src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_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_ElecCoul_VdwCSTab_GeomW4P1_VF_sse2_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_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              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128i          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m128           dummy_mask,cutoff_mask;
 110     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 111     __m128           one     = _mm_set1_ps(1.0);
 112     __m128           two     = _mm_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 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     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 137     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 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_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 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         fix3             = _mm_setzero_ps();
 183         fiy3             = _mm_setzero_ps();
 184         fiz3             = _mm_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm_setzero_ps();
 188         vvdwsum          = _mm_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                               x+j_coord_offsetC,x+j_coord_offsetD,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_ps(ix0,jx0);
 211             dy00             = _mm_sub_ps(iy0,jy0);
 212             dz00             = _mm_sub_ps(iz0,jz0);
 213             dx10             = _mm_sub_ps(ix1,jx0);
 214             dy10             = _mm_sub_ps(iy1,jy0);
 215             dz10             = _mm_sub_ps(iz1,jz0);
 216             dx20             = _mm_sub_ps(ix2,jx0);
 217             dy20             = _mm_sub_ps(iy2,jy0);
 218             dz20             = _mm_sub_ps(iz2,jz0);
 219             dx30             = _mm_sub_ps(ix3,jx0);
 220             dy30             = _mm_sub_ps(iy3,jy0);
 221             dz30             = _mm_sub_ps(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 228
 229             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 230             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 232             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 233
 234             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 236             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 240                                                               charge+jnrC+0,charge+jnrD+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243             vdwjidx0C        = 2*vdwtype[jnrC+0];
 244             vdwjidx0D        = 2*vdwtype[jnrD+0];
 245
 246             fjx0             = _mm_setzero_ps();
 247             fjy0             = _mm_setzero_ps();
 248             fjz0             = _mm_setzero_ps();
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             r00              = _mm_mul_ps(rsq00,rinv00);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 258                                          vdwparam+vdwioffset0+vdwjidx0B,
 259                                          vdwparam+vdwioffset0+vdwjidx0C,
 260                                          vdwparam+vdwioffset0+vdwjidx0D,
 261                                          &c6_00,&c12_00);
 262
 263             /* Calculate table index by multiplying r with table scale and truncate to integer */
 264             rt               = _mm_mul_ps(r00,vftabscale);
 265             vfitab           = _mm_cvttps_epi32(rt);
 266             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 267             vfitab           = _mm_slli_epi32(vfitab,3);
 268
 269             /* CUBIC SPLINE TABLE DISPERSION */
 270             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 271             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 272             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 273             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 274             _MM_TRANSPOSE4_PS(Y,F,G,H);
 275             Heps             = _mm_mul_ps(vfeps,H);
 276             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 277             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 278             vvdw6            = _mm_mul_ps(c6_00,VV);
 279             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 280             fvdw6            = _mm_mul_ps(c6_00,FF);
 281
 282             /* CUBIC SPLINE TABLE REPULSION */
 283             vfitab           = _mm_add_epi32(vfitab,ifour);
 284             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 285             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 286             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 287             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 288             _MM_TRANSPOSE4_PS(Y,F,G,H);
 289             Heps             = _mm_mul_ps(vfeps,H);
 290             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 291             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 292             vvdw12           = _mm_mul_ps(c12_00,VV);
 293             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 294             fvdw12           = _mm_mul_ps(c12_00,FF);
 295             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 296             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = fvdw;
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm_mul_ps(fscal,dx00);
 305             ty               = _mm_mul_ps(fscal,dy00);
 306             tz               = _mm_mul_ps(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm_add_ps(fix0,tx);
 310             fiy0             = _mm_add_ps(fiy0,ty);
 311             fiz0             = _mm_add_ps(fiz0,tz);
 312
 313             fjx0             = _mm_add_ps(fjx0,tx);
 314             fjy0             = _mm_add_ps(fjy0,ty);
 315             fjz0             = _mm_add_ps(fjz0,tz);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm_mul_ps(iq1,jq0);
 323
 324             /* COULOMB ELECTROSTATICS */
 325             velec            = _mm_mul_ps(qq10,rinv10);
 326             felec            = _mm_mul_ps(velec,rinvsq10);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_ps(fscal,dx10);
 335             ty               = _mm_mul_ps(fscal,dy10);
 336             tz               = _mm_mul_ps(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm_add_ps(fix1,tx);
 340             fiy1             = _mm_add_ps(fiy1,ty);
 341             fiz1             = _mm_add_ps(fiz1,tz);
 342
 343             fjx0             = _mm_add_ps(fjx0,tx);
 344             fjy0             = _mm_add_ps(fjy0,ty);
 345             fjz0             = _mm_add_ps(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm_mul_ps(iq2,jq0);
 353
 354             /* COULOMB ELECTROSTATICS */
 355             velec            = _mm_mul_ps(qq20,rinv20);
 356             felec            = _mm_mul_ps(velec,rinvsq20);
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm_add_ps(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm_mul_ps(fscal,dx20);
 365             ty               = _mm_mul_ps(fscal,dy20);
 366             tz               = _mm_mul_ps(fscal,dz20);
 367
 368             /* Update vectorial force */
 369             fix2             = _mm_add_ps(fix2,tx);
 370             fiy2             = _mm_add_ps(fiy2,ty);
 371             fiz2             = _mm_add_ps(fiz2,tz);
 372
 373             fjx0             = _mm_add_ps(fjx0,tx);
 374             fjy0             = _mm_add_ps(fjy0,ty);
 375             fjz0             = _mm_add_ps(fjz0,tz);
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             /* Compute parameters for interactions between i and j atoms */
 382             qq30             = _mm_mul_ps(iq3,jq0);
 383
 384             /* COULOMB ELECTROSTATICS */
 385             velec            = _mm_mul_ps(qq30,rinv30);
 386             felec            = _mm_mul_ps(velec,rinvsq30);
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velecsum         = _mm_add_ps(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm_mul_ps(fscal,dx30);
 395             ty               = _mm_mul_ps(fscal,dy30);
 396             tz               = _mm_mul_ps(fscal,dz30);
 397
 398             /* Update vectorial force */
 399             fix3             = _mm_add_ps(fix3,tx);
 400             fiy3             = _mm_add_ps(fiy3,ty);
 401             fiz3             = _mm_add_ps(fiz3,tz);
 402
 403             fjx0             = _mm_add_ps(fjx0,tx);
 404             fjy0             = _mm_add_ps(fjy0,ty);
 405             fjz0             = _mm_add_ps(fjz0,tz);
 406
 407             fjptrA             = f+j_coord_offsetA;
 408             fjptrB             = f+j_coord_offsetB;
 409             fjptrC             = f+j_coord_offsetC;
 410             fjptrD             = f+j_coord_offsetD;
 411
 412             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 413
 414             /* Inner loop uses 140 flops */
 415         }
 416
 417         if(jidx<j_index_end)
 418         {
 419
 420             /* Get j neighbor index, and coordinate index */
 421             jnrlistA         = jjnr[jidx];
 422             jnrlistB         = jjnr[jidx+1];
 423             jnrlistC         = jjnr[jidx+2];
 424             jnrlistD         = jjnr[jidx+3];
 425             /* Sign of each element will be negative for non-real atoms.
 426              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 427              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 428              */
 429             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 430             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 431             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 432             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 433             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 434             j_coord_offsetA  = DIM*jnrA;
 435             j_coord_offsetB  = DIM*jnrB;
 436             j_coord_offsetC  = DIM*jnrC;
 437             j_coord_offsetD  = DIM*jnrD;
 438
 439             /* load j atom coordinates */
 440             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 441                                               x+j_coord_offsetC,x+j_coord_offsetD,
 442                                               &jx0,&jy0,&jz0);
 443
 444             /* Calculate displacement vector */
 445             dx00             = _mm_sub_ps(ix0,jx0);
 446             dy00             = _mm_sub_ps(iy0,jy0);
 447             dz00             = _mm_sub_ps(iz0,jz0);
 448             dx10             = _mm_sub_ps(ix1,jx0);
 449             dy10             = _mm_sub_ps(iy1,jy0);
 450             dz10             = _mm_sub_ps(iz1,jz0);
 451             dx20             = _mm_sub_ps(ix2,jx0);
 452             dy20             = _mm_sub_ps(iy2,jy0);
 453             dz20             = _mm_sub_ps(iz2,jz0);
 454             dx30             = _mm_sub_ps(ix3,jx0);
 455             dy30             = _mm_sub_ps(iy3,jy0);
 456             dz30             = _mm_sub_ps(iz3,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 462             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 463
 464             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 465             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 466             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 467             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 468
 469             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 470             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 471             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 472
 473             /* Load parameters for j particles */
 474             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 475                                                               charge+jnrC+0,charge+jnrD+0);
 476             vdwjidx0A        = 2*vdwtype[jnrA+0];
 477             vdwjidx0B        = 2*vdwtype[jnrB+0];
 478             vdwjidx0C        = 2*vdwtype[jnrC+0];
 479             vdwjidx0D        = 2*vdwtype[jnrD+0];
 480
 481             fjx0             = _mm_setzero_ps();
 482             fjy0             = _mm_setzero_ps();
 483             fjz0             = _mm_setzero_ps();
 484
 485             /**************************
 486              * CALCULATE INTERACTIONS *
 487              **************************/
 488
 489             r00              = _mm_mul_ps(rsq00,rinv00);
 490             r00              = _mm_andnot_ps(dummy_mask,r00);
 491
 492             /* Compute parameters for interactions between i and j atoms */
 493             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 494                                          vdwparam+vdwioffset0+vdwjidx0B,
 495                                          vdwparam+vdwioffset0+vdwjidx0C,
 496                                          vdwparam+vdwioffset0+vdwjidx0D,
 497                                          &c6_00,&c12_00);
 498
 499             /* Calculate table index by multiplying r with table scale and truncate to integer */
 500             rt               = _mm_mul_ps(r00,vftabscale);
 501             vfitab           = _mm_cvttps_epi32(rt);
 502             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 503             vfitab           = _mm_slli_epi32(vfitab,3);
 504
 505             /* CUBIC SPLINE TABLE DISPERSION */
 506             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 507             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 508             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 509             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 510             _MM_TRANSPOSE4_PS(Y,F,G,H);
 511             Heps             = _mm_mul_ps(vfeps,H);
 512             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 513             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 514             vvdw6            = _mm_mul_ps(c6_00,VV);
 515             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 516             fvdw6            = _mm_mul_ps(c6_00,FF);
 517
 518             /* CUBIC SPLINE TABLE REPULSION */
 519             vfitab           = _mm_add_epi32(vfitab,ifour);
 520             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 521             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 522             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 523             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 524             _MM_TRANSPOSE4_PS(Y,F,G,H);
 525             Heps             = _mm_mul_ps(vfeps,H);
 526             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 527             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 528             vvdw12           = _mm_mul_ps(c12_00,VV);
 529             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 530             fvdw12           = _mm_mul_ps(c12_00,FF);
 531             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 532             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 536             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 537
 538             fscal            = fvdw;
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_ps(fscal,dx00);
 544             ty               = _mm_mul_ps(fscal,dy00);
 545             tz               = _mm_mul_ps(fscal,dz00);
 546
 547             /* Update vectorial force */
 548             fix0             = _mm_add_ps(fix0,tx);
 549             fiy0             = _mm_add_ps(fiy0,ty);
 550             fiz0             = _mm_add_ps(fiz0,tz);
 551
 552             fjx0             = _mm_add_ps(fjx0,tx);
 553             fjy0             = _mm_add_ps(fjy0,ty);
 554             fjz0             = _mm_add_ps(fjz0,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             qq10             = _mm_mul_ps(iq1,jq0);
 562
 563             /* COULOMB ELECTROSTATICS */
 564             velec            = _mm_mul_ps(qq10,rinv10);
 565             felec            = _mm_mul_ps(velec,rinvsq10);
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _mm_andnot_ps(dummy_mask,velec);
 569             velecsum         = _mm_add_ps(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_ps(fscal,dx10);
 577             ty               = _mm_mul_ps(fscal,dy10);
 578             tz               = _mm_mul_ps(fscal,dz10);
 579
 580             /* Update vectorial force */
 581             fix1             = _mm_add_ps(fix1,tx);
 582             fiy1             = _mm_add_ps(fiy1,ty);
 583             fiz1             = _mm_add_ps(fiz1,tz);
 584
 585             fjx0             = _mm_add_ps(fjx0,tx);
 586             fjy0             = _mm_add_ps(fjy0,ty);
 587             fjz0             = _mm_add_ps(fjz0,tz);
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq20             = _mm_mul_ps(iq2,jq0);
 595
 596             /* COULOMB ELECTROSTATICS */
 597             velec            = _mm_mul_ps(qq20,rinv20);
 598             felec            = _mm_mul_ps(velec,rinvsq20);
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velec            = _mm_andnot_ps(dummy_mask,velec);
 602             velecsum         = _mm_add_ps(velecsum,velec);
 603
 604             fscal            = felec;
 605
 606             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 607
 608             /* Calculate temporary vectorial force */
 609             tx               = _mm_mul_ps(fscal,dx20);
 610             ty               = _mm_mul_ps(fscal,dy20);
 611             tz               = _mm_mul_ps(fscal,dz20);
 612
 613             /* Update vectorial force */
 614             fix2             = _mm_add_ps(fix2,tx);
 615             fiy2             = _mm_add_ps(fiy2,ty);
 616             fiz2             = _mm_add_ps(fiz2,tz);
 617
 618             fjx0             = _mm_add_ps(fjx0,tx);
 619             fjy0             = _mm_add_ps(fjy0,ty);
 620             fjz0             = _mm_add_ps(fjz0,tz);
 621
 622             /**************************
 623              * CALCULATE INTERACTIONS *
 624              **************************/
 625
 626             /* Compute parameters for interactions between i and j atoms */
 627             qq30             = _mm_mul_ps(iq3,jq0);
 628
 629             /* COULOMB ELECTROSTATICS */
 630             velec            = _mm_mul_ps(qq30,rinv30);
 631             felec            = _mm_mul_ps(velec,rinvsq30);
 632
 633             /* Update potential sum for this i atom from the interaction with this j atom. */
 634             velec            = _mm_andnot_ps(dummy_mask,velec);
 635             velecsum         = _mm_add_ps(velecsum,velec);
 636
 637             fscal            = felec;
 638
 639             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 640
 641             /* Calculate temporary vectorial force */
 642             tx               = _mm_mul_ps(fscal,dx30);
 643             ty               = _mm_mul_ps(fscal,dy30);
 644             tz               = _mm_mul_ps(fscal,dz30);
 645
 646             /* Update vectorial force */
 647             fix3             = _mm_add_ps(fix3,tx);
 648             fiy3             = _mm_add_ps(fiy3,ty);
 649             fiz3             = _mm_add_ps(fiz3,tz);
 650
 651             fjx0             = _mm_add_ps(fjx0,tx);
 652             fjy0             = _mm_add_ps(fjy0,ty);
 653             fjz0             = _mm_add_ps(fjz0,tz);
 654
 655             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 656             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 657             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 658             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 659
 660             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 661
 662             /* Inner loop uses 141 flops */
 663         }
 664
 665         /* End of innermost loop */
 666
 667         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 668                                               f+i_coord_offset,fshift+i_shift_offset);
 669
 670         ggid                        = gid[iidx];
 671         /* Update potential energies */
 672         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 673         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 674
 675         /* Increment number of inner iterations */
 676         inneriter                  += j_index_end - j_index_start;
 677
 678         /* Outer loop uses 26 flops */
 679     }
 680
 681     /* Increment number of outer iterations */
 682     outeriter        += nri;
 683
 684     /* Update outer/inner flops */
 685
 686     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 687 }
 688 /*
 689  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse2_single
 690  * Electrostatics interaction: Coulomb
 691  * VdW interaction:            CubicSplineTable
 692  * Geometry:                   Water4-Particle
 693  * Calculate force/pot:        Force
 694  */
 695 void
 696 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse2_single
 697                     (t_nblist                    * gmx_restrict       nlist,
 698                      rvec                        * gmx_restrict          xx,
 699                      rvec                        * gmx_restrict          ff,
 700                      t_forcerec                  * gmx_restrict          fr,
 701                      t_mdatoms                   * gmx_restrict     mdatoms,
 702                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 703                      t_nrnb                      * gmx_restrict        nrnb)
 704 {
 705     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 706      * just 0 for non-waters.
 707      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 708      * jnr indices corresponding to data put in the four positions in the SIMD register.
 709      */
 710     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 711     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 712     int              jnrA,jnrB,jnrC,jnrD;
 713     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 714     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 715     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 716     real             rcutoff_scalar;
 717     real             *shiftvec,*fshift,*x,*f;
 718     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 719     real             scratch[4*DIM];
 720     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 721     int              vdwioffset0;
 722     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 723     int              vdwioffset1;
 724     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 725     int              vdwioffset2;
 726     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 727     int              vdwioffset3;
 728     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 729     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 730     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 731     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 732     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 733     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 734     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 735     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 736     real             *charge;
 737     int              nvdwtype;
 738     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 739     int              *vdwtype;
 740     real             *vdwparam;
 741     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 742     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 743     __m128i          vfitab;
 744     __m128i          ifour       = _mm_set1_epi32(4);
 745     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 746     real             *vftab;
 747     __m128           dummy_mask,cutoff_mask;
 748     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 749     __m128           one     = _mm_set1_ps(1.0);
 750     __m128           two     = _mm_set1_ps(2.0);
 751     x                = xx[0];
 752     f                = ff[0];
 753
 754     nri              = nlist->nri;
 755     iinr             = nlist->iinr;
 756     jindex           = nlist->jindex;
 757     jjnr             = nlist->jjnr;
 758     shiftidx         = nlist->shift;
 759     gid              = nlist->gid;
 760     shiftvec         = fr->shift_vec[0];
 761     fshift           = fr->fshift[0];
 762     facel            = _mm_set1_ps(fr->epsfac);
 763     charge           = mdatoms->chargeA;
 764     nvdwtype         = fr->ntype;
 765     vdwparam         = fr->nbfp;
 766     vdwtype          = mdatoms->typeA;
 767
 768     vftab            = kernel_data->table_vdw->data;
 769     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 770
 771     /* Setup water-specific parameters */
 772     inr              = nlist->iinr[0];
 773     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 774     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 775     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 776     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 777
 778     /* Avoid stupid compiler warnings */
 779     jnrA = jnrB = jnrC = jnrD = 0;
 780     j_coord_offsetA = 0;
 781     j_coord_offsetB = 0;
 782     j_coord_offsetC = 0;
 783     j_coord_offsetD = 0;
 784
 785     outeriter        = 0;
 786     inneriter        = 0;
 787
 788     for(iidx=0;iidx<4*DIM;iidx++)
 789     {
 790         scratch[iidx] = 0.0;
 791     }
 792
 793     /* Start outer loop over neighborlists */
 794     for(iidx=0; iidx<nri; iidx++)
 795     {
 796         /* Load shift vector for this list */
 797         i_shift_offset   = DIM*shiftidx[iidx];
 798
 799         /* Load limits for loop over neighbors */
 800         j_index_start    = jindex[iidx];
 801         j_index_end      = jindex[iidx+1];
 802
 803         /* Get outer coordinate index */
 804         inr              = iinr[iidx];
 805         i_coord_offset   = DIM*inr;
 806
 807         /* Load i particle coords and add shift vector */
 808         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 809                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 810
 811         fix0             = _mm_setzero_ps();
 812         fiy0             = _mm_setzero_ps();
 813         fiz0             = _mm_setzero_ps();
 814         fix1             = _mm_setzero_ps();
 815         fiy1             = _mm_setzero_ps();
 816         fiz1             = _mm_setzero_ps();
 817         fix2             = _mm_setzero_ps();
 818         fiy2             = _mm_setzero_ps();
 819         fiz2             = _mm_setzero_ps();
 820         fix3             = _mm_setzero_ps();
 821         fiy3             = _mm_setzero_ps();
 822         fiz3             = _mm_setzero_ps();
 823
 824         /* Start inner kernel loop */
 825         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 826         {
 827
 828             /* Get j neighbor index, and coordinate index */
 829             jnrA             = jjnr[jidx];
 830             jnrB             = jjnr[jidx+1];
 831             jnrC             = jjnr[jidx+2];
 832             jnrD             = jjnr[jidx+3];
 833             j_coord_offsetA  = DIM*jnrA;
 834             j_coord_offsetB  = DIM*jnrB;
 835             j_coord_offsetC  = DIM*jnrC;
 836             j_coord_offsetD  = DIM*jnrD;
 837
 838             /* load j atom coordinates */
 839             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 840                                               x+j_coord_offsetC,x+j_coord_offsetD,
 841                                               &jx0,&jy0,&jz0);
 842
 843             /* Calculate displacement vector */
 844             dx00             = _mm_sub_ps(ix0,jx0);
 845             dy00             = _mm_sub_ps(iy0,jy0);
 846             dz00             = _mm_sub_ps(iz0,jz0);
 847             dx10             = _mm_sub_ps(ix1,jx0);
 848             dy10             = _mm_sub_ps(iy1,jy0);
 849             dz10             = _mm_sub_ps(iz1,jz0);
 850             dx20             = _mm_sub_ps(ix2,jx0);
 851             dy20             = _mm_sub_ps(iy2,jy0);
 852             dz20             = _mm_sub_ps(iz2,jz0);
 853             dx30             = _mm_sub_ps(ix3,jx0);
 854             dy30             = _mm_sub_ps(iy3,jy0);
 855             dz30             = _mm_sub_ps(iz3,jz0);
 856
 857             /* Calculate squared distance and things based on it */
 858             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 859             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 860             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 861             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 862
 863             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 864             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 865             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 866             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 867
 868             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 869             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 870             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 871
 872             /* Load parameters for j particles */
 873             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 874                                                               charge+jnrC+0,charge+jnrD+0);
 875             vdwjidx0A        = 2*vdwtype[jnrA+0];
 876             vdwjidx0B        = 2*vdwtype[jnrB+0];
 877             vdwjidx0C        = 2*vdwtype[jnrC+0];
 878             vdwjidx0D        = 2*vdwtype[jnrD+0];
 879
 880             fjx0             = _mm_setzero_ps();
 881             fjy0             = _mm_setzero_ps();
 882             fjz0             = _mm_setzero_ps();
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             r00              = _mm_mul_ps(rsq00,rinv00);
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 892                                          vdwparam+vdwioffset0+vdwjidx0B,
 893                                          vdwparam+vdwioffset0+vdwjidx0C,
 894                                          vdwparam+vdwioffset0+vdwjidx0D,
 895                                          &c6_00,&c12_00);
 896
 897             /* Calculate table index by multiplying r with table scale and truncate to integer */
 898             rt               = _mm_mul_ps(r00,vftabscale);
 899             vfitab           = _mm_cvttps_epi32(rt);
 900             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 901             vfitab           = _mm_slli_epi32(vfitab,3);
 902
 903             /* CUBIC SPLINE TABLE DISPERSION */
 904             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 905             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 906             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 907             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 908             _MM_TRANSPOSE4_PS(Y,F,G,H);
 909             Heps             = _mm_mul_ps(vfeps,H);
 910             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 911             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 912             fvdw6            = _mm_mul_ps(c6_00,FF);
 913
 914             /* CUBIC SPLINE TABLE REPULSION */
 915             vfitab           = _mm_add_epi32(vfitab,ifour);
 916             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 917             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 918             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 919             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 920             _MM_TRANSPOSE4_PS(Y,F,G,H);
 921             Heps             = _mm_mul_ps(vfeps,H);
 922             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 923             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 924             fvdw12           = _mm_mul_ps(c12_00,FF);
 925             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 926
 927             fscal            = fvdw;
 928
 929             /* Calculate temporary vectorial force */
 930             tx               = _mm_mul_ps(fscal,dx00);
 931             ty               = _mm_mul_ps(fscal,dy00);
 932             tz               = _mm_mul_ps(fscal,dz00);
 933
 934             /* Update vectorial force */
 935             fix0             = _mm_add_ps(fix0,tx);
 936             fiy0             = _mm_add_ps(fiy0,ty);
 937             fiz0             = _mm_add_ps(fiz0,tz);
 938
 939             fjx0             = _mm_add_ps(fjx0,tx);
 940             fjy0             = _mm_add_ps(fjy0,ty);
 941             fjz0             = _mm_add_ps(fjz0,tz);
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             /* Compute parameters for interactions between i and j atoms */
 948             qq10             = _mm_mul_ps(iq1,jq0);
 949
 950             /* COULOMB ELECTROSTATICS */
 951             velec            = _mm_mul_ps(qq10,rinv10);
 952             felec            = _mm_mul_ps(velec,rinvsq10);
 953
 954             fscal            = felec;
 955
 956             /* Calculate temporary vectorial force */
 957             tx               = _mm_mul_ps(fscal,dx10);
 958             ty               = _mm_mul_ps(fscal,dy10);
 959             tz               = _mm_mul_ps(fscal,dz10);
 960
 961             /* Update vectorial force */
 962             fix1             = _mm_add_ps(fix1,tx);
 963             fiy1             = _mm_add_ps(fiy1,ty);
 964             fiz1             = _mm_add_ps(fiz1,tz);
 965
 966             fjx0             = _mm_add_ps(fjx0,tx);
 967             fjy0             = _mm_add_ps(fjy0,ty);
 968             fjz0             = _mm_add_ps(fjz0,tz);
 969
 970             /**************************
 971              * CALCULATE INTERACTIONS *
 972              **************************/
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             qq20             = _mm_mul_ps(iq2,jq0);
 976
 977             /* COULOMB ELECTROSTATICS */
 978             velec            = _mm_mul_ps(qq20,rinv20);
 979             felec            = _mm_mul_ps(velec,rinvsq20);
 980
 981             fscal            = felec;
 982
 983             /* Calculate temporary vectorial force */
 984             tx               = _mm_mul_ps(fscal,dx20);
 985             ty               = _mm_mul_ps(fscal,dy20);
 986             tz               = _mm_mul_ps(fscal,dz20);
 987
 988             /* Update vectorial force */
 989             fix2             = _mm_add_ps(fix2,tx);
 990             fiy2             = _mm_add_ps(fiy2,ty);
 991             fiz2             = _mm_add_ps(fiz2,tz);
 992
 993             fjx0             = _mm_add_ps(fjx0,tx);
 994             fjy0             = _mm_add_ps(fjy0,ty);
 995             fjz0             = _mm_add_ps(fjz0,tz);
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             /* Compute parameters for interactions between i and j atoms */
1002             qq30             = _mm_mul_ps(iq3,jq0);
1003
1004             /* COULOMB ELECTROSTATICS */
1005             velec            = _mm_mul_ps(qq30,rinv30);
1006             felec            = _mm_mul_ps(velec,rinvsq30);
1007
1008             fscal            = felec;
1009
1010             /* Calculate temporary vectorial force */
1011             tx               = _mm_mul_ps(fscal,dx30);
1012             ty               = _mm_mul_ps(fscal,dy30);
1013             tz               = _mm_mul_ps(fscal,dz30);
1014
1015             /* Update vectorial force */
1016             fix3             = _mm_add_ps(fix3,tx);
1017             fiy3             = _mm_add_ps(fiy3,ty);
1018             fiz3             = _mm_add_ps(fiz3,tz);
1019
1020             fjx0             = _mm_add_ps(fjx0,tx);
1021             fjy0             = _mm_add_ps(fjy0,ty);
1022             fjz0             = _mm_add_ps(fjz0,tz);
1023
1024             fjptrA             = f+j_coord_offsetA;
1025             fjptrB             = f+j_coord_offsetB;
1026             fjptrC             = f+j_coord_offsetC;
1027             fjptrD             = f+j_coord_offsetD;
1028
1029             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1030
1031             /* Inner loop uses 129 flops */
1032         }
1033
1034         if(jidx<j_index_end)
1035         {
1036
1037             /* Get j neighbor index, and coordinate index */
1038             jnrlistA         = jjnr[jidx];
1039             jnrlistB         = jjnr[jidx+1];
1040             jnrlistC         = jjnr[jidx+2];
1041             jnrlistD         = jjnr[jidx+3];
1042             /* Sign of each element will be negative for non-real atoms.
1043              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1044              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1045              */
1046             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1047             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1048             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1049             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1050             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1051             j_coord_offsetA  = DIM*jnrA;
1052             j_coord_offsetB  = DIM*jnrB;
1053             j_coord_offsetC  = DIM*jnrC;
1054             j_coord_offsetD  = DIM*jnrD;
1055
1056             /* load j atom coordinates */
1057             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1058                                               x+j_coord_offsetC,x+j_coord_offsetD,
1059                                               &jx0,&jy0,&jz0);
1060
1061             /* Calculate displacement vector */
1062             dx00             = _mm_sub_ps(ix0,jx0);
1063             dy00             = _mm_sub_ps(iy0,jy0);
1064             dz00             = _mm_sub_ps(iz0,jz0);
1065             dx10             = _mm_sub_ps(ix1,jx0);
1066             dy10             = _mm_sub_ps(iy1,jy0);
1067             dz10             = _mm_sub_ps(iz1,jz0);
1068             dx20             = _mm_sub_ps(ix2,jx0);
1069             dy20             = _mm_sub_ps(iy2,jy0);
1070             dz20             = _mm_sub_ps(iz2,jz0);
1071             dx30             = _mm_sub_ps(ix3,jx0);
1072             dy30             = _mm_sub_ps(iy3,jy0);
1073             dz30             = _mm_sub_ps(iz3,jz0);
1074
1075             /* Calculate squared distance and things based on it */
1076             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1077             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1078             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1079             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1080
1081             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1082             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1083             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1084             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1085
1086             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1087             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1088             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1089
1090             /* Load parameters for j particles */
1091             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1092                                                               charge+jnrC+0,charge+jnrD+0);
1093             vdwjidx0A        = 2*vdwtype[jnrA+0];
1094             vdwjidx0B        = 2*vdwtype[jnrB+0];
1095             vdwjidx0C        = 2*vdwtype[jnrC+0];
1096             vdwjidx0D        = 2*vdwtype[jnrD+0];
1097
1098             fjx0             = _mm_setzero_ps();
1099             fjy0             = _mm_setzero_ps();
1100             fjz0             = _mm_setzero_ps();
1101
1102             /**************************
1103              * CALCULATE INTERACTIONS *
1104              **************************/
1105
1106             r00              = _mm_mul_ps(rsq00,rinv00);
1107             r00              = _mm_andnot_ps(dummy_mask,r00);
1108
1109             /* Compute parameters for interactions between i and j atoms */
1110             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1111                                          vdwparam+vdwioffset0+vdwjidx0B,
1112                                          vdwparam+vdwioffset0+vdwjidx0C,
1113                                          vdwparam+vdwioffset0+vdwjidx0D,
1114                                          &c6_00,&c12_00);
1115
1116             /* Calculate table index by multiplying r with table scale and truncate to integer */
1117             rt               = _mm_mul_ps(r00,vftabscale);
1118             vfitab           = _mm_cvttps_epi32(rt);
1119             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1120             vfitab           = _mm_slli_epi32(vfitab,3);
1121
1122             /* CUBIC SPLINE TABLE DISPERSION */
1123             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1124             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1125             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1126             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1127             _MM_TRANSPOSE4_PS(Y,F,G,H);
1128             Heps             = _mm_mul_ps(vfeps,H);
1129             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1130             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1131             fvdw6            = _mm_mul_ps(c6_00,FF);
1132
1133             /* CUBIC SPLINE TABLE REPULSION */
1134             vfitab           = _mm_add_epi32(vfitab,ifour);
1135             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1136             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1137             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1138             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1139             _MM_TRANSPOSE4_PS(Y,F,G,H);
1140             Heps             = _mm_mul_ps(vfeps,H);
1141             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1142             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1143             fvdw12           = _mm_mul_ps(c12_00,FF);
1144             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1145
1146             fscal            = fvdw;
1147
1148             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm_mul_ps(fscal,dx00);
1152             ty               = _mm_mul_ps(fscal,dy00);
1153             tz               = _mm_mul_ps(fscal,dz00);
1154
1155             /* Update vectorial force */
1156             fix0             = _mm_add_ps(fix0,tx);
1157             fiy0             = _mm_add_ps(fiy0,ty);
1158             fiz0             = _mm_add_ps(fiz0,tz);
1159
1160             fjx0             = _mm_add_ps(fjx0,tx);
1161             fjy0             = _mm_add_ps(fjy0,ty);
1162             fjz0             = _mm_add_ps(fjz0,tz);
1163
1164             /**************************
1165              * CALCULATE INTERACTIONS *
1166              **************************/
1167
1168             /* Compute parameters for interactions between i and j atoms */
1169             qq10             = _mm_mul_ps(iq1,jq0);
1170
1171             /* COULOMB ELECTROSTATICS */
1172             velec            = _mm_mul_ps(qq10,rinv10);
1173             felec            = _mm_mul_ps(velec,rinvsq10);
1174
1175             fscal            = felec;
1176
1177             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1178
1179             /* Calculate temporary vectorial force */
1180             tx               = _mm_mul_ps(fscal,dx10);
1181             ty               = _mm_mul_ps(fscal,dy10);
1182             tz               = _mm_mul_ps(fscal,dz10);
1183
1184             /* Update vectorial force */
1185             fix1             = _mm_add_ps(fix1,tx);
1186             fiy1             = _mm_add_ps(fiy1,ty);
1187             fiz1             = _mm_add_ps(fiz1,tz);
1188
1189             fjx0             = _mm_add_ps(fjx0,tx);
1190             fjy0             = _mm_add_ps(fjy0,ty);
1191             fjz0             = _mm_add_ps(fjz0,tz);
1192
1193             /**************************
1194              * CALCULATE INTERACTIONS *
1195              **************************/
1196
1197             /* Compute parameters for interactions between i and j atoms */
1198             qq20             = _mm_mul_ps(iq2,jq0);
1199
1200             /* COULOMB ELECTROSTATICS */
1201             velec            = _mm_mul_ps(qq20,rinv20);
1202             felec            = _mm_mul_ps(velec,rinvsq20);
1203
1204             fscal            = felec;
1205
1206             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1207
1208             /* Calculate temporary vectorial force */
1209             tx               = _mm_mul_ps(fscal,dx20);
1210             ty               = _mm_mul_ps(fscal,dy20);
1211             tz               = _mm_mul_ps(fscal,dz20);
1212
1213             /* Update vectorial force */
1214             fix2             = _mm_add_ps(fix2,tx);
1215             fiy2             = _mm_add_ps(fiy2,ty);
1216             fiz2             = _mm_add_ps(fiz2,tz);
1217
1218             fjx0             = _mm_add_ps(fjx0,tx);
1219             fjy0             = _mm_add_ps(fjy0,ty);
1220             fjz0             = _mm_add_ps(fjz0,tz);
1221
1222             /**************************
1223              * CALCULATE INTERACTIONS *
1224              **************************/
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq30             = _mm_mul_ps(iq3,jq0);
1228
1229             /* COULOMB ELECTROSTATICS */
1230             velec            = _mm_mul_ps(qq30,rinv30);
1231             felec            = _mm_mul_ps(velec,rinvsq30);
1232
1233             fscal            = felec;
1234
1235             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1236
1237             /* Calculate temporary vectorial force */
1238             tx               = _mm_mul_ps(fscal,dx30);
1239             ty               = _mm_mul_ps(fscal,dy30);
1240             tz               = _mm_mul_ps(fscal,dz30);
1241
1242             /* Update vectorial force */
1243             fix3             = _mm_add_ps(fix3,tx);
1244             fiy3             = _mm_add_ps(fiy3,ty);
1245             fiz3             = _mm_add_ps(fiz3,tz);
1246
1247             fjx0             = _mm_add_ps(fjx0,tx);
1248             fjy0             = _mm_add_ps(fjy0,ty);
1249             fjz0             = _mm_add_ps(fjz0,tz);
1250
1251             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1252             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1253             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1254             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1255
1256             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1257
1258             /* Inner loop uses 130 flops */
1259         }
1260
1261         /* End of innermost loop */
1262
1263         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1264                                               f+i_coord_offset,fshift+i_shift_offset);
1265
1266         /* Increment number of inner iterations */
1267         inneriter                  += j_index_end - j_index_start;
1268
1269         /* Outer loop uses 24 flops */
1270     }
1271
1272     /* Increment number of outer iterations */
1273     outeriter        += nri;
1274
1275     /* Update outer/inner flops */
1276
1277     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1278 }