx<4*DIM;iidx++)
 726     {
 727         scratch[iidx] = 0.0;
 728     }
 729
 730     /* Start outer loop over neighborlists */
 731     for(iidx=0; iidx<nri; iidx++)
 732     {
 733         /* Load shift vector for this list */
 734         i_shift_offset   = DIM*shiftidx[iidx];
 735
 736         /* Load limits for loop over neighbors */
 737         j_index_start    = jindex[iidx];
 738         j_index_end      = jindex[iidx+1];
 739
 740         /* Get outer coordinate index */
 741         inr              = iinr[iidx];
 742         i_coord_offset   = DIM*inr;
 743
 744         /* Load i particle coords and add shift vector */
 745         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 746                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 747
 748         fix0             = _mm_setzero_ps();
 749         fiy0             = _mm_setzero_ps();
 750         fiz0             = _mm_setzero_ps();
 751         fix1             = _mm_setzero_ps();
 752         fiy1             = _mm_setzero_ps();
 753         fiz1             = _mm_setzero_ps();
 754         fix2             = _mm_setzero_ps();
 755         fiy2             = _mm_setzero_ps();
 756         fiz2             = _mm_setzero_ps();
 757
 758         /* Start inner kernel loop */
 759         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 760         {
 761
 762             /* Get j neighbor index, and coordinate index */
 763             jnrA             = jjnr[jidx];
 764             jnrB             = jjnr[jidx+1];
 765             jnrC             = jjnr[jidx+2];
 766             jnrD             = jjnr[jidx+3];
 767             j_coord_offsetA  = DIM*jnrA;
 768             j_coord_offsetB  = DIM*jnrB;
 769             j_coord_offsetC  = DIM*jnrC;
 770             j_coord_offsetD  = DIM*jnrD;
 771
 772             /* load j atom coordinates */
 773             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 774                                               x+j_coord_offsetC,x+j_coord_offsetD,
 775                                               &jx0,&jy0,&jz0);
 776
 777             /* Calculate displacement vector */
 778             dx00             = _mm_sub_ps(ix0,jx0);
 779             dy00             = _mm_sub_ps(iy0,jy0);
 780             dz00             = _mm_sub_ps(iz0,jz0);
 781             dx10             = _mm_sub_ps(ix1,jx0);
 782             dy10             = _mm_sub_ps(iy1,jy0);
 783             dz10             = _mm_sub_ps(iz1,jz0);
 784             dx20             = _mm_sub_ps(ix2,jx0);
 785             dy20             = _mm_sub_ps(iy2,jy0);
 786             dz20             = _mm_sub_ps(iz2,jz0);
 787
 788             /* Calculate squared distance and things based on it */
 789             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 790             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 791             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 792
 793             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 794             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 795             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 796
 797             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 798             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 799             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 800
 801             /* Load parameters for j particles */
 802             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 803                                                               charge+jnrC+0,charge+jnrD+0);
 804             vdwjidx0A        = 2*vdwtype[jnrA+0];
 805             vdwjidx0B        = 2*vdwtype[jnrB+0];
 806             vdwjidx0C        = 2*vdwtype[jnrC+0];
 807             vdwjidx0D        = 2*vdwtype[jnrD+0];
 808
 809             fjx0             = _mm_setzero_ps();
 810             fjy0             = _mm_setzero_ps();
 811             fjz0             = _mm_setzero_ps();
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             r00              = _mm_mul_ps(rsq00,rinv00);
 818
 819             /* Compute parameters for interactions between i and j atoms */
 820             qq00             = _mm_mul_ps(iq0,jq0);
 821             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 822                                          vdwparam+vdwioffset0+vdwjidx0B,
 823                                          vdwparam+vdwioffset0+vdwjidx0C,
 824                                          vdwparam+vdwioffset0+vdwjidx0D,
 825                                          &c6_00,&c12_00);
 826
 827             /* Calculate table index by multiplying r with table scale and truncate to integer */
 828             rt               = _mm_mul_ps(r00,vftabscale);
 829             vfitab           = _mm_cvttps_epi32(rt);
 830             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 831             vfitab           = _mm_slli_epi32(vfitab,3);
 832
 833             /* REACTION-FIELD ELECTROSTATICS */
 834             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 835
 836             /* CUBIC SPLINE TABLE DISPERSION */
 837             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 838             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 839             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 840             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 841             _MM_TRANSPOSE4_PS(Y,F,G,H);
 842             Heps             = _mm_mul_ps(vfeps,H);
 843             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 844             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 845             fvdw6            = _mm_mul_ps(c6_00,FF);
 846
 847             /* CUBIC SPLINE TABLE REPULSION */
 848             vfitab           = _mm_add_epi32(vfitab,ifour);
 849             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 850             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 851             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 852             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 853             _MM_TRANSPOSE4_PS(Y,F,G,H);
 854             Heps             = _mm_mul_ps(vfeps,H);
 855             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 856             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 857             fvdw12           = _mm_mul_ps(c12_00,FF);
 858             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 859
 860             fscal            = _mm_add_ps(felec,fvdw);
 861
 862             /* Calculate temporary vectorial force */
 863             tx               = _mm_mul_ps(fscal,dx00);
 864             ty               = _mm_mul_ps(fscal,dy00);
 865             tz               = _mm_mul_ps(fscal,dz00);
 866
 867             /* Update vectorial force */
 868             fix0             = _mm_add_ps(fix0,tx);
 869             fiy0             = _mm_add_ps(fiy0,ty);
 870             fiz0             = _mm_add_ps(fiz0,tz);
 871
 872             fjx0             = _mm_add_ps(fjx0,tx);
 873             fjy0             = _mm_add_ps(fjy0,ty);
 874             fjz0             = _mm_add_ps(fjz0,tz);
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             /* Compute parameters for interactions between i and j atoms */
 881             qq10             = _mm_mul_ps(iq1,jq0);
 882
 883             /* REACTION-FIELD ELECTROSTATICS */
 884             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 885
 886             fscal            = felec;
 887
 888             /* Calculate temporary vectorial force */
 889             tx               = _mm_mul_ps(fscal,dx10);
 890             ty               = _mm_mul_ps(fscal,dy10);
 891             tz               = _mm_mul_ps(fscal,dz10);
 892
 893             /* Update vectorial force */
 894             fix1             = _mm_add_ps(fix1,tx);
 895             fiy1             = _mm_add_ps(fiy1,ty);
 896             fiz1             = _mm_add_ps(fiz1,tz);
 897
 898             fjx0             = _mm_add_ps(fjx0,tx);
 899             fjy0             = _mm_add_ps(fjy0,ty);
 900             fjz0             = _mm_add_ps(fjz0,tz);
 901
 902             /**************************
 903              * CALCULATE INTERACTIONS *
 904              **************************/
 905
 906             /* Compute parameters for interactions between i and j atoms */
 907             qq20             = _mm_mul_ps(iq2,jq0);
 908
 909             /* REACTION-FIELD ELECTROSTATICS */
 910             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 911
 912             fscal            = felec;
 913
 914             /* Calculate temporary vectorial force */
 915             tx               = _mm_mul_ps(fscal,dx20);
 916             ty               = _mm_mul_ps(fscal,dy20);
 917             tz               = _mm_mul_ps(fscal,dz20);
 918
 919             /* Update vectorial force */
 920             fix2             = _mm_add_ps(fix2,tx);
 921             fiy2             = _mm_add_ps(fiy2,ty);
 922             fiz2             = _mm_add_ps(fiz2,tz);
 923
 924             fjx0             = _mm_add_ps(fjx0,tx);
 925             fjy0             = _mm_add_ps(fjy0,ty);
 926             fjz0             = _mm_add_ps(fjz0,tz);
 927
 928             fjptrA             = f+j_coord_offsetA;
 929             fjptrB             = f+j_coord_offsetB;
 930             fjptrC             = f+j_coord_offsetC;
 931             fjptrD             = f+j_coord_offsetD;
 932
 933             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 934
 935             /* Inner loop uses 108 flops */
 936         }
 937
 938         if(jidx<j_index_end)
 939         {
 940
 941             /* Get j neighbor index, and coordinate index */
 942             jnrlistA         = jjnr[jidx];
 943             jnrlistB         = jjnr[jidx+1];
 944             jnrlistC         = jjnr[jidx+2];
 945             jnrlistD         = jjnr[jidx+3];
 946             /* Sign of each element will be negative for non-real atoms.
 947              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 948              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 949              */
 950             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 951             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 952             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 953             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 954             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 955             j_coord_offsetA  = DIM*jnrA;
 956             j_coord_offsetB  = DIM*jnrB;
 957             j_coord_offsetC  = DIM*jnrC;
 958             j_coord_offsetD  = DIM*jnrD;
 959
 960             /* load j atom coordinates */
 961             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 962                                               x+j_coord_offsetC,x+j_coord_offsetD,
 963                                               &jx0,&jy0,&jz0);
 964
 965             /* Calculate displacement vector */
 966             dx00             = _mm_sub_ps(ix0,jx0);
 967             dy00             = _mm_sub_ps(iy0,jy0);
 968             dz00             = _mm_sub_ps(iz0,jz0);
 969             dx10             = _mm_sub_ps(ix1,jx0);
 970             dy10             = _mm_sub_ps(iy1,jy0);
 971             dz10             = _mm_sub_ps(iz1,jz0);
 972             dx20             = _mm_sub_ps(ix2,jx0);
 973             dy20             = _mm_sub_ps(iy2,jy0);
 974             dz20             = _mm_sub_ps(iz2,jz0);
 975
 976             /* Calculate squared distance and things based on it */
 977             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 978             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 979             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 980
 981             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 982             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 983             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 984
 985             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 986             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 987             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 988
 989             /* Load parameters for j particles */
 990             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 991                                                               charge+jnrC+0,charge+jnrD+0);
 992             vdwjidx0A        = 2*vdwtype[jnrA+0];
 993             vdwjidx0B        = 2*vdwtype[jnrB+0];
 994             vdwjidx0C        = 2*vdwtype[jnrC+0];
 995             vdwjidx0D        = 2*vdwtype[jnrD+0];
 996
 997             fjx0             = _mm_setzero_ps();
 998             fjy0             = _mm_setzero_ps();
 999             fjz0             = _mm_setzero_ps();
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             r00              = _mm_mul_ps(rsq00,rinv00);
1006             r00              = _mm_andnot_ps(dummy_mask,r00);
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             qq00             = _mm_mul_ps(iq0,jq0);
1010             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1011                                          vdwparam+vdwioffset0+vdwjidx0B,
1012                                          vdwparam+vdwioffset0+vdwjidx0C,
1013                                          vdwparam+vdwioffset0+vdwjidx0D,
1014                                          &c6_00,&c12_00);
1015
1016             /* Calculate table index by multiplying r with table scale and truncate to integer */
1017             rt               = _mm_mul_ps(r00,vftabscale);
1018             vfitab           = _mm_cvttps_epi32(rt);
1019             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1020             vfitab           = _mm_slli_epi32(vfitab,3);
1021
1022             /* REACTION-FIELD ELECTROSTATICS */
1023             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1024
1025             /* CUBIC SPLINE TABLE DISPERSION */
1026             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1027             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1028             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1029             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1030             _MM_TRANSPOSE4_PS(Y,F,G,H);
1031             Heps             = _mm_mul_ps(vfeps,H);
1032             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1033             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1034             fvdw6            = _mm_mul_ps(c6_00,FF);
1035
1036             /* CUBIC SPLINE TABLE REPULSION */
1037             vfitab           = _mm_add_epi32(vfitab,ifour);
1038             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1039             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1040             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1041             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1042             _MM_TRANSPOSE4_PS(Y,F,G,H);
1043             Heps             = _mm_mul_ps(vfeps,H);
1044             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1045             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1046             fvdw12           = _mm_mul_ps(c12_00,FF);
1047             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1048
1049             fscal            = _mm_add_ps(felec,fvdw);
1050
1051             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm_mul_ps(fscal,dx00);
1055             ty               = _mm_mul_ps(fscal,dy00);
1056             tz               = _mm_mul_ps(fscal,dz00);
1057
1058             /* Update vectorial force */
1059             fix0             = _mm_add_ps(fix0,tx);
1060             fiy0             = _mm_add_ps(fiy0,ty);
1061             fiz0             = _mm_add_ps(fiz0,tz);
1062
1063             fjx0             = _mm_add_ps(fjx0,tx);
1064             fjy0             = _mm_add_ps(fjy0,ty);
1065             fjz0             = _mm_add_ps(fjz0,tz);
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq10             = _mm_mul_ps(iq1,jq0);
1073
1074             /* REACTION-FIELD ELECTROSTATICS */
1075             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1076
1077             fscal            = felec;
1078
1079             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm_mul_ps(fscal,dx10);
1083             ty               = _mm_mul_ps(fscal,dy10);
1084             tz               = _mm_mul_ps(fscal,dz10);
1085
1086             /* Update vectorial force */
1087             fix1             = _mm_add_ps(fix1,tx);
1088             fiy1             = _mm_add_ps(fiy1,ty);
1089             fiz1             = _mm_add_ps(fiz1,tz);
1090
1091             fjx0             = _mm_add_ps(fjx0,tx);
1092             fjy0             = _mm_add_ps(fjy0,ty);
1093             fjz0             = _mm_add_ps(fjz0,tz);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq20             = _mm_mul_ps(iq2,jq0);
1101
1102             /* REACTION-FIELD ELECTROSTATICS */
1103             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm_mul_ps(fscal,dx20);
1111             ty               = _mm_mul_ps(fscal,dy20);
1112             tz               = _mm_mul_ps(fscal,dz20);
1113
1114             /* Update vectorial force */
1115             fix2             = _mm_add_ps(fix2,tx);
1116             fiy2             = _mm_add_ps(fiy2,ty);
1117             fiz2             = _mm_add_ps(fiz2,tz);
1118
1119             fjx0             = _mm_add_ps(fjx0,tx);
1120             fjy0             = _mm_add_ps(fjy0,ty);
1121             fjz0             = _mm_add_ps(fjz0,tz);
1122
1123             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1124             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1125             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1126             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1127
1128             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1129
1130             /* Inner loop uses 109 flops */
1131         }
1132
1133         /* End of innermost loop */
1134
1135         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1136                                               f+i_coord_offset,fshift+i_shift_offset);
1137
1138         /* Increment number of inner iterations */
1139         inneriter                  += j_index_end - j_index_start;
1140
1141         /* Outer loop uses 18 flops */
1142     }
1143
1144     /* Increment number of outer iterations */
1145     outeriter        += nri;
1146
1147     /* Update outer/inner flops */
1148
1149     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1150 }