src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm_set1_ps(fr->ic->k_rf);
 126     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 127     crf              = _mm_set1_ps(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 138     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 139     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->rcoulomb;
 144     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 145     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm_setzero_ps();
 181         fiy0             = _mm_setzero_ps();
 182         fiz0             = _mm_setzero_ps();
 183         fix1             = _mm_setzero_ps();
 184         fiy1             = _mm_setzero_ps();
 185         fiz1             = _mm_setzero_ps();
 186         fix2             = _mm_setzero_ps();
 187         fiy2             = _mm_setzero_ps();
 188         fiz2             = _mm_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_ps();
 192         vvdwsum          = _mm_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                               x+j_coord_offsetC,x+j_coord_offsetD,
 211                                               &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm_sub_ps(ix0,jx0);
 215             dy00             = _mm_sub_ps(iy0,jy0);
 216             dz00             = _mm_sub_ps(iz0,jz0);
 217             dx10             = _mm_sub_ps(ix1,jx0);
 218             dy10             = _mm_sub_ps(iy1,jy0);
 219             dz10             = _mm_sub_ps(iz1,jz0);
 220             dx20             = _mm_sub_ps(ix2,jx0);
 221             dy20             = _mm_sub_ps(iy2,jy0);
 222             dz20             = _mm_sub_ps(iz2,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 226             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 227             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 228
 229             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 230             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 232
 233             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 234             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 239                                                               charge+jnrC+0,charge+jnrD+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244
 245             fjx0             = _mm_setzero_ps();
 246             fjy0             = _mm_setzero_ps();
 247             fjz0             = _mm_setzero_ps();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             if (gmx_mm_any_lt(rsq00,rcutoff2))
 254             {
 255
 256             r00              = _mm_mul_ps(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq00             = _mm_mul_ps(iq0,jq0);
 260             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 261                                          vdwparam+vdwioffset0+vdwjidx0B,
 262                                          vdwparam+vdwioffset0+vdwjidx0C,
 263                                          vdwparam+vdwioffset0+vdwjidx0D,
 264                                          &c6_00,&c12_00);
 265
 266             /* Calculate table index by multiplying r with table scale and truncate to integer */
 267             rt               = _mm_mul_ps(r00,vftabscale);
 268             vfitab           = _mm_cvttps_epi32(rt);
 269             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 270             vfitab           = _mm_slli_epi32(vfitab,3);
 271
 272             /* REACTION-FIELD ELECTROSTATICS */
 273             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 274             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 275
 276             /* CUBIC SPLINE TABLE DISPERSION */
 277             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 278             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 279             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 280             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 281             _MM_TRANSPOSE4_PS(Y,F,G,H);
 282             Heps             = _mm_mul_ps(vfeps,H);
 283             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 284             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 285             vvdw6            = _mm_mul_ps(c6_00,VV);
 286             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 287             fvdw6            = _mm_mul_ps(c6_00,FF);
 288
 289             /* CUBIC SPLINE TABLE REPULSION */
 290             vfitab           = _mm_add_epi32(vfitab,ifour);
 291             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 292             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 293             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 294             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 295             _MM_TRANSPOSE4_PS(Y,F,G,H);
 296             Heps             = _mm_mul_ps(vfeps,H);
 297             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 298             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 299             vvdw12           = _mm_mul_ps(c12_00,VV);
 300             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 301             fvdw12           = _mm_mul_ps(c12_00,FF);
 302             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 303             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 304
 305             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _mm_and_ps(velec,cutoff_mask);
 309             velecsum         = _mm_add_ps(velecsum,velec);
 310             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 311             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 312
 313             fscal            = _mm_add_ps(felec,fvdw);
 314
 315             fscal            = _mm_and_ps(fscal,cutoff_mask);
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm_mul_ps(fscal,dx00);
 319             ty               = _mm_mul_ps(fscal,dy00);
 320             tz               = _mm_mul_ps(fscal,dz00);
 321
 322             /* Update vectorial force */
 323             fix0             = _mm_add_ps(fix0,tx);
 324             fiy0             = _mm_add_ps(fiy0,ty);
 325             fiz0             = _mm_add_ps(fiz0,tz);
 326
 327             fjx0             = _mm_add_ps(fjx0,tx);
 328             fjy0             = _mm_add_ps(fjy0,ty);
 329             fjz0             = _mm_add_ps(fjz0,tz);
 330
 331             }
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             if (gmx_mm_any_lt(rsq10,rcutoff2))
 338             {
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq10             = _mm_mul_ps(iq1,jq0);
 342
 343             /* REACTION-FIELD ELECTROSTATICS */
 344             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 345             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 346
 347             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velec            = _mm_and_ps(velec,cutoff_mask);
 351             velecsum         = _mm_add_ps(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             fscal            = _mm_and_ps(fscal,cutoff_mask);
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm_mul_ps(fscal,dx10);
 359             ty               = _mm_mul_ps(fscal,dy10);
 360             tz               = _mm_mul_ps(fscal,dz10);
 361
 362             /* Update vectorial force */
 363             fix1             = _mm_add_ps(fix1,tx);
 364             fiy1             = _mm_add_ps(fiy1,ty);
 365             fiz1             = _mm_add_ps(fiz1,tz);
 366
 367             fjx0             = _mm_add_ps(fjx0,tx);
 368             fjy0             = _mm_add_ps(fjy0,ty);
 369             fjz0             = _mm_add_ps(fjz0,tz);
 370
 371             }
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             if (gmx_mm_any_lt(rsq20,rcutoff2))
 378             {
 379
 380             /* Compute parameters for interactions between i and j atoms */
 381             qq20             = _mm_mul_ps(iq2,jq0);
 382
 383             /* REACTION-FIELD ELECTROSTATICS */
 384             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 385             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 386
 387             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velec            = _mm_and_ps(velec,cutoff_mask);
 391             velecsum         = _mm_add_ps(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             fscal            = _mm_and_ps(fscal,cutoff_mask);
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm_mul_ps(fscal,dx20);
 399             ty               = _mm_mul_ps(fscal,dy20);
 400             tz               = _mm_mul_ps(fscal,dz20);
 401
 402             /* Update vectorial force */
 403             fix2             = _mm_add_ps(fix2,tx);
 404             fiy2             = _mm_add_ps(fiy2,ty);
 405             fiz2             = _mm_add_ps(fiz2,tz);
 406
 407             fjx0             = _mm_add_ps(fjx0,tx);
 408             fjy0             = _mm_add_ps(fjy0,ty);
 409             fjz0             = _mm_add_ps(fjz0,tz);
 410
 411             }
 412
 413             fjptrA             = f+j_coord_offsetA;
 414             fjptrB             = f+j_coord_offsetB;
 415             fjptrC             = f+j_coord_offsetC;
 416             fjptrD             = f+j_coord_offsetD;
 417
 418             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 419
 420             /* Inner loop uses 144 flops */
 421         }
 422
 423         if(jidx<j_index_end)
 424         {
 425
 426             /* Get j neighbor index, and coordinate index */
 427             jnrlistA         = jjnr[jidx];
 428             jnrlistB         = jjnr[jidx+1];
 429             jnrlistC         = jjnr[jidx+2];
 430             jnrlistD         = jjnr[jidx+3];
 431             /* Sign of each element will be negative for non-real atoms.
 432              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 433              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 434              */
 435             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 436             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 437             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 438             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 439             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 440             j_coord_offsetA  = DIM*jnrA;
 441             j_coord_offsetB  = DIM*jnrB;
 442             j_coord_offsetC  = DIM*jnrC;
 443             j_coord_offsetD  = DIM*jnrD;
 444
 445             /* load j atom coordinates */
 446             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 447                                               x+j_coord_offsetC,x+j_coord_offsetD,
 448                                               &jx0,&jy0,&jz0);
 449
 450             /* Calculate displacement vector */
 451             dx00             = _mm_sub_ps(ix0,jx0);
 452             dy00             = _mm_sub_ps(iy0,jy0);
 453             dz00             = _mm_sub_ps(iz0,jz0);
 454             dx10             = _mm_sub_ps(ix1,jx0);
 455             dy10             = _mm_sub_ps(iy1,jy0);
 456             dz10             = _mm_sub_ps(iz1,jz0);
 457             dx20             = _mm_sub_ps(ix2,jx0);
 458             dy20             = _mm_sub_ps(iy2,jy0);
 459             dz20             = _mm_sub_ps(iz2,jz0);
 460
 461             /* Calculate squared distance and things based on it */
 462             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 463             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 464             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 465
 466             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 467             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 468             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 469
 470             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 471             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 472             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 473
 474             /* Load parameters for j particles */
 475             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 476                                                               charge+jnrC+0,charge+jnrD+0);
 477             vdwjidx0A        = 2*vdwtype[jnrA+0];
 478             vdwjidx0B        = 2*vdwtype[jnrB+0];
 479             vdwjidx0C        = 2*vdwtype[jnrC+0];
 480             vdwjidx0D        = 2*vdwtype[jnrD+0];
 481
 482             fjx0             = _mm_setzero_ps();
 483             fjy0             = _mm_setzero_ps();
 484             fjz0             = _mm_setzero_ps();
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             if (gmx_mm_any_lt(rsq00,rcutoff2))
 491             {
 492
 493             r00              = _mm_mul_ps(rsq00,rinv00);
 494             r00              = _mm_andnot_ps(dummy_mask,r00);
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             qq00             = _mm_mul_ps(iq0,jq0);
 498             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 499                                          vdwparam+vdwioffset0+vdwjidx0B,
 500                                          vdwparam+vdwioffset0+vdwjidx0C,
 501                                          vdwparam+vdwioffset0+vdwjidx0D,
 502                                          &c6_00,&c12_00);
 503
 504             /* Calculate table index by multiplying r with table scale and truncate to integer */
 505             rt               = _mm_mul_ps(r00,vftabscale);
 506             vfitab           = _mm_cvttps_epi32(rt);
 507             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 508             vfitab           = _mm_slli_epi32(vfitab,3);
 509
 510             /* REACTION-FIELD ELECTROSTATICS */
 511             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 512             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 513
 514             /* CUBIC SPLINE TABLE DISPERSION */
 515             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 516             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 517             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 518             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 519             _MM_TRANSPOSE4_PS(Y,F,G,H);
 520             Heps             = _mm_mul_ps(vfeps,H);
 521             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 522             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 523             vvdw6            = _mm_mul_ps(c6_00,VV);
 524             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 525             fvdw6            = _mm_mul_ps(c6_00,FF);
 526
 527             /* CUBIC SPLINE TABLE REPULSION */
 528             vfitab           = _mm_add_epi32(vfitab,ifour);
 529             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 530             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 531             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 532             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 533             _MM_TRANSPOSE4_PS(Y,F,G,H);
 534             Heps             = _mm_mul_ps(vfeps,H);
 535             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 536             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 537             vvdw12           = _mm_mul_ps(c12_00,VV);
 538             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 539             fvdw12           = _mm_mul_ps(c12_00,FF);
 540             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 541             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 542
 543             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 544
 545             /* Update potential sum for this i atom from the interaction with this j atom. */
 546             velec            = _mm_and_ps(velec,cutoff_mask);
 547             velec            = _mm_andnot_ps(dummy_mask,velec);
 548             velecsum         = _mm_add_ps(velecsum,velec);
 549             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 550             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 551             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 552
 553             fscal            = _mm_add_ps(felec,fvdw);
 554
 555             fscal            = _mm_and_ps(fscal,cutoff_mask);
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_ps(fscal,dx00);
 561             ty               = _mm_mul_ps(fscal,dy00);
 562             tz               = _mm_mul_ps(fscal,dz00);
 563
 564             /* Update vectorial force */
 565             fix0             = _mm_add_ps(fix0,tx);
 566             fiy0             = _mm_add_ps(fiy0,ty);
 567             fiz0             = _mm_add_ps(fiz0,tz);
 568
 569             fjx0             = _mm_add_ps(fjx0,tx);
 570             fjy0             = _mm_add_ps(fjy0,ty);
 571             fjz0             = _mm_add_ps(fjz0,tz);
 572
 573             }
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             if (gmx_mm_any_lt(rsq10,rcutoff2))
 580             {
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq10             = _mm_mul_ps(iq1,jq0);
 584
 585             /* REACTION-FIELD ELECTROSTATICS */
 586             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 587             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 588
 589             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velec            = _mm_and_ps(velec,cutoff_mask);
 593             velec            = _mm_andnot_ps(dummy_mask,velec);
 594             velecsum         = _mm_add_ps(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             fscal            = _mm_and_ps(fscal,cutoff_mask);
 599
 600             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 601
 602             /* Calculate temporary vectorial force */
 603             tx               = _mm_mul_ps(fscal,dx10);
 604             ty               = _mm_mul_ps(fscal,dy10);
 605             tz               = _mm_mul_ps(fscal,dz10);
 606
 607             /* Update vectorial force */
 608             fix1             = _mm_add_ps(fix1,tx);
 609             fiy1             = _mm_add_ps(fiy1,ty);
 610             fiz1             = _mm_add_ps(fiz1,tz);
 611
 612             fjx0             = _mm_add_ps(fjx0,tx);
 613             fjy0             = _mm_add_ps(fjy0,ty);
 614             fjz0             = _mm_add_ps(fjz0,tz);
 615
 616             }
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             if (gmx_mm_any_lt(rsq20,rcutoff2))
 623             {
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq20             = _mm_mul_ps(iq2,jq0);
 627
 628             /* REACTION-FIELD ELECTROSTATICS */
 629             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 630             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 631
 632             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velec            = _mm_and_ps(velec,cutoff_mask);
 636             velec            = _mm_andnot_ps(dummy_mask,velec);
 637             velecsum         = _mm_add_ps(velecsum,velec);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_and_ps(fscal,cutoff_mask);
 642
 643             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm_mul_ps(fscal,dx20);
 647             ty               = _mm_mul_ps(fscal,dy20);
 648             tz               = _mm_mul_ps(fscal,dz20);
 649
 650             /* Update vectorial force */
 651             fix2             = _mm_add_ps(fix2,tx);
 652             fiy2             = _mm_add_ps(fiy2,ty);
 653             fiz2             = _mm_add_ps(fiz2,tz);
 654
 655             fjx0             = _mm_add_ps(fjx0,tx);
 656             fjy0             = _mm_add_ps(fjy0,ty);
 657             fjz0             = _mm_add_ps(fjz0,tz);
 658
 659             }
 660
 661             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 662             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 663             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 664             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 665
 666             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 667
 668             /* Inner loop uses 145 flops */
 669         }
 670
 671         /* End of innermost loop */
 672
 673         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 674                                               f+i_coord_offset,fshift+i_shift_offset);
 675
 676         ggid                        = gid[iidx];
 677         /* Update potential energies */
 678         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 679         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 680
 681         /* Increment number of inner iterations */
 682         inneriter                  += j_index_end - j_index_start;
 683
 684         /* Outer loop uses 20 flops */
 685     }
 686
 687     /* Increment number of outer iterations */
 688     outeriter        += nri;
 689
 690     /* Update outer/inner flops */
 691
 692     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 693 }
 694 /*
 695  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_single
 696  * Electrostatics interaction: ReactionField
 697  * VdW interaction:            CubicSplineTable
 698  * Geometry:                   Water3-Particle
 699  * Calculate force/pot:        Force
 700  */
 701 void
 702 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_single
 703                     (t_nblist                    * gmx_restrict       nlist,
 704                      rvec                        * gmx_restrict          xx,
 705                      rvec                        * gmx_restrict          ff,
 706                      t_forcerec                  * gmx_restrict          fr,
 707                      t_mdatoms                   * gmx_restrict     mdatoms,
 708                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 709                      t_nrnb                      * gmx_restrict        nrnb)
 710 {
 711     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 712      * just 0 for non-waters.
 713      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 714      * jnr indices corresponding to data put in the four positions in the SIMD register.
 715      */
 716     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 717     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 718     int              jnrA,jnrB,jnrC,jnrD;
 719     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 720     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 721     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 722     real             rcutoff_scalar;
 723     real             *shiftvec,*fshift,*x,*f;
 724     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 725     real             scratch[4*DIM];
 726     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 727     int              vdwioffset0;
 728     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 729     int              vdwioffset1;
 730     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 731     int              vdwioffset2;
 732     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 733     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 734     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 735     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 736     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 737     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 738     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 739     real             *charge;
 740     int              nvdwtype;
 741     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 742     int              *vdwtype;
 743     real             *vdwparam;
 744     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 745     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 746     __m128i          vfitab;
 747     __m128i          ifour       = _mm_set1_epi32(4);
 748     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 749     real             *vftab;
 750     __m128           dummy_mask,cutoff_mask;
 751     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 752     __m128           one     = _mm_set1_ps(1.0);
 753     __m128           two     = _mm_set1_ps(2.0);
 754     x                = xx[0];
 755     f                = ff[0];
 756
 757     nri              = nlist->nri;
 758     iinr             = nlist->iinr;
 759     jindex           = nlist->jindex;
 760     jjnr             = nlist->jjnr;
 761     shiftidx         = nlist->shift;
 762     gid              = nlist->gid;
 763     shiftvec         = fr->shift_vec[0];
 764     fshift           = fr->fshift[0];
 765     facel            = _mm_set1_ps(fr->epsfac);
 766     charge           = mdatoms->chargeA;
 767     krf              = _mm_set1_ps(fr->ic->k_rf);
 768     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 769     crf              = _mm_set1_ps(fr->ic->c_rf);
 770     nvdwtype         = fr->ntype;
 771     vdwparam         = fr->nbfp;
 772     vdwtype          = mdatoms->typeA;
 773
 774     vftab            = kernel_data->table_vdw->data;
 775     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 776
 777     /* Setup water-specific parameters */
 778     inr              = nlist->iinr[0];
 779     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 780     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 781     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 782     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 783
 784     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 785     rcutoff_scalar   = fr->rcoulomb;
 786     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 787     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 788
 789     /* Avoid stupid compiler warnings */
 790     jnrA = jnrB = jnrC = jnrD = 0;
 791     j_coord_offsetA = 0;
 792     j_coord_offsetB = 0;
 793     j_coord_offsetC = 0;
 794     j_coord_offsetD = 0;
 795
 796     outeriter        = 0;
 797     inneriter        = 0;
 798
 799     for(iidx=0;iidx<4*DIM;iidx++)
 800     {
 801         scratch[iidx] = 0.0;
 802     }
 803
 804     /* Start outer loop over neighborlists */
 805     for(iidx=0; iidx<nri; iidx++)
 806     {
 807         /* Load shift vector for this list */
 808         i_shift_offset   = DIM*shiftidx[iidx];
 809
 810         /* Load limits for loop over neighbors */
 811         j_index_start    = jindex[iidx];
 812         j_index_end      = jindex[iidx+1];
 813
 814         /* Get outer coordinate index */
 815         inr              = iinr[iidx];
 816         i_coord_offset   = DIM*inr;
 817
 818         /* Load i particle coords and add shift vector */
 819         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 820                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 821
 822         fix0             = _mm_setzero_ps();
 823         fiy0             = _mm_setzero_ps();
 824         fiz0             = _mm_setzero_ps();
 825         fix1             = _mm_setzero_ps();
 826         fiy1             = _mm_setzero_ps();
 827         fiz1             = _mm_setzero_ps();
 828         fix2             = _mm_setzero_ps();
 829         fiy2             = _mm_setzero_ps();
 830         fiz2             = _mm_setzero_ps();
 831
 832         /* Start inner kernel loop */
 833         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrA             = jjnr[jidx];
 838             jnrB             = jjnr[jidx+1];
 839             jnrC             = jjnr[jidx+2];
 840             jnrD             = jjnr[jidx+3];
 841             j_coord_offsetA  = DIM*jnrA;
 842             j_coord_offsetB  = DIM*jnrB;
 843             j_coord_offsetC  = DIM*jnrC;
 844             j_coord_offsetD  = DIM*jnrD;
 845
 846             /* load j atom coordinates */
 847             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 848                                               x+j_coord_offsetC,x+j_coord_offsetD,
 849                                               &jx0,&jy0,&jz0);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm_sub_ps(ix0,jx0);
 853             dy00             = _mm_sub_ps(iy0,jy0);
 854             dz00             = _mm_sub_ps(iz0,jz0);
 855             dx10             = _mm_sub_ps(ix1,jx0);
 856             dy10             = _mm_sub_ps(iy1,jy0);
 857             dz10             = _mm_sub_ps(iz1,jz0);
 858             dx20             = _mm_sub_ps(ix2,jx0);
 859             dy20             = _mm_sub_ps(iy2,jy0);
 860             dz20             = _mm_sub_ps(iz2,jz0);
 861
 862             /* Calculate squared distance and things based on it */
 863             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 864             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 865             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 866
 867             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 868             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 869             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 870
 871             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 872             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 873             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 874
 875             /* Load parameters for j particles */
 876             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 877                                                               charge+jnrC+0,charge+jnrD+0);
 878             vdwjidx0A        = 2*vdwtype[jnrA+0];
 879             vdwjidx0B        = 2*vdwtype[jnrB+0];
 880             vdwjidx0C        = 2*vdwtype[jnrC+0];
 881             vdwjidx0D        = 2*vdwtype[jnrD+0];
 882
 883             fjx0             = _mm_setzero_ps();
 884             fjy0             = _mm_setzero_ps();
 885             fjz0             = _mm_setzero_ps();
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             if (gmx_mm_any_lt(rsq00,rcutoff2))
 892             {
 893
 894             r00              = _mm_mul_ps(rsq00,rinv00);
 895
 896             /* Compute parameters for interactions between i and j atoms */
 897             qq00             = _mm_mul_ps(iq0,jq0);
 898             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 899                                          vdwparam+vdwioffset0+vdwjidx0B,
 900                                          vdwparam+vdwioffset0+vdwjidx0C,
 901                                          vdwparam+vdwioffset0+vdwjidx0D,
 902                                          &c6_00,&c12_00);
 903
 904             /* Calculate table index by multiplying r with table scale and truncate to integer */
 905             rt               = _mm_mul_ps(r00,vftabscale);
 906             vfitab           = _mm_cvttps_epi32(rt);
 907             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 908             vfitab           = _mm_slli_epi32(vfitab,3);
 909
 910             /* REACTION-FIELD ELECTROSTATICS */
 911             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 912
 913             /* CUBIC SPLINE TABLE DISPERSION */
 914             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 915             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 916             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 917             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 918             _MM_TRANSPOSE4_PS(Y,F,G,H);
 919             Heps             = _mm_mul_ps(vfeps,H);
 920             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 921             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 922             fvdw6            = _mm_mul_ps(c6_00,FF);
 923
 924             /* CUBIC SPLINE TABLE REPULSION */
 925             vfitab           = _mm_add_epi32(vfitab,ifour);
 926             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 927             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 928             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 929             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 930             _MM_TRANSPOSE4_PS(Y,F,G,H);
 931             Heps             = _mm_mul_ps(vfeps,H);
 932             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 933             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 934             fvdw12           = _mm_mul_ps(c12_00,FF);
 935             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 936
 937             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 938
 939             fscal            = _mm_add_ps(felec,fvdw);
 940
 941             fscal            = _mm_and_ps(fscal,cutoff_mask);
 942
 943             /* Calculate temporary vectorial force */
 944             tx               = _mm_mul_ps(fscal,dx00);
 945             ty               = _mm_mul_ps(fscal,dy00);
 946             tz               = _mm_mul_ps(fscal,dz00);
 947
 948             /* Update vectorial force */
 949             fix0             = _mm_add_ps(fix0,tx);
 950             fiy0             = _mm_add_ps(fiy0,ty);
 951             fiz0             = _mm_add_ps(fiz0,tz);
 952
 953             fjx0             = _mm_add_ps(fjx0,tx);
 954             fjy0             = _mm_add_ps(fjy0,ty);
 955             fjz0             = _mm_add_ps(fjz0,tz);
 956
 957             }
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             if (gmx_mm_any_lt(rsq10,rcutoff2))
 964             {
 965
 966             /* Compute parameters for interactions between i and j atoms */
 967             qq10             = _mm_mul_ps(iq1,jq0);
 968
 969             /* REACTION-FIELD ELECTROSTATICS */
 970             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 971
 972             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 973
 974             fscal            = felec;
 975
 976             fscal            = _mm_and_ps(fscal,cutoff_mask);
 977
 978             /* Calculate temporary vectorial force */
 979             tx               = _mm_mul_ps(fscal,dx10);
 980             ty               = _mm_mul_ps(fscal,dy10);
 981             tz               = _mm_mul_ps(fscal,dz10);
 982
 983             /* Update vectorial force */
 984             fix1             = _mm_add_ps(fix1,tx);
 985             fiy1             = _mm_add_ps(fiy1,ty);
 986             fiz1             = _mm_add_ps(fiz1,tz);
 987
 988             fjx0             = _mm_add_ps(fjx0,tx);
 989             fjy0             = _mm_add_ps(fjy0,ty);
 990             fjz0             = _mm_add_ps(fjz0,tz);
 991
 992             }
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             if (gmx_mm_any_lt(rsq20,rcutoff2))
 999             {
1000
1001             /* Compute parameters for interactions between i and j atoms */
1002             qq20             = _mm_mul_ps(iq2,jq0);
1003
1004             /* REACTION-FIELD ELECTROSTATICS */
1005             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1006
1007             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1008
1009             fscal            = felec;
1010
1011             fscal            = _mm_and_ps(fscal,cutoff_mask);
1012
1013             /* Calculate temporary vectorial force */
1014             tx               = _mm_mul_ps(fscal,dx20);
1015             ty               = _mm_mul_ps(fscal,dy20);
1016             tz               = _mm_mul_ps(fscal,dz20);
1017
1018             /* Update vectorial force */
1019             fix2             = _mm_add_ps(fix2,tx);
1020             fiy2             = _mm_add_ps(fiy2,ty);
1021             fiz2             = _mm_add_ps(fiz2,tz);
1022
1023             fjx0             = _mm_add_ps(fjx0,tx);
1024             fjy0             = _mm_add_ps(fjy0,ty);
1025             fjz0             = _mm_add_ps(fjz0,tz);
1026
1027             }
1028
1029             fjptrA             = f+j_coord_offsetA;
1030             fjptrB             = f+j_coord_offsetB;
1031             fjptrC             = f+j_coord_offsetC;
1032             fjptrD             = f+j_coord_offsetD;
1033
1034             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1035
1036             /* Inner loop uses 117 flops */
1037         }
1038
1039         if(jidx<j_index_end)
1040         {
1041
1042             /* Get j neighbor index, and coordinate index */
1043             jnrlistA         = jjnr[jidx];
1044             jnrlistB         = jjnr[jidx+1];
1045             jnrlistC         = jjnr[jidx+2];
1046             jnrlistD         = jjnr[jidx+3];
1047             /* Sign of each element will be negative for non-real atoms.
1048              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1049              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1050              */
1051             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1052             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1053             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1054             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1055             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1056             j_coord_offsetA  = DIM*jnrA;
1057             j_coord_offsetB  = DIM*jnrB;
1058             j_coord_offsetC  = DIM*jnrC;
1059             j_coord_offsetD  = DIM*jnrD;
1060
1061             /* load j atom coordinates */
1062             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1063                                               x+j_coord_offsetC,x+j_coord_offsetD,
1064                                               &jx0,&jy0,&jz0);
1065
1066             /* Calculate displacement vector */
1067             dx00             = _mm_sub_ps(ix0,jx0);
1068             dy00             = _mm_sub_ps(iy0,jy0);
1069             dz00             = _mm_sub_ps(iz0,jz0);
1070             dx10             = _mm_sub_ps(ix1,jx0);
1071             dy10             = _mm_sub_ps(iy1,jy0);
1072             dz10             = _mm_sub_ps(iz1,jz0);
1073             dx20             = _mm_sub_ps(ix2,jx0);
1074             dy20             = _mm_sub_ps(iy2,jy0);
1075             dz20             = _mm_sub_ps(iz2,jz0);
1076
1077             /* Calculate squared distance and things based on it */
1078             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1079             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1080             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1081
1082             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1083             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1084             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1085
1086             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1087             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1088             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
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             if (gmx_mm_any_lt(rsq00,rcutoff2))
1107             {
1108
1109             r00              = _mm_mul_ps(rsq00,rinv00);
1110             r00              = _mm_andnot_ps(dummy_mask,r00);
1111
1112             /* Compute parameters for interactions between i and j atoms */
1113             qq00             = _mm_mul_ps(iq0,jq0);
1114             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1115                                          vdwparam+vdwioffset0+vdwjidx0B,
1116                                          vdwparam+vdwioffset0+vdwjidx0C,
1117                                          vdwparam+vdwioffset0+vdwjidx0D,
1118                                          &c6_00,&c12_00);
1119
1120             /* Calculate table index by multiplying r with table scale and truncate to integer */
1121             rt               = _mm_mul_ps(r00,vftabscale);
1122             vfitab           = _mm_cvttps_epi32(rt);
1123             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1124             vfitab           = _mm_slli_epi32(vfitab,3);
1125
1126             /* REACTION-FIELD ELECTROSTATICS */
1127             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1128
1129             /* CUBIC SPLINE TABLE DISPERSION */
1130             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1131             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1132             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1133             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1134             _MM_TRANSPOSE4_PS(Y,F,G,H);
1135             Heps             = _mm_mul_ps(vfeps,H);
1136             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1137             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1138             fvdw6            = _mm_mul_ps(c6_00,FF);
1139
1140             /* CUBIC SPLINE TABLE REPULSION */
1141             vfitab           = _mm_add_epi32(vfitab,ifour);
1142             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1143             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1144             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1145             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1146             _MM_TRANSPOSE4_PS(Y,F,G,H);
1147             Heps             = _mm_mul_ps(vfeps,H);
1148             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1149             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1150             fvdw12           = _mm_mul_ps(c12_00,FF);
1151             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1152
1153             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1154
1155             fscal            = _mm_add_ps(felec,fvdw);
1156
1157             fscal            = _mm_and_ps(fscal,cutoff_mask);
1158
1159             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1160
1161             /* Calculate temporary vectorial force */
1162             tx               = _mm_mul_ps(fscal,dx00);
1163             ty               = _mm_mul_ps(fscal,dy00);
1164             tz               = _mm_mul_ps(fscal,dz00);
1165
1166             /* Update vectorial force */
1167             fix0             = _mm_add_ps(fix0,tx);
1168             fiy0             = _mm_add_ps(fiy0,ty);
1169             fiz0             = _mm_add_ps(fiz0,tz);
1170
1171             fjx0             = _mm_add_ps(fjx0,tx);
1172             fjy0             = _mm_add_ps(fjy0,ty);
1173             fjz0             = _mm_add_ps(fjz0,tz);
1174
1175             }
1176
1177             /**************************
1178              * CALCULATE INTERACTIONS *
1179              **************************/
1180
1181             if (gmx_mm_any_lt(rsq10,rcutoff2))
1182             {
1183
1184             /* Compute parameters for interactions between i and j atoms */
1185             qq10             = _mm_mul_ps(iq1,jq0);
1186
1187             /* REACTION-FIELD ELECTROSTATICS */
1188             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1189
1190             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1191
1192             fscal            = felec;
1193
1194             fscal            = _mm_and_ps(fscal,cutoff_mask);
1195
1196             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1197
1198             /* Calculate temporary vectorial force */
1199             tx               = _mm_mul_ps(fscal,dx10);
1200             ty               = _mm_mul_ps(fscal,dy10);
1201             tz               = _mm_mul_ps(fscal,dz10);
1202
1203             /* Update vectorial force */
1204             fix1             = _mm_add_ps(fix1,tx);
1205             fiy1             = _mm_add_ps(fiy1,ty);
1206             fiz1             = _mm_add_ps(fiz1,tz);
1207
1208             fjx0             = _mm_add_ps(fjx0,tx);
1209             fjy0             = _mm_add_ps(fjy0,ty);
1210             fjz0             = _mm_add_ps(fjz0,tz);
1211
1212             }
1213
1214             /**************************
1215              * CALCULATE INTERACTIONS *
1216              **************************/
1217
1218             if (gmx_mm_any_lt(rsq20,rcutoff2))
1219             {
1220
1221             /* Compute parameters for interactions between i and j atoms */
1222             qq20             = _mm_mul_ps(iq2,jq0);
1223
1224             /* REACTION-FIELD ELECTROSTATICS */
1225             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1226
1227             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1228
1229             fscal            = felec;
1230
1231             fscal            = _mm_and_ps(fscal,cutoff_mask);
1232
1233             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1234
1235             /* Calculate temporary vectorial force */
1236             tx               = _mm_mul_ps(fscal,dx20);
1237             ty               = _mm_mul_ps(fscal,dy20);
1238             tz               = _mm_mul_ps(fscal,dz20);
1239
1240             /* Update vectorial force */
1241             fix2             = _mm_add_ps(fix2,tx);
1242             fiy2             = _mm_add_ps(fiy2,ty);
1243             fiz2             = _mm_add_ps(fiz2,tz);
1244
1245             fjx0             = _mm_add_ps(fjx0,tx);
1246             fjy0             = _mm_add_ps(fjy0,ty);
1247             fjz0             = _mm_add_ps(fjz0,tz);
1248
1249             }
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 118 flops */
1259         }
1260
1261         /* End of innermost loop */
1262
1263         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
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 18 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_W3_F,outeriter*18 + inneriter*118);
1278 }