src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwNone_GeomW3P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3P1_VF_sse2_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            None
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_ps(fr->ic->epsfac);
 115     charge           = mdatoms->chargeA;
 116
 117     vftab            = kernel_data->table_elec->data;
 118     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 123     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 124     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162         fix1             = _mm_setzero_ps();
 163         fiy1             = _mm_setzero_ps();
 164         fiz1             = _mm_setzero_ps();
 165         fix2             = _mm_setzero_ps();
 166         fiy2             = _mm_setzero_ps();
 167         fiz2             = _mm_setzero_ps();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_ps();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183             j_coord_offsetC  = DIM*jnrC;
 184             j_coord_offsetD  = DIM*jnrD;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               x+j_coord_offsetC,x+j_coord_offsetD,
 189                                               &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm_sub_ps(ix0,jx0);
 193             dy00             = _mm_sub_ps(iy0,jy0);
 194             dz00             = _mm_sub_ps(iz0,jz0);
 195             dx10             = _mm_sub_ps(ix1,jx0);
 196             dy10             = _mm_sub_ps(iy1,jy0);
 197             dz10             = _mm_sub_ps(iz1,jz0);
 198             dx20             = _mm_sub_ps(ix2,jx0);
 199             dy20             = _mm_sub_ps(iy2,jy0);
 200             dz20             = _mm_sub_ps(iz2,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 206
 207             rinv00           = sse2_invsqrt_f(rsq00);
 208             rinv10           = sse2_invsqrt_f(rsq10);
 209             rinv20           = sse2_invsqrt_f(rsq20);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                               charge+jnrC+0,charge+jnrD+0);
 214
 215             fjx0             = _mm_setzero_ps();
 216             fjy0             = _mm_setzero_ps();
 217             fjz0             = _mm_setzero_ps();
 218
 219             /**************************
 220              * CALCULATE INTERACTIONS *
 221              **************************/
 222
 223             r00              = _mm_mul_ps(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_ps(iq0,jq0);
 227
 228             /* Calculate table index by multiplying r with table scale and truncate to integer */
 229             rt               = _mm_mul_ps(r00,vftabscale);
 230             vfitab           = _mm_cvttps_epi32(rt);
 231             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 232             vfitab           = _mm_slli_epi32(vfitab,2);
 233
 234             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 235             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 236             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 237             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 238             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 239             _MM_TRANSPOSE4_PS(Y,F,G,H);
 240             Heps             = _mm_mul_ps(vfeps,H);
 241             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 242             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 243             velec            = _mm_mul_ps(qq00,VV);
 244             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 245             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velecsum         = _mm_add_ps(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm_mul_ps(fscal,dx00);
 254             ty               = _mm_mul_ps(fscal,dy00);
 255             tz               = _mm_mul_ps(fscal,dz00);
 256
 257             /* Update vectorial force */
 258             fix0             = _mm_add_ps(fix0,tx);
 259             fiy0             = _mm_add_ps(fiy0,ty);
 260             fiz0             = _mm_add_ps(fiz0,tz);
 261
 262             fjx0             = _mm_add_ps(fjx0,tx);
 263             fjy0             = _mm_add_ps(fjy0,ty);
 264             fjz0             = _mm_add_ps(fjz0,tz);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             r10              = _mm_mul_ps(rsq10,rinv10);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq10             = _mm_mul_ps(iq1,jq0);
 274
 275             /* Calculate table index by multiplying r with table scale and truncate to integer */
 276             rt               = _mm_mul_ps(r10,vftabscale);
 277             vfitab           = _mm_cvttps_epi32(rt);
 278             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 279             vfitab           = _mm_slli_epi32(vfitab,2);
 280
 281             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 282             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 283             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 284             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 285             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 286             _MM_TRANSPOSE4_PS(Y,F,G,H);
 287             Heps             = _mm_mul_ps(vfeps,H);
 288             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 289             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 290             velec            = _mm_mul_ps(qq10,VV);
 291             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 292             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm_add_ps(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm_mul_ps(fscal,dx10);
 301             ty               = _mm_mul_ps(fscal,dy10);
 302             tz               = _mm_mul_ps(fscal,dz10);
 303
 304             /* Update vectorial force */
 305             fix1             = _mm_add_ps(fix1,tx);
 306             fiy1             = _mm_add_ps(fiy1,ty);
 307             fiz1             = _mm_add_ps(fiz1,tz);
 308
 309             fjx0             = _mm_add_ps(fjx0,tx);
 310             fjy0             = _mm_add_ps(fjy0,ty);
 311             fjz0             = _mm_add_ps(fjz0,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             r20              = _mm_mul_ps(rsq20,rinv20);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq20             = _mm_mul_ps(iq2,jq0);
 321
 322             /* Calculate table index by multiplying r with table scale and truncate to integer */
 323             rt               = _mm_mul_ps(r20,vftabscale);
 324             vfitab           = _mm_cvttps_epi32(rt);
 325             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 326             vfitab           = _mm_slli_epi32(vfitab,2);
 327
 328             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 329             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 330             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 331             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 332             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 333             _MM_TRANSPOSE4_PS(Y,F,G,H);
 334             Heps             = _mm_mul_ps(vfeps,H);
 335             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 336             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 337             velec            = _mm_mul_ps(qq20,VV);
 338             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 339             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_ps(fscal,dx20);
 348             ty               = _mm_mul_ps(fscal,dy20);
 349             tz               = _mm_mul_ps(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm_add_ps(fix2,tx);
 353             fiy2             = _mm_add_ps(fiy2,ty);
 354             fiz2             = _mm_add_ps(fiz2,tz);
 355
 356             fjx0             = _mm_add_ps(fjx0,tx);
 357             fjy0             = _mm_add_ps(fjy0,ty);
 358             fjz0             = _mm_add_ps(fjz0,tz);
 359
 360             fjptrA             = f+j_coord_offsetA;
 361             fjptrB             = f+j_coord_offsetB;
 362             fjptrC             = f+j_coord_offsetC;
 363             fjptrD             = f+j_coord_offsetD;
 364
 365             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 366
 367             /* Inner loop uses 129 flops */
 368         }
 369
 370         if(jidx<j_index_end)
 371         {
 372
 373             /* Get j neighbor index, and coordinate index */
 374             jnrlistA         = jjnr[jidx];
 375             jnrlistB         = jjnr[jidx+1];
 376             jnrlistC         = jjnr[jidx+2];
 377             jnrlistD         = jjnr[jidx+3];
 378             /* Sign of each element will be negative for non-real atoms.
 379              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 380              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 381              */
 382             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 383             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 384             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 385             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 386             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 387             j_coord_offsetA  = DIM*jnrA;
 388             j_coord_offsetB  = DIM*jnrB;
 389             j_coord_offsetC  = DIM*jnrC;
 390             j_coord_offsetD  = DIM*jnrD;
 391
 392             /* load j atom coordinates */
 393             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 394                                               x+j_coord_offsetC,x+j_coord_offsetD,
 395                                               &jx0,&jy0,&jz0);
 396
 397             /* Calculate displacement vector */
 398             dx00             = _mm_sub_ps(ix0,jx0);
 399             dy00             = _mm_sub_ps(iy0,jy0);
 400             dz00             = _mm_sub_ps(iz0,jz0);
 401             dx10             = _mm_sub_ps(ix1,jx0);
 402             dy10             = _mm_sub_ps(iy1,jy0);
 403             dz10             = _mm_sub_ps(iz1,jz0);
 404             dx20             = _mm_sub_ps(ix2,jx0);
 405             dy20             = _mm_sub_ps(iy2,jy0);
 406             dz20             = _mm_sub_ps(iz2,jz0);
 407
 408             /* Calculate squared distance and things based on it */
 409             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 410             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 411             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 412
 413             rinv00           = sse2_invsqrt_f(rsq00);
 414             rinv10           = sse2_invsqrt_f(rsq10);
 415             rinv20           = sse2_invsqrt_f(rsq20);
 416
 417             /* Load parameters for j particles */
 418             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 419                                                               charge+jnrC+0,charge+jnrD+0);
 420
 421             fjx0             = _mm_setzero_ps();
 422             fjy0             = _mm_setzero_ps();
 423             fjz0             = _mm_setzero_ps();
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             r00              = _mm_mul_ps(rsq00,rinv00);
 430             r00              = _mm_andnot_ps(dummy_mask,r00);
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             qq00             = _mm_mul_ps(iq0,jq0);
 434
 435             /* Calculate table index by multiplying r with table scale and truncate to integer */
 436             rt               = _mm_mul_ps(r00,vftabscale);
 437             vfitab           = _mm_cvttps_epi32(rt);
 438             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 439             vfitab           = _mm_slli_epi32(vfitab,2);
 440
 441             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 442             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 443             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 444             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 445             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 446             _MM_TRANSPOSE4_PS(Y,F,G,H);
 447             Heps             = _mm_mul_ps(vfeps,H);
 448             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 449             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 450             velec            = _mm_mul_ps(qq00,VV);
 451             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 452             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velec            = _mm_andnot_ps(dummy_mask,velec);
 456             velecsum         = _mm_add_ps(velecsum,velec);
 457
 458             fscal            = felec;
 459
 460             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 461
 462             /* Calculate temporary vectorial force */
 463             tx               = _mm_mul_ps(fscal,dx00);
 464             ty               = _mm_mul_ps(fscal,dy00);
 465             tz               = _mm_mul_ps(fscal,dz00);
 466
 467             /* Update vectorial force */
 468             fix0             = _mm_add_ps(fix0,tx);
 469             fiy0             = _mm_add_ps(fiy0,ty);
 470             fiz0             = _mm_add_ps(fiz0,tz);
 471
 472             fjx0             = _mm_add_ps(fjx0,tx);
 473             fjy0             = _mm_add_ps(fjy0,ty);
 474             fjz0             = _mm_add_ps(fjz0,tz);
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             r10              = _mm_mul_ps(rsq10,rinv10);
 481             r10              = _mm_andnot_ps(dummy_mask,r10);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq10             = _mm_mul_ps(iq1,jq0);
 485
 486             /* Calculate table index by multiplying r with table scale and truncate to integer */
 487             rt               = _mm_mul_ps(r10,vftabscale);
 488             vfitab           = _mm_cvttps_epi32(rt);
 489             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 490             vfitab           = _mm_slli_epi32(vfitab,2);
 491
 492             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 493             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 494             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 495             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 496             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 497             _MM_TRANSPOSE4_PS(Y,F,G,H);
 498             Heps             = _mm_mul_ps(vfeps,H);
 499             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 500             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 501             velec            = _mm_mul_ps(qq10,VV);
 502             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 503             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm_andnot_ps(dummy_mask,velec);
 507             velecsum         = _mm_add_ps(velecsum,velec);
 508
 509             fscal            = felec;
 510
 511             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 512
 513             /* Calculate temporary vectorial force */
 514             tx               = _mm_mul_ps(fscal,dx10);
 515             ty               = _mm_mul_ps(fscal,dy10);
 516             tz               = _mm_mul_ps(fscal,dz10);
 517
 518             /* Update vectorial force */
 519             fix1             = _mm_add_ps(fix1,tx);
 520             fiy1             = _mm_add_ps(fiy1,ty);
 521             fiz1             = _mm_add_ps(fiz1,tz);
 522
 523             fjx0             = _mm_add_ps(fjx0,tx);
 524             fjy0             = _mm_add_ps(fjy0,ty);
 525             fjz0             = _mm_add_ps(fjz0,tz);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             r20              = _mm_mul_ps(rsq20,rinv20);
 532             r20              = _mm_andnot_ps(dummy_mask,r20);
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq20             = _mm_mul_ps(iq2,jq0);
 536
 537             /* Calculate table index by multiplying r with table scale and truncate to integer */
 538             rt               = _mm_mul_ps(r20,vftabscale);
 539             vfitab           = _mm_cvttps_epi32(rt);
 540             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 541             vfitab           = _mm_slli_epi32(vfitab,2);
 542
 543             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 544             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 545             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 546             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 547             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 548             _MM_TRANSPOSE4_PS(Y,F,G,H);
 549             Heps             = _mm_mul_ps(vfeps,H);
 550             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 551             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 552             velec            = _mm_mul_ps(qq20,VV);
 553             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 554             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_andnot_ps(dummy_mask,velec);
 558             velecsum         = _mm_add_ps(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_ps(fscal,dx20);
 566             ty               = _mm_mul_ps(fscal,dy20);
 567             tz               = _mm_mul_ps(fscal,dz20);
 568
 569             /* Update vectorial force */
 570             fix2             = _mm_add_ps(fix2,tx);
 571             fiy2             = _mm_add_ps(fiy2,ty);
 572             fiz2             = _mm_add_ps(fiz2,tz);
 573
 574             fjx0             = _mm_add_ps(fjx0,tx);
 575             fjy0             = _mm_add_ps(fjy0,ty);
 576             fjz0             = _mm_add_ps(fjz0,tz);
 577
 578             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 579             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 580             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 581             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 582
 583             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 584
 585             /* Inner loop uses 132 flops */
 586         }
 587
 588         /* End of innermost loop */
 589
 590         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 591                                               f+i_coord_offset,fshift+i_shift_offset);
 592
 593         ggid                        = gid[iidx];
 594         /* Update potential energies */
 595         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 596
 597         /* Increment number of inner iterations */
 598         inneriter                  += j_index_end - j_index_start;
 599
 600         /* Outer loop uses 19 flops */
 601     }
 602
 603     /* Increment number of outer iterations */
 604     outeriter        += nri;
 605
 606     /* Update outer/inner flops */
 607
 608     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*132);
 609 }
 610 /*
 611  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_sse2_single
 612  * Electrostatics interaction: CubicSplineTable
 613  * VdW interaction:            None
 614  * Geometry:                   Water3-Particle
 615  * Calculate force/pot:        Force
 616  */
 617 void
 618 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_sse2_single
 619                     (t_nblist                    * gmx_restrict       nlist,
 620                      rvec                        * gmx_restrict          xx,
 621                      rvec                        * gmx_restrict          ff,
 622                      struct t_forcerec           * gmx_restrict          fr,
 623                      t_mdatoms                   * gmx_restrict     mdatoms,
 624                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 625                      t_nrnb                      * gmx_restrict        nrnb)
 626 {
 627     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 628      * just 0 for non-waters.
 629      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 630      * jnr indices corresponding to data put in the four positions in the SIMD register.
 631      */
 632     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 633     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 634     int              jnrA,jnrB,jnrC,jnrD;
 635     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 636     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 637     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 638     real             rcutoff_scalar;
 639     real             *shiftvec,*fshift,*x,*f;
 640     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 641     real             scratch[4*DIM];
 642     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 643     int              vdwioffset0;
 644     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 645     int              vdwioffset1;
 646     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 647     int              vdwioffset2;
 648     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 649     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 650     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 651     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 652     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 653     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 654     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 655     real             *charge;
 656     __m128i          vfitab;
 657     __m128i          ifour       = _mm_set1_epi32(4);
 658     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 659     real             *vftab;
 660     __m128           dummy_mask,cutoff_mask;
 661     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 662     __m128           one     = _mm_set1_ps(1.0);
 663     __m128           two     = _mm_set1_ps(2.0);
 664     x                = xx[0];
 665     f                = ff[0];
 666
 667     nri              = nlist->nri;
 668     iinr             = nlist->iinr;
 669     jindex           = nlist->jindex;
 670     jjnr             = nlist->jjnr;
 671     shiftidx         = nlist->shift;
 672     gid              = nlist->gid;
 673     shiftvec         = fr->shift_vec[0];
 674     fshift           = fr->fshift[0];
 675     facel            = _mm_set1_ps(fr->ic->epsfac);
 676     charge           = mdatoms->chargeA;
 677
 678     vftab            = kernel_data->table_elec->data;
 679     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 680
 681     /* Setup water-specific parameters */
 682     inr              = nlist->iinr[0];
 683     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 684     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 685     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 686
 687     /* Avoid stupid compiler warnings */
 688     jnrA = jnrB = jnrC = jnrD = 0;
 689     j_coord_offsetA = 0;
 690     j_coord_offsetB = 0;
 691     j_coord_offsetC = 0;
 692     j_coord_offsetD = 0;
 693
 694     outeriter        = 0;
 695     inneriter        = 0;
 696
 697     for(iidx=0;iidx<4*DIM;iidx++)
 698     {
 699         scratch[iidx] = 0.0;
 700     }
 701
 702     /* Start outer loop over neighborlists */
 703     for(iidx=0; iidx<nri; iidx++)
 704     {
 705         /* Load shift vector for this list */
 706         i_shift_offset   = DIM*shiftidx[iidx];
 707
 708         /* Load limits for loop over neighbors */
 709         j_index_start    = jindex[iidx];
 710         j_index_end      = jindex[iidx+1];
 711
 712         /* Get outer coordinate index */
 713         inr              = iinr[iidx];
 714         i_coord_offset   = DIM*inr;
 715
 716         /* Load i particle coords and add shift vector */
 717         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 718                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 719
 720         fix0             = _mm_setzero_ps();
 721         fiy0             = _mm_setzero_ps();
 722         fiz0             = _mm_setzero_ps();
 723         fix1             = _mm_setzero_ps();
 724         fiy1             = _mm_setzero_ps();
 725         fiz1             = _mm_setzero_ps();
 726         fix2             = _mm_setzero_ps();
 727         fiy2             = _mm_setzero_ps();
 728         fiz2             = _mm_setzero_ps();
 729
 730         /* Start inner kernel loop */
 731         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 732         {
 733
 734             /* Get j neighbor index, and coordinate index */
 735             jnrA             = jjnr[jidx];
 736             jnrB             = jjnr[jidx+1];
 737             jnrC             = jjnr[jidx+2];
 738             jnrD             = jjnr[jidx+3];
 739             j_coord_offsetA  = DIM*jnrA;
 740             j_coord_offsetB  = DIM*jnrB;
 741             j_coord_offsetC  = DIM*jnrC;
 742             j_coord_offsetD  = DIM*jnrD;
 743
 744             /* load j atom coordinates */
 745             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 746                                               x+j_coord_offsetC,x+j_coord_offsetD,
 747                                               &jx0,&jy0,&jz0);
 748
 749             /* Calculate displacement vector */
 750             dx00             = _mm_sub_ps(ix0,jx0);
 751             dy00             = _mm_sub_ps(iy0,jy0);
 752             dz00             = _mm_sub_ps(iz0,jz0);
 753             dx10             = _mm_sub_ps(ix1,jx0);
 754             dy10             = _mm_sub_ps(iy1,jy0);
 755             dz10             = _mm_sub_ps(iz1,jz0);
 756             dx20             = _mm_sub_ps(ix2,jx0);
 757             dy20             = _mm_sub_ps(iy2,jy0);
 758             dz20             = _mm_sub_ps(iz2,jz0);
 759
 760             /* Calculate squared distance and things based on it */
 761             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 762             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 763             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 764
 765             rinv00           = sse2_invsqrt_f(rsq00);
 766             rinv10           = sse2_invsqrt_f(rsq10);
 767             rinv20           = sse2_invsqrt_f(rsq20);
 768
 769             /* Load parameters for j particles */
 770             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 771                                                               charge+jnrC+0,charge+jnrD+0);
 772
 773             fjx0             = _mm_setzero_ps();
 774             fjy0             = _mm_setzero_ps();
 775             fjz0             = _mm_setzero_ps();
 776
 777             /**************************
 778              * CALCULATE INTERACTIONS *
 779              **************************/
 780
 781             r00              = _mm_mul_ps(rsq00,rinv00);
 782
 783             /* Compute parameters for interactions between i and j atoms */
 784             qq00             = _mm_mul_ps(iq0,jq0);
 785
 786             /* Calculate table index by multiplying r with table scale and truncate to integer */
 787             rt               = _mm_mul_ps(r00,vftabscale);
 788             vfitab           = _mm_cvttps_epi32(rt);
 789             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 790             vfitab           = _mm_slli_epi32(vfitab,2);
 791
 792             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 793             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 794             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 795             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 796             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 797             _MM_TRANSPOSE4_PS(Y,F,G,H);
 798             Heps             = _mm_mul_ps(vfeps,H);
 799             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 800             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 801             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 802
 803             fscal            = felec;
 804
 805             /* Calculate temporary vectorial force */
 806             tx               = _mm_mul_ps(fscal,dx00);
 807             ty               = _mm_mul_ps(fscal,dy00);
 808             tz               = _mm_mul_ps(fscal,dz00);
 809
 810             /* Update vectorial force */
 811             fix0             = _mm_add_ps(fix0,tx);
 812             fiy0             = _mm_add_ps(fiy0,ty);
 813             fiz0             = _mm_add_ps(fiz0,tz);
 814
 815             fjx0             = _mm_add_ps(fjx0,tx);
 816             fjy0             = _mm_add_ps(fjy0,ty);
 817             fjz0             = _mm_add_ps(fjz0,tz);
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             r10              = _mm_mul_ps(rsq10,rinv10);
 824
 825             /* Compute parameters for interactions between i and j atoms */
 826             qq10             = _mm_mul_ps(iq1,jq0);
 827
 828             /* Calculate table index by multiplying r with table scale and truncate to integer */
 829             rt               = _mm_mul_ps(r10,vftabscale);
 830             vfitab           = _mm_cvttps_epi32(rt);
 831             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 832             vfitab           = _mm_slli_epi32(vfitab,2);
 833
 834             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 835             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 836             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 837             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 838             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 839             _MM_TRANSPOSE4_PS(Y,F,G,H);
 840             Heps             = _mm_mul_ps(vfeps,H);
 841             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 842             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 843             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 844
 845             fscal            = felec;
 846
 847             /* Calculate temporary vectorial force */
 848             tx               = _mm_mul_ps(fscal,dx10);
 849             ty               = _mm_mul_ps(fscal,dy10);
 850             tz               = _mm_mul_ps(fscal,dz10);
 851
 852             /* Update vectorial force */
 853             fix1             = _mm_add_ps(fix1,tx);
 854             fiy1             = _mm_add_ps(fiy1,ty);
 855             fiz1             = _mm_add_ps(fiz1,tz);
 856
 857             fjx0             = _mm_add_ps(fjx0,tx);
 858             fjy0             = _mm_add_ps(fjy0,ty);
 859             fjz0             = _mm_add_ps(fjz0,tz);
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             r20              = _mm_mul_ps(rsq20,rinv20);
 866
 867             /* Compute parameters for interactions between i and j atoms */
 868             qq20             = _mm_mul_ps(iq2,jq0);
 869
 870             /* Calculate table index by multiplying r with table scale and truncate to integer */
 871             rt               = _mm_mul_ps(r20,vftabscale);
 872             vfitab           = _mm_cvttps_epi32(rt);
 873             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 874             vfitab           = _mm_slli_epi32(vfitab,2);
 875
 876             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 877             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 878             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 879             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 880             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 881             _MM_TRANSPOSE4_PS(Y,F,G,H);
 882             Heps             = _mm_mul_ps(vfeps,H);
 883             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 884             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 885             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 886
 887             fscal            = felec;
 888
 889             /* Calculate temporary vectorial force */
 890             tx               = _mm_mul_ps(fscal,dx20);
 891             ty               = _mm_mul_ps(fscal,dy20);
 892             tz               = _mm_mul_ps(fscal,dz20);
 893
 894             /* Update vectorial force */
 895             fix2             = _mm_add_ps(fix2,tx);
 896             fiy2             = _mm_add_ps(fiy2,ty);
 897             fiz2             = _mm_add_ps(fiz2,tz);
 898
 899             fjx0             = _mm_add_ps(fjx0,tx);
 900             fjy0             = _mm_add_ps(fjy0,ty);
 901             fjz0             = _mm_add_ps(fjz0,tz);
 902
 903             fjptrA             = f+j_coord_offsetA;
 904             fjptrB             = f+j_coord_offsetB;
 905             fjptrC             = f+j_coord_offsetC;
 906             fjptrD             = f+j_coord_offsetD;
 907
 908             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 909
 910             /* Inner loop uses 117 flops */
 911         }
 912
 913         if(jidx<j_index_end)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrlistA         = jjnr[jidx];
 918             jnrlistB         = jjnr[jidx+1];
 919             jnrlistC         = jjnr[jidx+2];
 920             jnrlistD         = jjnr[jidx+3];
 921             /* Sign of each element will be negative for non-real atoms.
 922              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 923              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 924              */
 925             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 926             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 927             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 928             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 929             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 930             j_coord_offsetA  = DIM*jnrA;
 931             j_coord_offsetB  = DIM*jnrB;
 932             j_coord_offsetC  = DIM*jnrC;
 933             j_coord_offsetD  = DIM*jnrD;
 934
 935             /* load j atom coordinates */
 936             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 937                                               x+j_coord_offsetC,x+j_coord_offsetD,
 938                                               &jx0,&jy0,&jz0);
 939
 940             /* Calculate displacement vector */
 941             dx00             = _mm_sub_ps(ix0,jx0);
 942             dy00             = _mm_sub_ps(iy0,jy0);
 943             dz00             = _mm_sub_ps(iz0,jz0);
 944             dx10             = _mm_sub_ps(ix1,jx0);
 945             dy10             = _mm_sub_ps(iy1,jy0);
 946             dz10             = _mm_sub_ps(iz1,jz0);
 947             dx20             = _mm_sub_ps(ix2,jx0);
 948             dy20             = _mm_sub_ps(iy2,jy0);
 949             dz20             = _mm_sub_ps(iz2,jz0);
 950
 951             /* Calculate squared distance and things based on it */
 952             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 953             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 954             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 955
 956             rinv00           = sse2_invsqrt_f(rsq00);
 957             rinv10           = sse2_invsqrt_f(rsq10);
 958             rinv20           = sse2_invsqrt_f(rsq20);
 959
 960             /* Load parameters for j particles */
 961             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 962                                                               charge+jnrC+0,charge+jnrD+0);
 963
 964             fjx0             = _mm_setzero_ps();
 965             fjy0             = _mm_setzero_ps();
 966             fjz0             = _mm_setzero_ps();
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             r00              = _mm_mul_ps(rsq00,rinv00);
 973             r00              = _mm_andnot_ps(dummy_mask,r00);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq00             = _mm_mul_ps(iq0,jq0);
 977
 978             /* Calculate table index by multiplying r with table scale and truncate to integer */
 979             rt               = _mm_mul_ps(r00,vftabscale);
 980             vfitab           = _mm_cvttps_epi32(rt);
 981             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 982             vfitab           = _mm_slli_epi32(vfitab,2);
 983
 984             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 985             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 986             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 987             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 988             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 989             _MM_TRANSPOSE4_PS(Y,F,G,H);
 990             Heps             = _mm_mul_ps(vfeps,H);
 991             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 992             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 993             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 994
 995             fscal            = felec;
 996
 997             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 998
 999             /* Calculate temporary vectorial force */
1000             tx               = _mm_mul_ps(fscal,dx00);
1001             ty               = _mm_mul_ps(fscal,dy00);
1002             tz               = _mm_mul_ps(fscal,dz00);
1003
1004             /* Update vectorial force */
1005             fix0             = _mm_add_ps(fix0,tx);
1006             fiy0             = _mm_add_ps(fiy0,ty);
1007             fiz0             = _mm_add_ps(fiz0,tz);
1008
1009             fjx0             = _mm_add_ps(fjx0,tx);
1010             fjy0             = _mm_add_ps(fjy0,ty);
1011             fjz0             = _mm_add_ps(fjz0,tz);
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             r10              = _mm_mul_ps(rsq10,rinv10);
1018             r10              = _mm_andnot_ps(dummy_mask,r10);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq10             = _mm_mul_ps(iq1,jq0);
1022
1023             /* Calculate table index by multiplying r with table scale and truncate to integer */
1024             rt               = _mm_mul_ps(r10,vftabscale);
1025             vfitab           = _mm_cvttps_epi32(rt);
1026             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1027             vfitab           = _mm_slli_epi32(vfitab,2);
1028
1029             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1030             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1031             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1032             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1033             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1034             _MM_TRANSPOSE4_PS(Y,F,G,H);
1035             Heps             = _mm_mul_ps(vfeps,H);
1036             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1037             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1038             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1039
1040             fscal            = felec;
1041
1042             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1043
1044             /* Calculate temporary vectorial force */
1045             tx               = _mm_mul_ps(fscal,dx10);
1046             ty               = _mm_mul_ps(fscal,dy10);
1047             tz               = _mm_mul_ps(fscal,dz10);
1048
1049             /* Update vectorial force */
1050             fix1             = _mm_add_ps(fix1,tx);
1051             fiy1             = _mm_add_ps(fiy1,ty);
1052             fiz1             = _mm_add_ps(fiz1,tz);
1053
1054             fjx0             = _mm_add_ps(fjx0,tx);
1055             fjy0             = _mm_add_ps(fjy0,ty);
1056             fjz0             = _mm_add_ps(fjz0,tz);
1057
1058             /**************************
1059              * CALCULATE INTERACTIONS *
1060              **************************/
1061
1062             r20              = _mm_mul_ps(rsq20,rinv20);
1063             r20              = _mm_andnot_ps(dummy_mask,r20);
1064
1065             /* Compute parameters for interactions between i and j atoms */
1066             qq20             = _mm_mul_ps(iq2,jq0);
1067
1068             /* Calculate table index by multiplying r with table scale and truncate to integer */
1069             rt               = _mm_mul_ps(r20,vftabscale);
1070             vfitab           = _mm_cvttps_epi32(rt);
1071             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1072             vfitab           = _mm_slli_epi32(vfitab,2);
1073
1074             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1075             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1076             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1077             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1078             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1079             _MM_TRANSPOSE4_PS(Y,F,G,H);
1080             Heps             = _mm_mul_ps(vfeps,H);
1081             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1082             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1083             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1084
1085             fscal            = felec;
1086
1087             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm_mul_ps(fscal,dx20);
1091             ty               = _mm_mul_ps(fscal,dy20);
1092             tz               = _mm_mul_ps(fscal,dz20);
1093
1094             /* Update vectorial force */
1095             fix2             = _mm_add_ps(fix2,tx);
1096             fiy2             = _mm_add_ps(fiy2,ty);
1097             fiz2             = _mm_add_ps(fiz2,tz);
1098
1099             fjx0             = _mm_add_ps(fjx0,tx);
1100             fjy0             = _mm_add_ps(fjy0,ty);
1101             fjz0             = _mm_add_ps(fjz0,tz);
1102
1103             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1104             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1105             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1106             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1107
1108             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1109
1110             /* Inner loop uses 120 flops */
1111         }
1112
1113         /* End of innermost loop */
1114
1115         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1116                                               f+i_coord_offset,fshift+i_shift_offset);
1117
1118         /* Increment number of inner iterations */
1119         inneriter                  += j_index_end - j_index_start;
1120
1121         /* Outer loop uses 18 flops */
1122     }
1123
1124     /* Increment number of outer iterations */
1125     outeriter        += nri;
1126
1127     /* Update outer/inner flops */
1128
1129     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*120);
1130 }