src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm_set1_pd(fr->ic->k_rf);
 124     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_pd(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 170
 171         fix0             = _mm_setzero_pd();
 172         fiy0             = _mm_setzero_pd();
 173         fiz0             = _mm_setzero_pd();
 174         fix1             = _mm_setzero_pd();
 175         fiy1             = _mm_setzero_pd();
 176         fiz1             = _mm_setzero_pd();
 177         fix2             = _mm_setzero_pd();
 178         fiy2             = _mm_setzero_pd();
 179         fiz2             = _mm_setzero_pd();
 180         fix3             = _mm_setzero_pd();
 181         fiy3             = _mm_setzero_pd();
 182         fiz3             = _mm_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_pd();
 186         vvdwsum          = _mm_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_pd(ix0,jx0);
 204             dy00             = _mm_sub_pd(iy0,jy0);
 205             dz00             = _mm_sub_pd(iz0,jz0);
 206             dx10             = _mm_sub_pd(ix1,jx0);
 207             dy10             = _mm_sub_pd(iy1,jy0);
 208             dz10             = _mm_sub_pd(iz1,jz0);
 209             dx20             = _mm_sub_pd(ix2,jx0);
 210             dy20             = _mm_sub_pd(iy2,jy0);
 211             dz20             = _mm_sub_pd(iz2,jz0);
 212             dx30             = _mm_sub_pd(ix3,jx0);
 213             dy30             = _mm_sub_pd(iy3,jy0);
 214             dz30             = _mm_sub_pd(iz3,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 220             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 221
 222             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 223             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 224             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 225             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 226
 227             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 229             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235
 236             fjx0             = _mm_setzero_pd();
 237             fjy0             = _mm_setzero_pd();
 238             fjz0             = _mm_setzero_pd();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             r00              = _mm_mul_pd(rsq00,rinv00);
 245
 246             /* Compute parameters for interactions between i and j atoms */
 247             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 249
 250             /* Calculate table index by multiplying r with table scale and truncate to integer */
 251             rt               = _mm_mul_pd(r00,vftabscale);
 252             vfitab           = _mm_cvttpd_epi32(rt);
 253             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 254             vfitab           = _mm_slli_epi32(vfitab,3);
 255
 256             /* CUBIC SPLINE TABLE DISPERSION */
 257             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 258             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 259             GMX_MM_TRANSPOSE2_PD(Y,F);
 260             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 261             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 262             GMX_MM_TRANSPOSE2_PD(G,H);
 263             Heps             = _mm_mul_pd(vfeps,H);
 264             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 265             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 266             vvdw6            = _mm_mul_pd(c6_00,VV);
 267             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 268             fvdw6            = _mm_mul_pd(c6_00,FF);
 269
 270             /* CUBIC SPLINE TABLE REPULSION */
 271             vfitab           = _mm_add_epi32(vfitab,ifour);
 272             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 273             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 274             GMX_MM_TRANSPOSE2_PD(Y,F);
 275             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 276             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 277             GMX_MM_TRANSPOSE2_PD(G,H);
 278             Heps             = _mm_mul_pd(vfeps,H);
 279             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 280             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 281             vvdw12           = _mm_mul_pd(c12_00,VV);
 282             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 283             fvdw12           = _mm_mul_pd(c12_00,FF);
 284             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 285             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 289
 290             fscal            = fvdw;
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm_mul_pd(fscal,dx00);
 294             ty               = _mm_mul_pd(fscal,dy00);
 295             tz               = _mm_mul_pd(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm_add_pd(fix0,tx);
 299             fiy0             = _mm_add_pd(fiy0,ty);
 300             fiz0             = _mm_add_pd(fiz0,tz);
 301
 302             fjx0             = _mm_add_pd(fjx0,tx);
 303             fjy0             = _mm_add_pd(fjy0,ty);
 304             fjz0             = _mm_add_pd(fjz0,tz);
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             if (gmx_mm_any_lt(rsq10,rcutoff2))
 311             {
 312
 313             /* Compute parameters for interactions between i and j atoms */
 314             qq10             = _mm_mul_pd(iq1,jq0);
 315
 316             /* REACTION-FIELD ELECTROSTATICS */
 317             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 318             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 319
 320             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velec            = _mm_and_pd(velec,cutoff_mask);
 324             velecsum         = _mm_add_pd(velecsum,velec);
 325
 326             fscal            = felec;
 327
 328             fscal            = _mm_and_pd(fscal,cutoff_mask);
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_pd(fscal,dx10);
 332             ty               = _mm_mul_pd(fscal,dy10);
 333             tz               = _mm_mul_pd(fscal,dz10);
 334
 335             /* Update vectorial force */
 336             fix1             = _mm_add_pd(fix1,tx);
 337             fiy1             = _mm_add_pd(fiy1,ty);
 338             fiz1             = _mm_add_pd(fiz1,tz);
 339
 340             fjx0             = _mm_add_pd(fjx0,tx);
 341             fjy0             = _mm_add_pd(fjy0,ty);
 342             fjz0             = _mm_add_pd(fjz0,tz);
 343
 344             }
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             if (gmx_mm_any_lt(rsq20,rcutoff2))
 351             {
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_pd(iq2,jq0);
 355
 356             /* REACTION-FIELD ELECTROSTATICS */
 357             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 358             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 359
 360             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velec            = _mm_and_pd(velec,cutoff_mask);
 364             velecsum         = _mm_add_pd(velecsum,velec);
 365
 366             fscal            = felec;
 367
 368             fscal            = _mm_and_pd(fscal,cutoff_mask);
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm_mul_pd(fscal,dx20);
 372             ty               = _mm_mul_pd(fscal,dy20);
 373             tz               = _mm_mul_pd(fscal,dz20);
 374
 375             /* Update vectorial force */
 376             fix2             = _mm_add_pd(fix2,tx);
 377             fiy2             = _mm_add_pd(fiy2,ty);
 378             fiz2             = _mm_add_pd(fiz2,tz);
 379
 380             fjx0             = _mm_add_pd(fjx0,tx);
 381             fjy0             = _mm_add_pd(fjy0,ty);
 382             fjz0             = _mm_add_pd(fjz0,tz);
 383
 384             }
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             if (gmx_mm_any_lt(rsq30,rcutoff2))
 391             {
 392
 393             /* Compute parameters for interactions between i and j atoms */
 394             qq30             = _mm_mul_pd(iq3,jq0);
 395
 396             /* REACTION-FIELD ELECTROSTATICS */
 397             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 398             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 399
 400             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             velec            = _mm_and_pd(velec,cutoff_mask);
 404             velecsum         = _mm_add_pd(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             fscal            = _mm_and_pd(fscal,cutoff_mask);
 409
 410             /* Calculate temporary vectorial force */
 411             tx               = _mm_mul_pd(fscal,dx30);
 412             ty               = _mm_mul_pd(fscal,dy30);
 413             tz               = _mm_mul_pd(fscal,dz30);
 414
 415             /* Update vectorial force */
 416             fix3             = _mm_add_pd(fix3,tx);
 417             fiy3             = _mm_add_pd(fiy3,ty);
 418             fiz3             = _mm_add_pd(fiz3,tz);
 419
 420             fjx0             = _mm_add_pd(fjx0,tx);
 421             fjy0             = _mm_add_pd(fjy0,ty);
 422             fjz0             = _mm_add_pd(fjz0,tz);
 423
 424             }
 425
 426             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 427
 428             /* Inner loop uses 167 flops */
 429         }
 430
 431         if(jidx<j_index_end)
 432         {
 433
 434             jnrA             = jjnr[jidx];
 435             j_coord_offsetA  = DIM*jnrA;
 436
 437             /* load j atom coordinates */
 438             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 439                                               &jx0,&jy0,&jz0);
 440
 441             /* Calculate displacement vector */
 442             dx00             = _mm_sub_pd(ix0,jx0);
 443             dy00             = _mm_sub_pd(iy0,jy0);
 444             dz00             = _mm_sub_pd(iz0,jz0);
 445             dx10             = _mm_sub_pd(ix1,jx0);
 446             dy10             = _mm_sub_pd(iy1,jy0);
 447             dz10             = _mm_sub_pd(iz1,jz0);
 448             dx20             = _mm_sub_pd(ix2,jx0);
 449             dy20             = _mm_sub_pd(iy2,jy0);
 450             dz20             = _mm_sub_pd(iz2,jz0);
 451             dx30             = _mm_sub_pd(ix3,jx0);
 452             dy30             = _mm_sub_pd(iy3,jy0);
 453             dz30             = _mm_sub_pd(iz3,jz0);
 454
 455             /* Calculate squared distance and things based on it */
 456             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 457             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 458             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 459             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 460
 461             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 462             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 463             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 464             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 465
 466             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 467             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 468             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 469
 470             /* Load parameters for j particles */
 471             jq0              = _mm_load_sd(charge+jnrA+0);
 472             vdwjidx0A        = 2*vdwtype[jnrA+0];
 473
 474             fjx0             = _mm_setzero_pd();
 475             fjy0             = _mm_setzero_pd();
 476             fjz0             = _mm_setzero_pd();
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             r00              = _mm_mul_pd(rsq00,rinv00);
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 486
 487             /* Calculate table index by multiplying r with table scale and truncate to integer */
 488             rt               = _mm_mul_pd(r00,vftabscale);
 489             vfitab           = _mm_cvttpd_epi32(rt);
 490             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 491             vfitab           = _mm_slli_epi32(vfitab,3);
 492
 493             /* CUBIC SPLINE TABLE DISPERSION */
 494             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 495             F                = _mm_setzero_pd();
 496             GMX_MM_TRANSPOSE2_PD(Y,F);
 497             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 498             H                = _mm_setzero_pd();
 499             GMX_MM_TRANSPOSE2_PD(G,H);
 500             Heps             = _mm_mul_pd(vfeps,H);
 501             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 502             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 503             vvdw6            = _mm_mul_pd(c6_00,VV);
 504             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 505             fvdw6            = _mm_mul_pd(c6_00,FF);
 506
 507             /* CUBIC SPLINE TABLE REPULSION */
 508             vfitab           = _mm_add_epi32(vfitab,ifour);
 509             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 510             F                = _mm_setzero_pd();
 511             GMX_MM_TRANSPOSE2_PD(Y,F);
 512             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 513             H                = _mm_setzero_pd();
 514             GMX_MM_TRANSPOSE2_PD(G,H);
 515             Heps             = _mm_mul_pd(vfeps,H);
 516             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 517             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 518             vvdw12           = _mm_mul_pd(c12_00,VV);
 519             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 520             fvdw12           = _mm_mul_pd(c12_00,FF);
 521             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 522             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 526             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 527
 528             fscal            = fvdw;
 529
 530             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm_mul_pd(fscal,dx00);
 534             ty               = _mm_mul_pd(fscal,dy00);
 535             tz               = _mm_mul_pd(fscal,dz00);
 536
 537             /* Update vectorial force */
 538             fix0             = _mm_add_pd(fix0,tx);
 539             fiy0             = _mm_add_pd(fiy0,ty);
 540             fiz0             = _mm_add_pd(fiz0,tz);
 541
 542             fjx0             = _mm_add_pd(fjx0,tx);
 543             fjy0             = _mm_add_pd(fjy0,ty);
 544             fjz0             = _mm_add_pd(fjz0,tz);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             if (gmx_mm_any_lt(rsq10,rcutoff2))
 551             {
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq10             = _mm_mul_pd(iq1,jq0);
 555
 556             /* REACTION-FIELD ELECTROSTATICS */
 557             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 558             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 559
 560             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velec            = _mm_and_pd(velec,cutoff_mask);
 564             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 565             velecsum         = _mm_add_pd(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             fscal            = _mm_and_pd(fscal,cutoff_mask);
 570
 571             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm_mul_pd(fscal,dx10);
 575             ty               = _mm_mul_pd(fscal,dy10);
 576             tz               = _mm_mul_pd(fscal,dz10);
 577
 578             /* Update vectorial force */
 579             fix1             = _mm_add_pd(fix1,tx);
 580             fiy1             = _mm_add_pd(fiy1,ty);
 581             fiz1             = _mm_add_pd(fiz1,tz);
 582
 583             fjx0             = _mm_add_pd(fjx0,tx);
 584             fjy0             = _mm_add_pd(fjy0,ty);
 585             fjz0             = _mm_add_pd(fjz0,tz);
 586
 587             }
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             if (gmx_mm_any_lt(rsq20,rcutoff2))
 594             {
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq20             = _mm_mul_pd(iq2,jq0);
 598
 599             /* REACTION-FIELD ELECTROSTATICS */
 600             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 601             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 602
 603             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_and_pd(velec,cutoff_mask);
 607             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 608             velecsum         = _mm_add_pd(velecsum,velec);
 609
 610             fscal            = felec;
 611
 612             fscal            = _mm_and_pd(fscal,cutoff_mask);
 613
 614             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 615
 616             /* Calculate temporary vectorial force */
 617             tx               = _mm_mul_pd(fscal,dx20);
 618             ty               = _mm_mul_pd(fscal,dy20);
 619             tz               = _mm_mul_pd(fscal,dz20);
 620
 621             /* Update vectorial force */
 622             fix2             = _mm_add_pd(fix2,tx);
 623             fiy2             = _mm_add_pd(fiy2,ty);
 624             fiz2             = _mm_add_pd(fiz2,tz);
 625
 626             fjx0             = _mm_add_pd(fjx0,tx);
 627             fjy0             = _mm_add_pd(fjy0,ty);
 628             fjz0             = _mm_add_pd(fjz0,tz);
 629
 630             }
 631
 632             /**************************
 633              * CALCULATE INTERACTIONS *
 634              **************************/
 635
 636             if (gmx_mm_any_lt(rsq30,rcutoff2))
 637             {
 638
 639             /* Compute parameters for interactions between i and j atoms */
 640             qq30             = _mm_mul_pd(iq3,jq0);
 641
 642             /* REACTION-FIELD ELECTROSTATICS */
 643             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 644             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 645
 646             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm_and_pd(velec,cutoff_mask);
 650             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 651             velecsum         = _mm_add_pd(velecsum,velec);
 652
 653             fscal            = felec;
 654
 655             fscal            = _mm_and_pd(fscal,cutoff_mask);
 656
 657             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm_mul_pd(fscal,dx30);
 661             ty               = _mm_mul_pd(fscal,dy30);
 662             tz               = _mm_mul_pd(fscal,dz30);
 663
 664             /* Update vectorial force */
 665             fix3             = _mm_add_pd(fix3,tx);
 666             fiy3             = _mm_add_pd(fiy3,ty);
 667             fiz3             = _mm_add_pd(fiz3,tz);
 668
 669             fjx0             = _mm_add_pd(fjx0,tx);
 670             fjy0             = _mm_add_pd(fjy0,ty);
 671             fjz0             = _mm_add_pd(fjz0,tz);
 672
 673             }
 674
 675             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 676
 677             /* Inner loop uses 167 flops */
 678         }
 679
 680         /* End of innermost loop */
 681
 682         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 683                                               f+i_coord_offset,fshift+i_shift_offset);
 684
 685         ggid                        = gid[iidx];
 686         /* Update potential energies */
 687         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 688         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 689
 690         /* Increment number of inner iterations */
 691         inneriter                  += j_index_end - j_index_start;
 692
 693         /* Outer loop uses 26 flops */
 694     }
 695
 696     /* Increment number of outer iterations */
 697     outeriter        += nri;
 698
 699     /* Update outer/inner flops */
 700
 701     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 702 }
 703 /*
 704  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_double
 705  * Electrostatics interaction: ReactionField
 706  * VdW interaction:            CubicSplineTable
 707  * Geometry:                   Water4-Particle
 708  * Calculate force/pot:        Force
 709  */
 710 void
 711 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_double
 712                     (t_nblist                    * gmx_restrict       nlist,
 713                      rvec                        * gmx_restrict          xx,
 714                      rvec                        * gmx_restrict          ff,
 715                      t_forcerec                  * gmx_restrict          fr,
 716                      t_mdatoms                   * gmx_restrict     mdatoms,
 717                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 718                      t_nrnb                      * gmx_restrict        nrnb)
 719 {
 720     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 721      * just 0 for non-waters.
 722      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 723      * jnr indices corresponding to data put in the four positions in the SIMD register.
 724      */
 725     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 726     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 727     int              jnrA,jnrB;
 728     int              j_coord_offsetA,j_coord_offsetB;
 729     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 730     real             rcutoff_scalar;
 731     real             *shiftvec,*fshift,*x,*f;
 732     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 733     int              vdwioffset0;
 734     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 735     int              vdwioffset1;
 736     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 737     int              vdwioffset2;
 738     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 739     int              vdwioffset3;
 740     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 741     int              vdwjidx0A,vdwjidx0B;
 742     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 743     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 744     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 745     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 746     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 747     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 748     real             *charge;
 749     int              nvdwtype;
 750     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 751     int              *vdwtype;
 752     real             *vdwparam;
 753     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 754     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 755     __m128i          vfitab;
 756     __m128i          ifour       = _mm_set1_epi32(4);
 757     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 758     real             *vftab;
 759     __m128d          dummy_mask,cutoff_mask;
 760     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 761     __m128d          one     = _mm_set1_pd(1.0);
 762     __m128d          two     = _mm_set1_pd(2.0);
 763     x                = xx[0];
 764     f                = ff[0];
 765
 766     nri              = nlist->nri;
 767     iinr             = nlist->iinr;
 768     jindex           = nlist->jindex;
 769     jjnr             = nlist->jjnr;
 770     shiftidx         = nlist->shift;
 771     gid              = nlist->gid;
 772     shiftvec         = fr->shift_vec[0];
 773     fshift           = fr->fshift[0];
 774     facel            = _mm_set1_pd(fr->epsfac);
 775     charge           = mdatoms->chargeA;
 776     krf              = _mm_set1_pd(fr->ic->k_rf);
 777     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 778     crf              = _mm_set1_pd(fr->ic->c_rf);
 779     nvdwtype         = fr->ntype;
 780     vdwparam         = fr->nbfp;
 781     vdwtype          = mdatoms->typeA;
 782
 783     vftab            = kernel_data->table_vdw->data;
 784     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 785
 786     /* Setup water-specific parameters */
 787     inr              = nlist->iinr[0];
 788     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 789     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 790     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 791     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 792
 793     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 794     rcutoff_scalar   = fr->rcoulomb;
 795     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 796     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 797
 798     /* Avoid stupid compiler warnings */
 799     jnrA = jnrB = 0;
 800     j_coord_offsetA = 0;
 801     j_coord_offsetB = 0;
 802
 803     outeriter        = 0;
 804     inneriter        = 0;
 805
 806     /* Start outer loop over neighborlists */
 807     for(iidx=0; iidx<nri; iidx++)
 808     {
 809         /* Load shift vector for this list */
 810         i_shift_offset   = DIM*shiftidx[iidx];
 811
 812         /* Load limits for loop over neighbors */
 813         j_index_start    = jindex[iidx];
 814         j_index_end      = jindex[iidx+1];
 815
 816         /* Get outer coordinate index */
 817         inr              = iinr[iidx];
 818         i_coord_offset   = DIM*inr;
 819
 820         /* Load i particle coords and add shift vector */
 821         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 822                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 823
 824         fix0             = _mm_setzero_pd();
 825         fiy0             = _mm_setzero_pd();
 826         fiz0             = _mm_setzero_pd();
 827         fix1             = _mm_setzero_pd();
 828         fiy1             = _mm_setzero_pd();
 829         fiz1             = _mm_setzero_pd();
 830         fix2             = _mm_setzero_pd();
 831         fiy2             = _mm_setzero_pd();
 832         fiz2             = _mm_setzero_pd();
 833         fix3             = _mm_setzero_pd();
 834         fiy3             = _mm_setzero_pd();
 835         fiz3             = _mm_setzero_pd();
 836
 837         /* Start inner kernel loop */
 838         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 839         {
 840
 841             /* Get j neighbor index, and coordinate index */
 842             jnrA             = jjnr[jidx];
 843             jnrB             = jjnr[jidx+1];
 844             j_coord_offsetA  = DIM*jnrA;
 845             j_coord_offsetB  = DIM*jnrB;
 846
 847             /* load j atom coordinates */
 848             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 849                                               &jx0,&jy0,&jz0);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm_sub_pd(ix0,jx0);
 853             dy00             = _mm_sub_pd(iy0,jy0);
 854             dz00             = _mm_sub_pd(iz0,jz0);
 855             dx10             = _mm_sub_pd(ix1,jx0);
 856             dy10             = _mm_sub_pd(iy1,jy0);
 857             dz10             = _mm_sub_pd(iz1,jz0);
 858             dx20             = _mm_sub_pd(ix2,jx0);
 859             dy20             = _mm_sub_pd(iy2,jy0);
 860             dz20             = _mm_sub_pd(iz2,jz0);
 861             dx30             = _mm_sub_pd(ix3,jx0);
 862             dy30             = _mm_sub_pd(iy3,jy0);
 863             dz30             = _mm_sub_pd(iz3,jz0);
 864
 865             /* Calculate squared distance and things based on it */
 866             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 867             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 868             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 869             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 870
 871             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 872             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 873             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 874             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 875
 876             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 877             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 878             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 879
 880             /* Load parameters for j particles */
 881             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 882             vdwjidx0A        = 2*vdwtype[jnrA+0];
 883             vdwjidx0B        = 2*vdwtype[jnrB+0];
 884
 885             fjx0             = _mm_setzero_pd();
 886             fjy0             = _mm_setzero_pd();
 887             fjz0             = _mm_setzero_pd();
 888
 889             /**************************
 890              * CALCULATE INTERACTIONS *
 891              **************************/
 892
 893             r00              = _mm_mul_pd(rsq00,rinv00);
 894
 895             /* Compute parameters for interactions between i and j atoms */
 896             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 897                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 898
 899             /* Calculate table index by multiplying r with table scale and truncate to integer */
 900             rt               = _mm_mul_pd(r00,vftabscale);
 901             vfitab           = _mm_cvttpd_epi32(rt);
 902             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 903             vfitab           = _mm_slli_epi32(vfitab,3);
 904
 905             /* CUBIC SPLINE TABLE DISPERSION */
 906             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 907             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 908             GMX_MM_TRANSPOSE2_PD(Y,F);
 909             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 910             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 911             GMX_MM_TRANSPOSE2_PD(G,H);
 912             Heps             = _mm_mul_pd(vfeps,H);
 913             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 914             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 915             fvdw6            = _mm_mul_pd(c6_00,FF);
 916
 917             /* CUBIC SPLINE TABLE REPULSION */
 918             vfitab           = _mm_add_epi32(vfitab,ifour);
 919             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 920             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 921             GMX_MM_TRANSPOSE2_PD(Y,F);
 922             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 923             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 924             GMX_MM_TRANSPOSE2_PD(G,H);
 925             Heps             = _mm_mul_pd(vfeps,H);
 926             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 927             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 928             fvdw12           = _mm_mul_pd(c12_00,FF);
 929             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 930
 931             fscal            = fvdw;
 932
 933             /* Calculate temporary vectorial force */
 934             tx               = _mm_mul_pd(fscal,dx00);
 935             ty               = _mm_mul_pd(fscal,dy00);
 936             tz               = _mm_mul_pd(fscal,dz00);
 937
 938             /* Update vectorial force */
 939             fix0             = _mm_add_pd(fix0,tx);
 940             fiy0             = _mm_add_pd(fiy0,ty);
 941             fiz0             = _mm_add_pd(fiz0,tz);
 942
 943             fjx0             = _mm_add_pd(fjx0,tx);
 944             fjy0             = _mm_add_pd(fjy0,ty);
 945             fjz0             = _mm_add_pd(fjz0,tz);
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             if (gmx_mm_any_lt(rsq10,rcutoff2))
 952             {
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq10             = _mm_mul_pd(iq1,jq0);
 956
 957             /* REACTION-FIELD ELECTROSTATICS */
 958             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 959
 960             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 961
 962             fscal            = felec;
 963
 964             fscal            = _mm_and_pd(fscal,cutoff_mask);
 965
 966             /* Calculate temporary vectorial force */
 967             tx               = _mm_mul_pd(fscal,dx10);
 968             ty               = _mm_mul_pd(fscal,dy10);
 969             tz               = _mm_mul_pd(fscal,dz10);
 970
 971             /* Update vectorial force */
 972             fix1             = _mm_add_pd(fix1,tx);
 973             fiy1             = _mm_add_pd(fiy1,ty);
 974             fiz1             = _mm_add_pd(fiz1,tz);
 975
 976             fjx0             = _mm_add_pd(fjx0,tx);
 977             fjy0             = _mm_add_pd(fjy0,ty);
 978             fjz0             = _mm_add_pd(fjz0,tz);
 979
 980             }
 981
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             if (gmx_mm_any_lt(rsq20,rcutoff2))
 987             {
 988
 989             /* Compute parameters for interactions between i and j atoms */
 990             qq20             = _mm_mul_pd(iq2,jq0);
 991
 992             /* REACTION-FIELD ELECTROSTATICS */
 993             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 994
 995             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 996
 997             fscal            = felec;
 998
 999             fscal            = _mm_and_pd(fscal,cutoff_mask);
1000
1001             /* Calculate temporary vectorial force */
1002             tx               = _mm_mul_pd(fscal,dx20);
1003             ty               = _mm_mul_pd(fscal,dy20);
1004             tz               = _mm_mul_pd(fscal,dz20);
1005
1006             /* Update vectorial force */
1007             fix2             = _mm_add_pd(fix2,tx);
1008             fiy2             = _mm_add_pd(fiy2,ty);
1009             fiz2             = _mm_add_pd(fiz2,tz);
1010
1011             fjx0             = _mm_add_pd(fjx0,tx);
1012             fjy0             = _mm_add_pd(fjy0,ty);
1013             fjz0             = _mm_add_pd(fjz0,tz);
1014
1015             }
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             if (gmx_mm_any_lt(rsq30,rcutoff2))
1022             {
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             qq30             = _mm_mul_pd(iq3,jq0);
1026
1027             /* REACTION-FIELD ELECTROSTATICS */
1028             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1029
1030             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm_and_pd(fscal,cutoff_mask);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm_mul_pd(fscal,dx30);
1038             ty               = _mm_mul_pd(fscal,dy30);
1039             tz               = _mm_mul_pd(fscal,dz30);
1040
1041             /* Update vectorial force */
1042             fix3             = _mm_add_pd(fix3,tx);
1043             fiy3             = _mm_add_pd(fiy3,ty);
1044             fiz3             = _mm_add_pd(fiz3,tz);
1045
1046             fjx0             = _mm_add_pd(fjx0,tx);
1047             fjy0             = _mm_add_pd(fjy0,ty);
1048             fjz0             = _mm_add_pd(fjz0,tz);
1049
1050             }
1051
1052             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1053
1054             /* Inner loop uses 141 flops */
1055         }
1056
1057         if(jidx<j_index_end)
1058         {
1059
1060             jnrA             = jjnr[jidx];
1061             j_coord_offsetA  = DIM*jnrA;
1062
1063             /* load j atom coordinates */
1064             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1065                                               &jx0,&jy0,&jz0);
1066
1067             /* Calculate displacement vector */
1068             dx00             = _mm_sub_pd(ix0,jx0);
1069             dy00             = _mm_sub_pd(iy0,jy0);
1070             dz00             = _mm_sub_pd(iz0,jz0);
1071             dx10             = _mm_sub_pd(ix1,jx0);
1072             dy10             = _mm_sub_pd(iy1,jy0);
1073             dz10             = _mm_sub_pd(iz1,jz0);
1074             dx20             = _mm_sub_pd(ix2,jx0);
1075             dy20             = _mm_sub_pd(iy2,jy0);
1076             dz20             = _mm_sub_pd(iz2,jz0);
1077             dx30             = _mm_sub_pd(ix3,jx0);
1078             dy30             = _mm_sub_pd(iy3,jy0);
1079             dz30             = _mm_sub_pd(iz3,jz0);
1080
1081             /* Calculate squared distance and things based on it */
1082             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1083             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1084             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1085             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1086
1087             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1088             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1089             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1090             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1091
1092             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1093             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1094             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1095
1096             /* Load parameters for j particles */
1097             jq0              = _mm_load_sd(charge+jnrA+0);
1098             vdwjidx0A        = 2*vdwtype[jnrA+0];
1099
1100             fjx0             = _mm_setzero_pd();
1101             fjy0             = _mm_setzero_pd();
1102             fjz0             = _mm_setzero_pd();
1103
1104             /**************************
1105              * CALCULATE INTERACTIONS *
1106              **************************/
1107
1108             r00              = _mm_mul_pd(rsq00,rinv00);
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1112
1113             /* Calculate table index by multiplying r with table scale and truncate to integer */
1114             rt               = _mm_mul_pd(r00,vftabscale);
1115             vfitab           = _mm_cvttpd_epi32(rt);
1116             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1117             vfitab           = _mm_slli_epi32(vfitab,3);
1118
1119             /* CUBIC SPLINE TABLE DISPERSION */
1120             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1121             F                = _mm_setzero_pd();
1122             GMX_MM_TRANSPOSE2_PD(Y,F);
1123             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1124             H                = _mm_setzero_pd();
1125             GMX_MM_TRANSPOSE2_PD(G,H);
1126             Heps             = _mm_mul_pd(vfeps,H);
1127             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1128             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1129             fvdw6            = _mm_mul_pd(c6_00,FF);
1130
1131             /* CUBIC SPLINE TABLE REPULSION */
1132             vfitab           = _mm_add_epi32(vfitab,ifour);
1133             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1134             F                = _mm_setzero_pd();
1135             GMX_MM_TRANSPOSE2_PD(Y,F);
1136             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1137             H                = _mm_setzero_pd();
1138             GMX_MM_TRANSPOSE2_PD(G,H);
1139             Heps             = _mm_mul_pd(vfeps,H);
1140             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1141             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1142             fvdw12           = _mm_mul_pd(c12_00,FF);
1143             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1144
1145             fscal            = fvdw;
1146
1147             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1148
1149             /* Calculate temporary vectorial force */
1150             tx               = _mm_mul_pd(fscal,dx00);
1151             ty               = _mm_mul_pd(fscal,dy00);
1152             tz               = _mm_mul_pd(fscal,dz00);
1153
1154             /* Update vectorial force */
1155             fix0             = _mm_add_pd(fix0,tx);
1156             fiy0             = _mm_add_pd(fiy0,ty);
1157             fiz0             = _mm_add_pd(fiz0,tz);
1158
1159             fjx0             = _mm_add_pd(fjx0,tx);
1160             fjy0             = _mm_add_pd(fjy0,ty);
1161             fjz0             = _mm_add_pd(fjz0,tz);
1162
1163             /**************************
1164              * CALCULATE INTERACTIONS *
1165              **************************/
1166
1167             if (gmx_mm_any_lt(rsq10,rcutoff2))
1168             {
1169
1170             /* Compute parameters for interactions between i and j atoms */
1171             qq10             = _mm_mul_pd(iq1,jq0);
1172
1173             /* REACTION-FIELD ELECTROSTATICS */
1174             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1175
1176             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1177
1178             fscal            = felec;
1179
1180             fscal            = _mm_and_pd(fscal,cutoff_mask);
1181
1182             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm_mul_pd(fscal,dx10);
1186             ty               = _mm_mul_pd(fscal,dy10);
1187             tz               = _mm_mul_pd(fscal,dz10);
1188
1189             /* Update vectorial force */
1190             fix1             = _mm_add_pd(fix1,tx);
1191             fiy1             = _mm_add_pd(fiy1,ty);
1192             fiz1             = _mm_add_pd(fiz1,tz);
1193
1194             fjx0             = _mm_add_pd(fjx0,tx);
1195             fjy0             = _mm_add_pd(fjy0,ty);
1196             fjz0             = _mm_add_pd(fjz0,tz);
1197
1198             }
1199
1200             /**************************
1201              * CALCULATE INTERACTIONS *
1202              **************************/
1203
1204             if (gmx_mm_any_lt(rsq20,rcutoff2))
1205             {
1206
1207             /* Compute parameters for interactions between i and j atoms */
1208             qq20             = _mm_mul_pd(iq2,jq0);
1209
1210             /* REACTION-FIELD ELECTROSTATICS */
1211             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1212
1213             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1214
1215             fscal            = felec;
1216
1217             fscal            = _mm_and_pd(fscal,cutoff_mask);
1218
1219             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1220
1221             /* Calculate temporary vectorial force */
1222             tx               = _mm_mul_pd(fscal,dx20);
1223             ty               = _mm_mul_pd(fscal,dy20);
1224             tz               = _mm_mul_pd(fscal,dz20);
1225
1226             /* Update vectorial force */
1227             fix2             = _mm_add_pd(fix2,tx);
1228             fiy2             = _mm_add_pd(fiy2,ty);
1229             fiz2             = _mm_add_pd(fiz2,tz);
1230
1231             fjx0             = _mm_add_pd(fjx0,tx);
1232             fjy0             = _mm_add_pd(fjy0,ty);
1233             fjz0             = _mm_add_pd(fjz0,tz);
1234
1235             }
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             if (gmx_mm_any_lt(rsq30,rcutoff2))
1242             {
1243
1244             /* Compute parameters for interactions between i and j atoms */
1245             qq30             = _mm_mul_pd(iq3,jq0);
1246
1247             /* REACTION-FIELD ELECTROSTATICS */
1248             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1249
1250             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1251
1252             fscal            = felec;
1253
1254             fscal            = _mm_and_pd(fscal,cutoff_mask);
1255
1256             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1257
1258             /* Calculate temporary vectorial force */
1259             tx               = _mm_mul_pd(fscal,dx30);
1260             ty               = _mm_mul_pd(fscal,dy30);
1261             tz               = _mm_mul_pd(fscal,dz30);
1262
1263             /* Update vectorial force */
1264             fix3             = _mm_add_pd(fix3,tx);
1265             fiy3             = _mm_add_pd(fiy3,ty);
1266             fiz3             = _mm_add_pd(fiz3,tz);
1267
1268             fjx0             = _mm_add_pd(fjx0,tx);
1269             fjy0             = _mm_add_pd(fjy0,ty);
1270             fjz0             = _mm_add_pd(fjz0,tz);
1271
1272             }
1273
1274             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1275
1276             /* Inner loop uses 141 flops */
1277         }
1278
1279         /* End of innermost loop */
1280
1281         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1282                                               f+i_coord_offset,fshift+i_shift_offset);
1283
1284         /* Increment number of inner iterations */
1285         inneriter                  += j_index_end - j_index_start;
1286
1287         /* Outer loop uses 24 flops */
1288     }
1289
1290     /* Increment number of outer iterations */
1291     outeriter        += nri;
1292
1293     /* Update outer/inner flops */
1294
1295     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*141);
1296 }