src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwNone_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 "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_ElecCSTab_VdwNone_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwNone_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     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 101     real             *vftab;
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119
 120     vftab            = kernel_data->table_elec->data;
 121     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 126     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 127     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165         fix1             = _mm_setzero_ps();
 166         fiy1             = _mm_setzero_ps();
 167         fiz1             = _mm_setzero_ps();
 168         fix2             = _mm_setzero_ps();
 169         fiy2             = _mm_setzero_ps();
 170         fiz2             = _mm_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186             j_coord_offsetC  = DIM*jnrC;
 187             j_coord_offsetD  = DIM*jnrD;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               x+j_coord_offsetC,x+j_coord_offsetD,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_ps(ix0,jx0);
 196             dy00             = _mm_sub_ps(iy0,jy0);
 197             dz00             = _mm_sub_ps(iz0,jz0);
 198             dx10             = _mm_sub_ps(ix1,jx0);
 199             dy10             = _mm_sub_ps(iy1,jy0);
 200             dz10             = _mm_sub_ps(iz1,jz0);
 201             dx20             = _mm_sub_ps(ix2,jx0);
 202             dy20             = _mm_sub_ps(iy2,jy0);
 203             dz20             = _mm_sub_ps(iz2,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 207             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 208             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 209
 210             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 211             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 212             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                               charge+jnrC+0,charge+jnrD+0);
 217
 218             fjx0             = _mm_setzero_ps();
 219             fjy0             = _mm_setzero_ps();
 220             fjz0             = _mm_setzero_ps();
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             r00              = _mm_mul_ps(rsq00,rinv00);
 227
 228             /* Compute parameters for interactions between i and j atoms */
 229             qq00             = _mm_mul_ps(iq0,jq0);
 230
 231             /* Calculate table index by multiplying r with table scale and truncate to integer */
 232             rt               = _mm_mul_ps(r00,vftabscale);
 233             vfitab           = _mm_cvttps_epi32(rt);
 234             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 235             vfitab           = _mm_slli_epi32(vfitab,2);
 236
 237             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 238             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 239             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 240             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 241             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 242             _MM_TRANSPOSE4_PS(Y,F,G,H);
 243             Heps             = _mm_mul_ps(vfeps,H);
 244             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 245             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 246             velec            = _mm_mul_ps(qq00,VV);
 247             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 248             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 249
 250             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             velecsum         = _mm_add_ps(velecsum,velec);
 252
 253             fscal            = felec;
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm_mul_ps(fscal,dx00);
 257             ty               = _mm_mul_ps(fscal,dy00);
 258             tz               = _mm_mul_ps(fscal,dz00);
 259
 260             /* Update vectorial force */
 261             fix0             = _mm_add_ps(fix0,tx);
 262             fiy0             = _mm_add_ps(fiy0,ty);
 263             fiz0             = _mm_add_ps(fiz0,tz);
 264
 265             fjx0             = _mm_add_ps(fjx0,tx);
 266             fjy0             = _mm_add_ps(fjy0,ty);
 267             fjz0             = _mm_add_ps(fjz0,tz);
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             r10              = _mm_mul_ps(rsq10,rinv10);
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             qq10             = _mm_mul_ps(iq1,jq0);
 277
 278             /* Calculate table index by multiplying r with table scale and truncate to integer */
 279             rt               = _mm_mul_ps(r10,vftabscale);
 280             vfitab           = _mm_cvttps_epi32(rt);
 281             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 282             vfitab           = _mm_slli_epi32(vfitab,2);
 283
 284             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 285             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 286             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 287             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 288             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 289             _MM_TRANSPOSE4_PS(Y,F,G,H);
 290             Heps             = _mm_mul_ps(vfeps,H);
 291             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 292             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 293             velec            = _mm_mul_ps(qq10,VV);
 294             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 295             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_ps(fscal,dx10);
 304             ty               = _mm_mul_ps(fscal,dy10);
 305             tz               = _mm_mul_ps(fscal,dz10);
 306
 307             /* Update vectorial force */
 308             fix1             = _mm_add_ps(fix1,tx);
 309             fiy1             = _mm_add_ps(fiy1,ty);
 310             fiz1             = _mm_add_ps(fiz1,tz);
 311
 312             fjx0             = _mm_add_ps(fjx0,tx);
 313             fjy0             = _mm_add_ps(fjy0,ty);
 314             fjz0             = _mm_add_ps(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r20              = _mm_mul_ps(rsq20,rinv20);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq20             = _mm_mul_ps(iq2,jq0);
 324
 325             /* Calculate table index by multiplying r with table scale and truncate to integer */
 326             rt               = _mm_mul_ps(r20,vftabscale);
 327             vfitab           = _mm_cvttps_epi32(rt);
 328             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 329             vfitab           = _mm_slli_epi32(vfitab,2);
 330
 331             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 332             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 333             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 334             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 335             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 336             _MM_TRANSPOSE4_PS(Y,F,G,H);
 337             Heps             = _mm_mul_ps(vfeps,H);
 338             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 339             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 340             velec            = _mm_mul_ps(qq20,VV);
 341             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 342             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm_mul_ps(fscal,dx20);
 351             ty               = _mm_mul_ps(fscal,dy20);
 352             tz               = _mm_mul_ps(fscal,dz20);
 353
 354             /* Update vectorial force */
 355             fix2             = _mm_add_ps(fix2,tx);
 356             fiy2             = _mm_add_ps(fiy2,ty);
 357             fiz2             = _mm_add_ps(fiz2,tz);
 358
 359             fjx0             = _mm_add_ps(fjx0,tx);
 360             fjy0             = _mm_add_ps(fjy0,ty);
 361             fjz0             = _mm_add_ps(fjz0,tz);
 362
 363             fjptrA             = f+j_coord_offsetA;
 364             fjptrB             = f+j_coord_offsetB;
 365             fjptrC             = f+j_coord_offsetC;
 366             fjptrD             = f+j_coord_offsetD;
 367
 368             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 369
 370             /* Inner loop uses 129 flops */
 371         }
 372
 373         if(jidx<j_index_end)
 374         {
 375
 376             /* Get j neighbor index, and coordinate index */
 377             jnrlistA         = jjnr[jidx];
 378             jnrlistB         = jjnr[jidx+1];
 379             jnrlistC         = jjnr[jidx+2];
 380             jnrlistD         = jjnr[jidx+3];
 381             /* Sign of each element will be negative for non-real atoms.
 382              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 383              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 384              */
 385             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 386             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 387             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 388             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 389             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 390             j_coord_offsetA  = DIM*jnrA;
 391             j_coord_offsetB  = DIM*jnrB;
 392             j_coord_offsetC  = DIM*jnrC;
 393             j_coord_offsetD  = DIM*jnrD;
 394
 395             /* load j atom coordinates */
 396             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 397                                               x+j_coord_offsetC,x+j_coord_offsetD,
 398                                               &jx0,&jy0,&jz0);
 399
 400             /* Calculate displacement vector */
 401             dx00             = _mm_sub_ps(ix0,jx0);
 402             dy00             = _mm_sub_ps(iy0,jy0);
 403             dz00             = _mm_sub_ps(iz0,jz0);
 404             dx10             = _mm_sub_ps(ix1,jx0);
 405             dy10             = _mm_sub_ps(iy1,jy0);
 406             dz10             = _mm_sub_ps(iz1,jz0);
 407             dx20             = _mm_sub_ps(ix2,jx0);
 408             dy20             = _mm_sub_ps(iy2,jy0);
 409             dz20             = _mm_sub_ps(iz2,jz0);
 410
 411             /* Calculate squared distance and things based on it */
 412             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 413             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 414             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 415
 416             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 417             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 418             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 419
 420             /* Load parameters for j particles */
 421             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 422                                                               charge+jnrC+0,charge+jnrD+0);
 423
 424             fjx0             = _mm_setzero_ps();
 425             fjy0             = _mm_setzero_ps();
 426             fjz0             = _mm_setzero_ps();
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             r00              = _mm_mul_ps(rsq00,rinv00);
 433             r00              = _mm_andnot_ps(dummy_mask,r00);
 434
 435             /* Compute parameters for interactions between i and j atoms */
 436             qq00             = _mm_mul_ps(iq0,jq0);
 437
 438             /* Calculate table index by multiplying r with table scale and truncate to integer */
 439             rt               = _mm_mul_ps(r00,vftabscale);
 440             vfitab           = _mm_cvttps_epi32(rt);
 441             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 442             vfitab           = _mm_slli_epi32(vfitab,2);
 443
 444             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 445             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 446             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 447             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 448             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 449             _MM_TRANSPOSE4_PS(Y,F,G,H);
 450             Heps             = _mm_mul_ps(vfeps,H);
 451             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 452             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 453             velec            = _mm_mul_ps(qq00,VV);
 454             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 455             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm_andnot_ps(dummy_mask,velec);
 459             velecsum         = _mm_add_ps(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 464
 465             /* Calculate temporary vectorial force */
 466             tx               = _mm_mul_ps(fscal,dx00);
 467             ty               = _mm_mul_ps(fscal,dy00);
 468             tz               = _mm_mul_ps(fscal,dz00);
 469
 470             /* Update vectorial force */
 471             fix0             = _mm_add_ps(fix0,tx);
 472             fiy0             = _mm_add_ps(fiy0,ty);
 473             fiz0             = _mm_add_ps(fiz0,tz);
 474
 475             fjx0             = _mm_add_ps(fjx0,tx);
 476             fjy0             = _mm_add_ps(fjy0,ty);
 477             fjz0             = _mm_add_ps(fjz0,tz);
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             r10              = _mm_mul_ps(rsq10,rinv10);
 484             r10              = _mm_andnot_ps(dummy_mask,r10);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq10             = _mm_mul_ps(iq1,jq0);
 488
 489             /* Calculate table index by multiplying r with table scale and truncate to integer */
 490             rt               = _mm_mul_ps(r10,vftabscale);
 491             vfitab           = _mm_cvttps_epi32(rt);
 492             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 493             vfitab           = _mm_slli_epi32(vfitab,2);
 494
 495             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 496             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 497             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 498             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 499             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 500             _MM_TRANSPOSE4_PS(Y,F,G,H);
 501             Heps             = _mm_mul_ps(vfeps,H);
 502             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 503             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 504             velec            = _mm_mul_ps(qq10,VV);
 505             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 506             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velec            = _mm_andnot_ps(dummy_mask,velec);
 510             velecsum         = _mm_add_ps(velecsum,velec);
 511
 512             fscal            = felec;
 513
 514             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 515
 516             /* Calculate temporary vectorial force */
 517             tx               = _mm_mul_ps(fscal,dx10);
 518             ty               = _mm_mul_ps(fscal,dy10);
 519             tz               = _mm_mul_ps(fscal,dz10);
 520
 521             /* Update vectorial force */
 522             fix1             = _mm_add_ps(fix1,tx);
 523             fiy1             = _mm_add_ps(fiy1,ty);
 524             fiz1             = _mm_add_ps(fiz1,tz);
 525
 526             fjx0             = _mm_add_ps(fjx0,tx);
 527             fjy0             = _mm_add_ps(fjy0,ty);
 528             fjz0             = _mm_add_ps(fjz0,tz);
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r20              = _mm_mul_ps(rsq20,rinv20);
 535             r20              = _mm_andnot_ps(dummy_mask,r20);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq20             = _mm_mul_ps(iq2,jq0);
 539
 540             /* Calculate table index by multiplying r with table scale and truncate to integer */
 541             rt               = _mm_mul_ps(r20,vftabscale);
 542             vfitab           = _mm_cvttps_epi32(rt);
 543             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 544             vfitab           = _mm_slli_epi32(vfitab,2);
 545
 546             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 547             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 548             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 549             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 550             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 551             _MM_TRANSPOSE4_PS(Y,F,G,H);
 552             Heps             = _mm_mul_ps(vfeps,H);
 553             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 554             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 555             velec            = _mm_mul_ps(qq20,VV);
 556             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 557             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 558
 559             /* Update potential sum for this i atom from the interaction with this j atom. */
 560             velec            = _mm_andnot_ps(dummy_mask,velec);
 561             velecsum         = _mm_add_ps(velecsum,velec);
 562
 563             fscal            = felec;
 564
 565             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 566
 567             /* Calculate temporary vectorial force */
 568             tx               = _mm_mul_ps(fscal,dx20);
 569             ty               = _mm_mul_ps(fscal,dy20);
 570             tz               = _mm_mul_ps(fscal,dz20);
 571
 572             /* Update vectorial force */
 573             fix2             = _mm_add_ps(fix2,tx);
 574             fiy2             = _mm_add_ps(fiy2,ty);
 575             fiz2             = _mm_add_ps(fiz2,tz);
 576
 577             fjx0             = _mm_add_ps(fjx0,tx);
 578             fjy0             = _mm_add_ps(fjy0,ty);
 579             fjz0             = _mm_add_ps(fjz0,tz);
 580
 581             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 582             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 583             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 584             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 585
 586             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 587
 588             /* Inner loop uses 132 flops */
 589         }
 590
 591         /* End of innermost loop */
 592
 593         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 594                                               f+i_coord_offset,fshift+i_shift_offset);
 595
 596         ggid                        = gid[iidx];
 597         /* Update potential energies */
 598         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 599
 600         /* Increment number of inner iterations */
 601         inneriter                  += j_index_end - j_index_start;
 602
 603         /* Outer loop uses 19 flops */
 604     }
 605
 606     /* Increment number of outer iterations */
 607     outeriter        += nri;
 608
 609     /* Update outer/inner flops */
 610
 611     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*132);
 612 }
 613 /*
 614  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_sse4_1_single
 615  * Electrostatics interaction: CubicSplineTable
 616  * VdW interaction:            None
 617  * Geometry:                   Water3-Particle
 618  * Calculate force/pot:        Force
 619  */
 620 void
 621 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_sse4_1_single
 622                     (t_nblist                    * gmx_restrict       nlist,
 623                      rvec                        * gmx_restrict          xx,
 624                      rvec                        * gmx_restrict          ff,
 625                      t_forcerec                  * gmx_restrict          fr,
 626                      t_mdatoms                   * gmx_restrict     mdatoms,
 627                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 628                      t_nrnb                      * gmx_restrict        nrnb)
 629 {
 630     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 631      * just 0 for non-waters.
 632      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 633      * jnr indices corresponding to data put in the four positions in the SIMD register.
 634      */
 635     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 636     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 637     int              jnrA,jnrB,jnrC,jnrD;
 638     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 639     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 640     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 641     real             rcutoff_scalar;
 642     real             *shiftvec,*fshift,*x,*f;
 643     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 644     real             scratch[4*DIM];
 645     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 646     int              vdwioffset0;
 647     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 648     int              vdwioffset1;
 649     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 650     int              vdwioffset2;
 651     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 652     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 653     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 654     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 655     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 656     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 657     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 658     real             *charge;
 659     __m128i          vfitab;
 660     __m128i          ifour       = _mm_set1_epi32(4);
 661     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 662     real             *vftab;
 663     __m128           dummy_mask,cutoff_mask;
 664     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 665     __m128           one     = _mm_set1_ps(1.0);
 666     __m128           two     = _mm_set1_ps(2.0);
 667     x                = xx[0];
 668     f                = ff[0];
 669
 670     nri              = nlist->nri;
 671     iinr             = nlist->iinr;
 672     jindex           = nlist->jindex;
 673     jjnr             = nlist->jjnr;
 674     shiftidx         = nlist->shift;
 675     gid              = nlist->gid;
 676     shiftvec         = fr->shift_vec[0];
 677     fshift           = fr->fshift[0];
 678     facel            = _mm_set1_ps(fr->epsfac);
 679     charge           = mdatoms->chargeA;
 680
 681     vftab            = kernel_data->table_elec->data;
 682     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 683
 684     /* Setup water-specific parameters */
 685     inr              = nlist->iinr[0];
 686     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 687     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 688     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 689
 690     /* Avoid stupid compiler warnings */
 691     jnrA = jnrB = jnrC = jnrD = 0;
 692     j_coord_offsetA = 0;
 693     j_coord_offsetB = 0;
 694     j_coord_offsetC = 0;
 695     j_coord_offsetD = 0;
 696
 697     outeriter        = 0;
 698     inneriter        = 0;
 699
 700     for(iidx=0;iidx<4*DIM;iidx++)
 701     {
 702         scratch[iidx] = 0.0;
 703     }
 704
 705     /* Start outer loop over neighborlists */
 706     for(iidx=0; iidx<nri; iidx++)
 707     {
 708         /* Load shift vector for this list */
 709         i_shift_offset   = DIM*shiftidx[iidx];
 710
 711         /* Load limits for loop over neighbors */
 712         j_index_start    = jindex[iidx];
 713         j_index_end      = jindex[iidx+1];
 714
 715         /* Get outer coordinate index */
 716         inr              = iinr[iidx];
 717         i_coord_offset   = DIM*inr;
 718
 719         /* Load i particle coords and add shift vector */
 720         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 721                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 722
 723         fix0             = _mm_setzero_ps();
 724         fiy0             = _mm_setzero_ps();
 725         fiz0             = _mm_setzero_ps();
 726         fix1             = _mm_setzero_ps();
 727         fiy1             = _mm_setzero_ps();
 728         fiz1             = _mm_setzero_ps();
 729         fix2             = _mm_setzero_ps();
 730         fiy2             = _mm_setzero_ps();
 731         fiz2             = _mm_setzero_ps();
 732
 733         /* Start inner kernel loop */
 734         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 735         {
 736
 737             /* Get j neighbor index, and coordinate index */
 738             jnrA             = jjnr[jidx];
 739             jnrB             = jjnr[jidx+1];
 740             jnrC             = jjnr[jidx+2];
 741             jnrD             = jjnr[jidx+3];
 742             j_coord_offsetA  = DIM*jnrA;
 743             j_coord_offsetB  = DIM*jnrB;
 744             j_coord_offsetC  = DIM*jnrC;
 745             j_coord_offsetD  = DIM*jnrD;
 746
 747             /* load j atom coordinates */
 748             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 749                                               x+j_coord_offsetC,x+j_coord_offsetD,
 750                                               &jx0,&jy0,&jz0);
 751
 752             /* Calculate displacement vector */
 753             dx00             = _mm_sub_ps(ix0,jx0);
 754             dy00             = _mm_sub_ps(iy0,jy0);
 755             dz00             = _mm_sub_ps(iz0,jz0);
 756             dx10             = _mm_sub_ps(ix1,jx0);
 757             dy10             = _mm_sub_ps(iy1,jy0);
 758             dz10             = _mm_sub_ps(iz1,jz0);
 759             dx20             = _mm_sub_ps(ix2,jx0);
 760             dy20             = _mm_sub_ps(iy2,jy0);
 761             dz20             = _mm_sub_ps(iz2,jz0);
 762
 763             /* Calculate squared distance and things based on it */
 764             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 765             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 766             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 767
 768             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 769             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 770             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 771
 772             /* Load parameters for j particles */
 773             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 774                                                               charge+jnrC+0,charge+jnrD+0);
 775
 776             fjx0             = _mm_setzero_ps();
 777             fjy0             = _mm_setzero_ps();
 778             fjz0             = _mm_setzero_ps();
 779
 780             /**************************
 781              * CALCULATE INTERACTIONS *
 782              **************************/
 783
 784             r00              = _mm_mul_ps(rsq00,rinv00);
 785
 786             /* Compute parameters for interactions between i and j atoms */
 787             qq00             = _mm_mul_ps(iq0,jq0);
 788
 789             /* Calculate table index by multiplying r with table scale and truncate to integer */
 790             rt               = _mm_mul_ps(r00,vftabscale);
 791             vfitab           = _mm_cvttps_epi32(rt);
 792             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 793             vfitab           = _mm_slli_epi32(vfitab,2);
 794
 795             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 796             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 797             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 798             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 799             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 800             _MM_TRANSPOSE4_PS(Y,F,G,H);
 801             Heps             = _mm_mul_ps(vfeps,H);
 802             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 803             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 804             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 805
 806             fscal            = felec;
 807
 808             /* Calculate temporary vectorial force */
 809             tx               = _mm_mul_ps(fscal,dx00);
 810             ty               = _mm_mul_ps(fscal,dy00);
 811             tz               = _mm_mul_ps(fscal,dz00);
 812
 813             /* Update vectorial force */
 814             fix0             = _mm_add_ps(fix0,tx);
 815             fiy0             = _mm_add_ps(fiy0,ty);
 816             fiz0             = _mm_add_ps(fiz0,tz);
 817
 818             fjx0             = _mm_add_ps(fjx0,tx);
 819             fjy0             = _mm_add_ps(fjy0,ty);
 820             fjz0             = _mm_add_ps(fjz0,tz);
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             r10              = _mm_mul_ps(rsq10,rinv10);
 827
 828             /* Compute parameters for interactions between i and j atoms */
 829             qq10             = _mm_mul_ps(iq1,jq0);
 830
 831             /* Calculate table index by multiplying r with table scale and truncate to integer */
 832             rt               = _mm_mul_ps(r10,vftabscale);
 833             vfitab           = _mm_cvttps_epi32(rt);
 834             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 835             vfitab           = _mm_slli_epi32(vfitab,2);
 836
 837             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 838             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 839             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 840             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 841             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 842             _MM_TRANSPOSE4_PS(Y,F,G,H);
 843             Heps             = _mm_mul_ps(vfeps,H);
 844             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 845             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 846             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 847
 848             fscal            = felec;
 849
 850             /* Calculate temporary vectorial force */
 851             tx               = _mm_mul_ps(fscal,dx10);
 852             ty               = _mm_mul_ps(fscal,dy10);
 853             tz               = _mm_mul_ps(fscal,dz10);
 854
 855             /* Update vectorial force */
 856             fix1             = _mm_add_ps(fix1,tx);
 857             fiy1             = _mm_add_ps(fiy1,ty);
 858             fiz1             = _mm_add_ps(fiz1,tz);
 859
 860             fjx0             = _mm_add_ps(fjx0,tx);
 861             fjy0             = _mm_add_ps(fjy0,ty);
 862             fjz0             = _mm_add_ps(fjz0,tz);
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             r20              = _mm_mul_ps(rsq20,rinv20);
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq20             = _mm_mul_ps(iq2,jq0);
 872
 873             /* Calculate table index by multiplying r with table scale and truncate to integer */
 874             rt               = _mm_mul_ps(r20,vftabscale);
 875             vfitab           = _mm_cvttps_epi32(rt);
 876             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 877             vfitab           = _mm_slli_epi32(vfitab,2);
 878
 879             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 880             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 881             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 882             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 883             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 884             _MM_TRANSPOSE4_PS(Y,F,G,H);
 885             Heps             = _mm_mul_ps(vfeps,H);
 886             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 887             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 888             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 889
 890             fscal            = felec;
 891
 892             /* Calculate temporary vectorial force */
 893             tx               = _mm_mul_ps(fscal,dx20);
 894             ty               = _mm_mul_ps(fscal,dy20);
 895             tz               = _mm_mul_ps(fscal,dz20);
 896
 897             /* Update vectorial force */
 898             fix2             = _mm_add_ps(fix2,tx);
 899             fiy2             = _mm_add_ps(fiy2,ty);
 900             fiz2             = _mm_add_ps(fiz2,tz);
 901
 902             fjx0             = _mm_add_ps(fjx0,tx);
 903             fjy0             = _mm_add_ps(fjy0,ty);
 904             fjz0             = _mm_add_ps(fjz0,tz);
 905
 906             fjptrA             = f+j_coord_offsetA;
 907             fjptrB             = f+j_coord_offsetB;
 908             fjptrC             = f+j_coord_offsetC;
 909             fjptrD             = f+j_coord_offsetD;
 910
 911             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 912
 913             /* Inner loop uses 117 flops */
 914         }
 915
 916         if(jidx<j_index_end)
 917         {
 918
 919             /* Get j neighbor index, and coordinate index */
 920             jnrlistA         = jjnr[jidx];
 921             jnrlistB         = jjnr[jidx+1];
 922             jnrlistC         = jjnr[jidx+2];
 923             jnrlistD         = jjnr[jidx+3];
 924             /* Sign of each element will be negative for non-real atoms.
 925              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 926              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 927              */
 928             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 929             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 930             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 931             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 932             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 933             j_coord_offsetA  = DIM*jnrA;
 934             j_coord_offsetB  = DIM*jnrB;
 935             j_coord_offsetC  = DIM*jnrC;
 936             j_coord_offsetD  = DIM*jnrD;
 937
 938             /* load j atom coordinates */
 939             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 940                                               x+j_coord_offsetC,x+j_coord_offsetD,
 941                                               &jx0,&jy0,&jz0);
 942
 943             /* Calculate displacement vector */
 944             dx00             = _mm_sub_ps(ix0,jx0);
 945             dy00             = _mm_sub_ps(iy0,jy0);
 946             dz00             = _mm_sub_ps(iz0,jz0);
 947             dx10             = _mm_sub_ps(ix1,jx0);
 948             dy10             = _mm_sub_ps(iy1,jy0);
 949             dz10             = _mm_sub_ps(iz1,jz0);
 950             dx20             = _mm_sub_ps(ix2,jx0);
 951             dy20             = _mm_sub_ps(iy2,jy0);
 952             dz20             = _mm_sub_ps(iz2,jz0);
 953
 954             /* Calculate squared distance and things based on it */
 955             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 956             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 957             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 958
 959             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 960             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 961             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 962
 963             /* Load parameters for j particles */
 964             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 965                                                               charge+jnrC+0,charge+jnrD+0);
 966
 967             fjx0             = _mm_setzero_ps();
 968             fjy0             = _mm_setzero_ps();
 969             fjz0             = _mm_setzero_ps();
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             r00              = _mm_mul_ps(rsq00,rinv00);
 976             r00              = _mm_andnot_ps(dummy_mask,r00);
 977
 978             /* Compute parameters for interactions between i and j atoms */
 979             qq00             = _mm_mul_ps(iq0,jq0);
 980
 981             /* Calculate table index by multiplying r with table scale and truncate to integer */
 982             rt               = _mm_mul_ps(r00,vftabscale);
 983             vfitab           = _mm_cvttps_epi32(rt);
 984             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 985             vfitab           = _mm_slli_epi32(vfitab,2);
 986
 987             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 988             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 989             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 990             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 991             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 992             _MM_TRANSPOSE4_PS(Y,F,G,H);
 993             Heps             = _mm_mul_ps(vfeps,H);
 994             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 995             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 996             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 997
 998             fscal            = felec;
 999
1000             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1001
1002             /* Calculate temporary vectorial force */
1003             tx               = _mm_mul_ps(fscal,dx00);
1004             ty               = _mm_mul_ps(fscal,dy00);
1005             tz               = _mm_mul_ps(fscal,dz00);
1006
1007             /* Update vectorial force */
1008             fix0             = _mm_add_ps(fix0,tx);
1009             fiy0             = _mm_add_ps(fiy0,ty);
1010             fiz0             = _mm_add_ps(fiz0,tz);
1011
1012             fjx0             = _mm_add_ps(fjx0,tx);
1013             fjy0             = _mm_add_ps(fjy0,ty);
1014             fjz0             = _mm_add_ps(fjz0,tz);
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             r10              = _mm_mul_ps(rsq10,rinv10);
1021             r10              = _mm_andnot_ps(dummy_mask,r10);
1022
1023             /* Compute parameters for interactions between i and j atoms */
1024             qq10             = _mm_mul_ps(iq1,jq0);
1025
1026             /* Calculate table index by multiplying r with table scale and truncate to integer */
1027             rt               = _mm_mul_ps(r10,vftabscale);
1028             vfitab           = _mm_cvttps_epi32(rt);
1029             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1030             vfitab           = _mm_slli_epi32(vfitab,2);
1031
1032             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1033             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1034             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1035             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1036             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1037             _MM_TRANSPOSE4_PS(Y,F,G,H);
1038             Heps             = _mm_mul_ps(vfeps,H);
1039             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1040             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1041             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1042
1043             fscal            = felec;
1044
1045             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm_mul_ps(fscal,dx10);
1049             ty               = _mm_mul_ps(fscal,dy10);
1050             tz               = _mm_mul_ps(fscal,dz10);
1051
1052             /* Update vectorial force */
1053             fix1             = _mm_add_ps(fix1,tx);
1054             fiy1             = _mm_add_ps(fiy1,ty);
1055             fiz1             = _mm_add_ps(fiz1,tz);
1056
1057             fjx0             = _mm_add_ps(fjx0,tx);
1058             fjy0             = _mm_add_ps(fjy0,ty);
1059             fjz0             = _mm_add_ps(fjz0,tz);
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             r20              = _mm_mul_ps(rsq20,rinv20);
1066             r20              = _mm_andnot_ps(dummy_mask,r20);
1067
1068             /* Compute parameters for interactions between i and j atoms */
1069             qq20             = _mm_mul_ps(iq2,jq0);
1070
1071             /* Calculate table index by multiplying r with table scale and truncate to integer */
1072             rt               = _mm_mul_ps(r20,vftabscale);
1073             vfitab           = _mm_cvttps_epi32(rt);
1074             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1075             vfitab           = _mm_slli_epi32(vfitab,2);
1076
1077             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1078             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1079             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1080             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1081             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1082             _MM_TRANSPOSE4_PS(Y,F,G,H);
1083             Heps             = _mm_mul_ps(vfeps,H);
1084             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1085             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1086             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1087
1088             fscal            = felec;
1089
1090             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm_mul_ps(fscal,dx20);
1094             ty               = _mm_mul_ps(fscal,dy20);
1095             tz               = _mm_mul_ps(fscal,dz20);
1096
1097             /* Update vectorial force */
1098             fix2             = _mm_add_ps(fix2,tx);
1099             fiy2             = _mm_add_ps(fiy2,ty);
1100             fiz2             = _mm_add_ps(fiz2,tz);
1101
1102             fjx0             = _mm_add_ps(fjx0,tx);
1103             fjy0             = _mm_add_ps(fjy0,ty);
1104             fjz0             = _mm_add_ps(fjz0,tz);
1105
1106             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1107             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1108             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1109             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1110
1111             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1112
1113             /* Inner loop uses 120 flops */
1114         }
1115
1116         /* End of innermost loop */
1117
1118         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1119                                               f+i_coord_offset,fshift+i_shift_offset);
1120
1121         /* Increment number of inner iterations */
1122         inneriter                  += j_index_end - j_index_start;
1123
1124         /* Outer loop uses 18 flops */
1125     }
1126
1127     /* Increment number of outer iterations */
1128     outeriter        += nri;
1129
1130     /* Update outer/inner flops */
1131
1132     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*120);
1133 }