src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_sse2_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 sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse2_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse2_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     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_elec->data;
 128     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 162
 163         fix0             = _mm_setzero_pd();
 164         fiy0             = _mm_setzero_pd();
 165         fiz0             = _mm_setzero_pd();
 166         fix1             = _mm_setzero_pd();
 167         fiy1             = _mm_setzero_pd();
 168         fiz1             = _mm_setzero_pd();
 169         fix2             = _mm_setzero_pd();
 170         fiy2             = _mm_setzero_pd();
 171         fiz2             = _mm_setzero_pd();
 172         fix3             = _mm_setzero_pd();
 173         fiy3             = _mm_setzero_pd();
 174         fiz3             = _mm_setzero_pd();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_pd();
 178         vvdwsum          = _mm_setzero_pd();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_pd(ix0,jx0);
 196             dy00             = _mm_sub_pd(iy0,jy0);
 197             dz00             = _mm_sub_pd(iz0,jz0);
 198             dx10             = _mm_sub_pd(ix1,jx0);
 199             dy10             = _mm_sub_pd(iy1,jy0);
 200             dz10             = _mm_sub_pd(iz1,jz0);
 201             dx20             = _mm_sub_pd(ix2,jx0);
 202             dy20             = _mm_sub_pd(iy2,jy0);
 203             dz20             = _mm_sub_pd(iz2,jz0);
 204             dx30             = _mm_sub_pd(ix3,jx0);
 205             dy30             = _mm_sub_pd(iy3,jy0);
 206             dz30             = _mm_sub_pd(iz3,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 210             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 211             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 212             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 213
 214             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 215             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 216             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 217
 218             rinvsq00         = gmx_mm_inv_pd(rsq00);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224
 225             fjx0             = _mm_setzero_pd();
 226             fjy0             = _mm_setzero_pd();
 227             fjz0             = _mm_setzero_pd();
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 235                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 236
 237             /* LENNARD-JONES DISPERSION/REPULSION */
 238
 239             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 240             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 241             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 242             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 243             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 247
 248             fscal            = fvdw;
 249
 250             /* Calculate temporary vectorial force */
 251             tx               = _mm_mul_pd(fscal,dx00);
 252             ty               = _mm_mul_pd(fscal,dy00);
 253             tz               = _mm_mul_pd(fscal,dz00);
 254
 255             /* Update vectorial force */
 256             fix0             = _mm_add_pd(fix0,tx);
 257             fiy0             = _mm_add_pd(fiy0,ty);
 258             fiz0             = _mm_add_pd(fiz0,tz);
 259
 260             fjx0             = _mm_add_pd(fjx0,tx);
 261             fjy0             = _mm_add_pd(fjy0,ty);
 262             fjz0             = _mm_add_pd(fjz0,tz);
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             r10              = _mm_mul_pd(rsq10,rinv10);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq10             = _mm_mul_pd(iq1,jq0);
 272
 273             /* Calculate table index by multiplying r with table scale and truncate to integer */
 274             rt               = _mm_mul_pd(r10,vftabscale);
 275             vfitab           = _mm_cvttpd_epi32(rt);
 276             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 277             vfitab           = _mm_slli_epi32(vfitab,2);
 278
 279             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 280             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 281             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 282             GMX_MM_TRANSPOSE2_PD(Y,F);
 283             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 284             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 285             GMX_MM_TRANSPOSE2_PD(G,H);
 286             Heps             = _mm_mul_pd(vfeps,H);
 287             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 288             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 289             velec            = _mm_mul_pd(qq10,VV);
 290             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 291             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velecsum         = _mm_add_pd(velecsum,velec);
 295
 296             fscal            = felec;
 297
 298             /* Calculate temporary vectorial force */
 299             tx               = _mm_mul_pd(fscal,dx10);
 300             ty               = _mm_mul_pd(fscal,dy10);
 301             tz               = _mm_mul_pd(fscal,dz10);
 302
 303             /* Update vectorial force */
 304             fix1             = _mm_add_pd(fix1,tx);
 305             fiy1             = _mm_add_pd(fiy1,ty);
 306             fiz1             = _mm_add_pd(fiz1,tz);
 307
 308             fjx0             = _mm_add_pd(fjx0,tx);
 309             fjy0             = _mm_add_pd(fjy0,ty);
 310             fjz0             = _mm_add_pd(fjz0,tz);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             r20              = _mm_mul_pd(rsq20,rinv20);
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq20             = _mm_mul_pd(iq2,jq0);
 320
 321             /* Calculate table index by multiplying r with table scale and truncate to integer */
 322             rt               = _mm_mul_pd(r20,vftabscale);
 323             vfitab           = _mm_cvttpd_epi32(rt);
 324             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 325             vfitab           = _mm_slli_epi32(vfitab,2);
 326
 327             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 328             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 329             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 330             GMX_MM_TRANSPOSE2_PD(Y,F);
 331             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 332             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 333             GMX_MM_TRANSPOSE2_PD(G,H);
 334             Heps             = _mm_mul_pd(vfeps,H);
 335             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 336             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 337             velec            = _mm_mul_pd(qq20,VV);
 338             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 339             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_pd(fscal,dx20);
 348             ty               = _mm_mul_pd(fscal,dy20);
 349             tz               = _mm_mul_pd(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm_add_pd(fix2,tx);
 353             fiy2             = _mm_add_pd(fiy2,ty);
 354             fiz2             = _mm_add_pd(fiz2,tz);
 355
 356             fjx0             = _mm_add_pd(fjx0,tx);
 357             fjy0             = _mm_add_pd(fjy0,ty);
 358             fjz0             = _mm_add_pd(fjz0,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r30              = _mm_mul_pd(rsq30,rinv30);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq30             = _mm_mul_pd(iq3,jq0);
 368
 369             /* Calculate table index by multiplying r with table scale and truncate to integer */
 370             rt               = _mm_mul_pd(r30,vftabscale);
 371             vfitab           = _mm_cvttpd_epi32(rt);
 372             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 373             vfitab           = _mm_slli_epi32(vfitab,2);
 374
 375             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 376             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 377             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 378             GMX_MM_TRANSPOSE2_PD(Y,F);
 379             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 380             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 381             GMX_MM_TRANSPOSE2_PD(G,H);
 382             Heps             = _mm_mul_pd(vfeps,H);
 383             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 384             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 385             velec            = _mm_mul_pd(qq30,VV);
 386             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 387             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velecsum         = _mm_add_pd(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm_mul_pd(fscal,dx30);
 396             ty               = _mm_mul_pd(fscal,dy30);
 397             tz               = _mm_mul_pd(fscal,dz30);
 398
 399             /* Update vectorial force */
 400             fix3             = _mm_add_pd(fix3,tx);
 401             fiy3             = _mm_add_pd(fiy3,ty);
 402             fiz3             = _mm_add_pd(fiz3,tz);
 403
 404             fjx0             = _mm_add_pd(fjx0,tx);
 405             fjy0             = _mm_add_pd(fjy0,ty);
 406             fjz0             = _mm_add_pd(fjz0,tz);
 407
 408             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 409
 410             /* Inner loop uses 164 flops */
 411         }
 412
 413         if(jidx<j_index_end)
 414         {
 415
 416             jnrA             = jjnr[jidx];
 417             j_coord_offsetA  = DIM*jnrA;
 418
 419             /* load j atom coordinates */
 420             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 421                                               &jx0,&jy0,&jz0);
 422
 423             /* Calculate displacement vector */
 424             dx00             = _mm_sub_pd(ix0,jx0);
 425             dy00             = _mm_sub_pd(iy0,jy0);
 426             dz00             = _mm_sub_pd(iz0,jz0);
 427             dx10             = _mm_sub_pd(ix1,jx0);
 428             dy10             = _mm_sub_pd(iy1,jy0);
 429             dz10             = _mm_sub_pd(iz1,jz0);
 430             dx20             = _mm_sub_pd(ix2,jx0);
 431             dy20             = _mm_sub_pd(iy2,jy0);
 432             dz20             = _mm_sub_pd(iz2,jz0);
 433             dx30             = _mm_sub_pd(ix3,jx0);
 434             dy30             = _mm_sub_pd(iy3,jy0);
 435             dz30             = _mm_sub_pd(iz3,jz0);
 436
 437             /* Calculate squared distance and things based on it */
 438             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 439             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 440             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 441             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 442
 443             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 444             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 445             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 446
 447             rinvsq00         = gmx_mm_inv_pd(rsq00);
 448
 449             /* Load parameters for j particles */
 450             jq0              = _mm_load_sd(charge+jnrA+0);
 451             vdwjidx0A        = 2*vdwtype[jnrA+0];
 452
 453             fjx0             = _mm_setzero_pd();
 454             fjy0             = _mm_setzero_pd();
 455             fjz0             = _mm_setzero_pd();
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 463
 464             /* LENNARD-JONES DISPERSION/REPULSION */
 465
 466             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 467             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 468             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 469             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 470             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 474             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 475
 476             fscal            = fvdw;
 477
 478             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 479
 480             /* Calculate temporary vectorial force */
 481             tx               = _mm_mul_pd(fscal,dx00);
 482             ty               = _mm_mul_pd(fscal,dy00);
 483             tz               = _mm_mul_pd(fscal,dz00);
 484
 485             /* Update vectorial force */
 486             fix0             = _mm_add_pd(fix0,tx);
 487             fiy0             = _mm_add_pd(fiy0,ty);
 488             fiz0             = _mm_add_pd(fiz0,tz);
 489
 490             fjx0             = _mm_add_pd(fjx0,tx);
 491             fjy0             = _mm_add_pd(fjy0,ty);
 492             fjz0             = _mm_add_pd(fjz0,tz);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r10              = _mm_mul_pd(rsq10,rinv10);
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq10             = _mm_mul_pd(iq1,jq0);
 502
 503             /* Calculate table index by multiplying r with table scale and truncate to integer */
 504             rt               = _mm_mul_pd(r10,vftabscale);
 505             vfitab           = _mm_cvttpd_epi32(rt);
 506             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 507             vfitab           = _mm_slli_epi32(vfitab,2);
 508
 509             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 510             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 511             F                = _mm_setzero_pd();
 512             GMX_MM_TRANSPOSE2_PD(Y,F);
 513             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 514             H                = _mm_setzero_pd();
 515             GMX_MM_TRANSPOSE2_PD(G,H);
 516             Heps             = _mm_mul_pd(vfeps,H);
 517             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 518             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 519             velec            = _mm_mul_pd(qq10,VV);
 520             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 521             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 525             velecsum         = _mm_add_pd(velecsum,velec);
 526
 527             fscal            = felec;
 528
 529             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_pd(fscal,dx10);
 533             ty               = _mm_mul_pd(fscal,dy10);
 534             tz               = _mm_mul_pd(fscal,dz10);
 535
 536             /* Update vectorial force */
 537             fix1             = _mm_add_pd(fix1,tx);
 538             fiy1             = _mm_add_pd(fiy1,ty);
 539             fiz1             = _mm_add_pd(fiz1,tz);
 540
 541             fjx0             = _mm_add_pd(fjx0,tx);
 542             fjy0             = _mm_add_pd(fjy0,ty);
 543             fjz0             = _mm_add_pd(fjz0,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             r20              = _mm_mul_pd(rsq20,rinv20);
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq20             = _mm_mul_pd(iq2,jq0);
 553
 554             /* Calculate table index by multiplying r with table scale and truncate to integer */
 555             rt               = _mm_mul_pd(r20,vftabscale);
 556             vfitab           = _mm_cvttpd_epi32(rt);
 557             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 558             vfitab           = _mm_slli_epi32(vfitab,2);
 559
 560             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 561             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 562             F                = _mm_setzero_pd();
 563             GMX_MM_TRANSPOSE2_PD(Y,F);
 564             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 565             H                = _mm_setzero_pd();
 566             GMX_MM_TRANSPOSE2_PD(G,H);
 567             Heps             = _mm_mul_pd(vfeps,H);
 568             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 569             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 570             velec            = _mm_mul_pd(qq20,VV);
 571             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 572             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 576             velecsum         = _mm_add_pd(velecsum,velec);
 577
 578             fscal            = felec;
 579
 580             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm_mul_pd(fscal,dx20);
 584             ty               = _mm_mul_pd(fscal,dy20);
 585             tz               = _mm_mul_pd(fscal,dz20);
 586
 587             /* Update vectorial force */
 588             fix2             = _mm_add_pd(fix2,tx);
 589             fiy2             = _mm_add_pd(fiy2,ty);
 590             fiz2             = _mm_add_pd(fiz2,tz);
 591
 592             fjx0             = _mm_add_pd(fjx0,tx);
 593             fjy0             = _mm_add_pd(fjy0,ty);
 594             fjz0             = _mm_add_pd(fjz0,tz);
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             r30              = _mm_mul_pd(rsq30,rinv30);
 601
 602             /* Compute parameters for interactions between i and j atoms */
 603             qq30             = _mm_mul_pd(iq3,jq0);
 604
 605             /* Calculate table index by multiplying r with table scale and truncate to integer */
 606             rt               = _mm_mul_pd(r30,vftabscale);
 607             vfitab           = _mm_cvttpd_epi32(rt);
 608             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 609             vfitab           = _mm_slli_epi32(vfitab,2);
 610
 611             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 612             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 613             F                = _mm_setzero_pd();
 614             GMX_MM_TRANSPOSE2_PD(Y,F);
 615             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 616             H                = _mm_setzero_pd();
 617             GMX_MM_TRANSPOSE2_PD(G,H);
 618             Heps             = _mm_mul_pd(vfeps,H);
 619             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 620             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 621             velec            = _mm_mul_pd(qq30,VV);
 622             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 623             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 624
 625             /* Update potential sum for this i atom from the interaction with this j atom. */
 626             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 627             velecsum         = _mm_add_pd(velecsum,velec);
 628
 629             fscal            = felec;
 630
 631             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm_mul_pd(fscal,dx30);
 635             ty               = _mm_mul_pd(fscal,dy30);
 636             tz               = _mm_mul_pd(fscal,dz30);
 637
 638             /* Update vectorial force */
 639             fix3             = _mm_add_pd(fix3,tx);
 640             fiy3             = _mm_add_pd(fiy3,ty);
 641             fiz3             = _mm_add_pd(fiz3,tz);
 642
 643             fjx0             = _mm_add_pd(fjx0,tx);
 644             fjy0             = _mm_add_pd(fjy0,ty);
 645             fjz0             = _mm_add_pd(fjz0,tz);
 646
 647             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 648
 649             /* Inner loop uses 164 flops */
 650         }
 651
 652         /* End of innermost loop */
 653
 654         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 655                                               f+i_coord_offset,fshift+i_shift_offset);
 656
 657         ggid                        = gid[iidx];
 658         /* Update potential energies */
 659         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 660         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 661
 662         /* Increment number of inner iterations */
 663         inneriter                  += j_index_end - j_index_start;
 664
 665         /* Outer loop uses 26 flops */
 666     }
 667
 668     /* Increment number of outer iterations */
 669     outeriter        += nri;
 670
 671     /* Update outer/inner flops */
 672
 673     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
 674 }
 675 /*
 676  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_double
 677  * Electrostatics interaction: CubicSplineTable
 678  * VdW interaction:            LennardJones
 679  * Geometry:                   Water4-Particle
 680  * Calculate force/pot:        Force
 681  */
 682 void
 683 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_double
 684                     (t_nblist                    * gmx_restrict       nlist,
 685                      rvec                        * gmx_restrict          xx,
 686                      rvec                        * gmx_restrict          ff,
 687                      t_forcerec                  * gmx_restrict          fr,
 688                      t_mdatoms                   * gmx_restrict     mdatoms,
 689                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 690                      t_nrnb                      * gmx_restrict        nrnb)
 691 {
 692     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 693      * just 0 for non-waters.
 694      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 695      * jnr indices corresponding to data put in the four positions in the SIMD register.
 696      */
 697     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 698     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 699     int              jnrA,jnrB;
 700     int              j_coord_offsetA,j_coord_offsetB;
 701     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 702     real             rcutoff_scalar;
 703     real             *shiftvec,*fshift,*x,*f;
 704     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 705     int              vdwioffset0;
 706     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 707     int              vdwioffset1;
 708     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 709     int              vdwioffset2;
 710     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 711     int              vdwioffset3;
 712     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 713     int              vdwjidx0A,vdwjidx0B;
 714     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 715     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 716     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 717     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 718     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 719     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 720     real             *charge;
 721     int              nvdwtype;
 722     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 723     int              *vdwtype;
 724     real             *vdwparam;
 725     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 726     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 727     __m128i          vfitab;
 728     __m128i          ifour       = _mm_set1_epi32(4);
 729     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 730     real             *vftab;
 731     __m128d          dummy_mask,cutoff_mask;
 732     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 733     __m128d          one     = _mm_set1_pd(1.0);
 734     __m128d          two     = _mm_set1_pd(2.0);
 735     x                = xx[0];
 736     f                = ff[0];
 737
 738     nri              = nlist->nri;
 739     iinr             = nlist->iinr;
 740     jindex           = nlist->jindex;
 741     jjnr             = nlist->jjnr;
 742     shiftidx         = nlist->shift;
 743     gid              = nlist->gid;
 744     shiftvec         = fr->shift_vec[0];
 745     fshift           = fr->fshift[0];
 746     facel            = _mm_set1_pd(fr->epsfac);
 747     charge           = mdatoms->chargeA;
 748     nvdwtype         = fr->ntype;
 749     vdwparam         = fr->nbfp;
 750     vdwtype          = mdatoms->typeA;
 751
 752     vftab            = kernel_data->table_elec->data;
 753     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 754
 755     /* Setup water-specific parameters */
 756     inr              = nlist->iinr[0];
 757     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 758     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 759     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 760     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 761
 762     /* Avoid stupid compiler warnings */
 763     jnrA = jnrB = 0;
 764     j_coord_offsetA = 0;
 765     j_coord_offsetB = 0;
 766
 767     outeriter        = 0;
 768     inneriter        = 0;
 769
 770     /* Start outer loop over neighborlists */
 771     for(iidx=0; iidx<nri; iidx++)
 772     {
 773         /* Load shift vector for this list */
 774         i_shift_offset   = DIM*shiftidx[iidx];
 775
 776         /* Load limits for loop over neighbors */
 777         j_index_start    = jindex[iidx];
 778         j_index_end      = jindex[iidx+1];
 779
 780         /* Get outer coordinate index */
 781         inr              = iinr[iidx];
 782         i_coord_offset   = DIM*inr;
 783
 784         /* Load i particle coords and add shift vector */
 785         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 786                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 787
 788         fix0             = _mm_setzero_pd();
 789         fiy0             = _mm_setzero_pd();
 790         fiz0             = _mm_setzero_pd();
 791         fix1             = _mm_setzero_pd();
 792         fiy1             = _mm_setzero_pd();
 793         fiz1             = _mm_setzero_pd();
 794         fix2             = _mm_setzero_pd();
 795         fiy2             = _mm_setzero_pd();
 796         fiz2             = _mm_setzero_pd();
 797         fix3             = _mm_setzero_pd();
 798         fiy3             = _mm_setzero_pd();
 799         fiz3             = _mm_setzero_pd();
 800
 801         /* Start inner kernel loop */
 802         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 803         {
 804
 805             /* Get j neighbor index, and coordinate index */
 806             jnrA             = jjnr[jidx];
 807             jnrB             = jjnr[jidx+1];
 808             j_coord_offsetA  = DIM*jnrA;
 809             j_coord_offsetB  = DIM*jnrB;
 810
 811             /* load j atom coordinates */
 812             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 813                                               &jx0,&jy0,&jz0);
 814
 815             /* Calculate displacement vector */
 816             dx00             = _mm_sub_pd(ix0,jx0);
 817             dy00             = _mm_sub_pd(iy0,jy0);
 818             dz00             = _mm_sub_pd(iz0,jz0);
 819             dx10             = _mm_sub_pd(ix1,jx0);
 820             dy10             = _mm_sub_pd(iy1,jy0);
 821             dz10             = _mm_sub_pd(iz1,jz0);
 822             dx20             = _mm_sub_pd(ix2,jx0);
 823             dy20             = _mm_sub_pd(iy2,jy0);
 824             dz20             = _mm_sub_pd(iz2,jz0);
 825             dx30             = _mm_sub_pd(ix3,jx0);
 826             dy30             = _mm_sub_pd(iy3,jy0);
 827             dz30             = _mm_sub_pd(iz3,jz0);
 828
 829             /* Calculate squared distance and things based on it */
 830             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 831             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 832             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 833             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 834
 835             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 836             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 837             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 838
 839             rinvsq00         = gmx_mm_inv_pd(rsq00);
 840
 841             /* Load parameters for j particles */
 842             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 843             vdwjidx0A        = 2*vdwtype[jnrA+0];
 844             vdwjidx0B        = 2*vdwtype[jnrB+0];
 845
 846             fjx0             = _mm_setzero_pd();
 847             fjy0             = _mm_setzero_pd();
 848             fjz0             = _mm_setzero_pd();
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             /* Compute parameters for interactions between i and j atoms */
 855             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 856                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 857
 858             /* LENNARD-JONES DISPERSION/REPULSION */
 859
 860             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 861             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 862
 863             fscal            = fvdw;
 864
 865             /* Calculate temporary vectorial force */
 866             tx               = _mm_mul_pd(fscal,dx00);
 867             ty               = _mm_mul_pd(fscal,dy00);
 868             tz               = _mm_mul_pd(fscal,dz00);
 869
 870             /* Update vectorial force */
 871             fix0             = _mm_add_pd(fix0,tx);
 872             fiy0             = _mm_add_pd(fiy0,ty);
 873             fiz0             = _mm_add_pd(fiz0,tz);
 874
 875             fjx0             = _mm_add_pd(fjx0,tx);
 876             fjy0             = _mm_add_pd(fjy0,ty);
 877             fjz0             = _mm_add_pd(fjz0,tz);
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             r10              = _mm_mul_pd(rsq10,rinv10);
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq10             = _mm_mul_pd(iq1,jq0);
 887
 888             /* Calculate table index by multiplying r with table scale and truncate to integer */
 889             rt               = _mm_mul_pd(r10,vftabscale);
 890             vfitab           = _mm_cvttpd_epi32(rt);
 891             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 892             vfitab           = _mm_slli_epi32(vfitab,2);
 893
 894             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 895             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 896             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 897             GMX_MM_TRANSPOSE2_PD(Y,F);
 898             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 899             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 900             GMX_MM_TRANSPOSE2_PD(G,H);
 901             Heps             = _mm_mul_pd(vfeps,H);
 902             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 903             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 904             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 905
 906             fscal            = felec;
 907
 908             /* Calculate temporary vectorial force */
 909             tx               = _mm_mul_pd(fscal,dx10);
 910             ty               = _mm_mul_pd(fscal,dy10);
 911             tz               = _mm_mul_pd(fscal,dz10);
 912
 913             /* Update vectorial force */
 914             fix1             = _mm_add_pd(fix1,tx);
 915             fiy1             = _mm_add_pd(fiy1,ty);
 916             fiz1             = _mm_add_pd(fiz1,tz);
 917
 918             fjx0             = _mm_add_pd(fjx0,tx);
 919             fjy0             = _mm_add_pd(fjy0,ty);
 920             fjz0             = _mm_add_pd(fjz0,tz);
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             r20              = _mm_mul_pd(rsq20,rinv20);
 927
 928             /* Compute parameters for interactions between i and j atoms */
 929             qq20             = _mm_mul_pd(iq2,jq0);
 930
 931             /* Calculate table index by multiplying r with table scale and truncate to integer */
 932             rt               = _mm_mul_pd(r20,vftabscale);
 933             vfitab           = _mm_cvttpd_epi32(rt);
 934             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 935             vfitab           = _mm_slli_epi32(vfitab,2);
 936
 937             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 938             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 939             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 940             GMX_MM_TRANSPOSE2_PD(Y,F);
 941             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 942             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 943             GMX_MM_TRANSPOSE2_PD(G,H);
 944             Heps             = _mm_mul_pd(vfeps,H);
 945             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 946             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 947             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 948
 949             fscal            = felec;
 950
 951             /* Calculate temporary vectorial force */
 952             tx               = _mm_mul_pd(fscal,dx20);
 953             ty               = _mm_mul_pd(fscal,dy20);
 954             tz               = _mm_mul_pd(fscal,dz20);
 955
 956             /* Update vectorial force */
 957             fix2             = _mm_add_pd(fix2,tx);
 958             fiy2             = _mm_add_pd(fiy2,ty);
 959             fiz2             = _mm_add_pd(fiz2,tz);
 960
 961             fjx0             = _mm_add_pd(fjx0,tx);
 962             fjy0             = _mm_add_pd(fjy0,ty);
 963             fjz0             = _mm_add_pd(fjz0,tz);
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             r30              = _mm_mul_pd(rsq30,rinv30);
 970
 971             /* Compute parameters for interactions between i and j atoms */
 972             qq30             = _mm_mul_pd(iq3,jq0);
 973
 974             /* Calculate table index by multiplying r with table scale and truncate to integer */
 975             rt               = _mm_mul_pd(r30,vftabscale);
 976             vfitab           = _mm_cvttpd_epi32(rt);
 977             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 978             vfitab           = _mm_slli_epi32(vfitab,2);
 979
 980             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 981             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 982             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 983             GMX_MM_TRANSPOSE2_PD(Y,F);
 984             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 985             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 986             GMX_MM_TRANSPOSE2_PD(G,H);
 987             Heps             = _mm_mul_pd(vfeps,H);
 988             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 989             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 990             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 991
 992             fscal            = felec;
 993
 994             /* Calculate temporary vectorial force */
 995             tx               = _mm_mul_pd(fscal,dx30);
 996             ty               = _mm_mul_pd(fscal,dy30);
 997             tz               = _mm_mul_pd(fscal,dz30);
 998
 999             /* Update vectorial force */
1000             fix3             = _mm_add_pd(fix3,tx);
1001             fiy3             = _mm_add_pd(fiy3,ty);
1002             fiz3             = _mm_add_pd(fiz3,tz);
1003
1004             fjx0             = _mm_add_pd(fjx0,tx);
1005             fjy0             = _mm_add_pd(fjy0,ty);
1006             fjz0             = _mm_add_pd(fjz0,tz);
1007
1008             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1009
1010             /* Inner loop uses 147 flops */
1011         }
1012
1013         if(jidx<j_index_end)
1014         {
1015
1016             jnrA             = jjnr[jidx];
1017             j_coord_offsetA  = DIM*jnrA;
1018
1019             /* load j atom coordinates */
1020             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1021                                               &jx0,&jy0,&jz0);
1022
1023             /* Calculate displacement vector */
1024             dx00             = _mm_sub_pd(ix0,jx0);
1025             dy00             = _mm_sub_pd(iy0,jy0);
1026             dz00             = _mm_sub_pd(iz0,jz0);
1027             dx10             = _mm_sub_pd(ix1,jx0);
1028             dy10             = _mm_sub_pd(iy1,jy0);
1029             dz10             = _mm_sub_pd(iz1,jz0);
1030             dx20             = _mm_sub_pd(ix2,jx0);
1031             dy20             = _mm_sub_pd(iy2,jy0);
1032             dz20             = _mm_sub_pd(iz2,jz0);
1033             dx30             = _mm_sub_pd(ix3,jx0);
1034             dy30             = _mm_sub_pd(iy3,jy0);
1035             dz30             = _mm_sub_pd(iz3,jz0);
1036
1037             /* Calculate squared distance and things based on it */
1038             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1039             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1040             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1041             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1042
1043             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1044             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1045             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1046
1047             rinvsq00         = gmx_mm_inv_pd(rsq00);
1048
1049             /* Load parameters for j particles */
1050             jq0              = _mm_load_sd(charge+jnrA+0);
1051             vdwjidx0A        = 2*vdwtype[jnrA+0];
1052
1053             fjx0             = _mm_setzero_pd();
1054             fjy0             = _mm_setzero_pd();
1055             fjz0             = _mm_setzero_pd();
1056
1057             /**************************
1058              * CALCULATE INTERACTIONS *
1059              **************************/
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1063
1064             /* LENNARD-JONES DISPERSION/REPULSION */
1065
1066             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1067             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1068
1069             fscal            = fvdw;
1070
1071             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1072
1073             /* Calculate temporary vectorial force */
1074             tx               = _mm_mul_pd(fscal,dx00);
1075             ty               = _mm_mul_pd(fscal,dy00);
1076             tz               = _mm_mul_pd(fscal,dz00);
1077
1078             /* Update vectorial force */
1079             fix0             = _mm_add_pd(fix0,tx);
1080             fiy0             = _mm_add_pd(fiy0,ty);
1081             fiz0             = _mm_add_pd(fiz0,tz);
1082
1083             fjx0             = _mm_add_pd(fjx0,tx);
1084             fjy0             = _mm_add_pd(fjy0,ty);
1085             fjz0             = _mm_add_pd(fjz0,tz);
1086
1087             /**************************
1088              * CALCULATE INTERACTIONS *
1089              **************************/
1090
1091             r10              = _mm_mul_pd(rsq10,rinv10);
1092
1093             /* Compute parameters for interactions between i and j atoms */
1094             qq10             = _mm_mul_pd(iq1,jq0);
1095
1096             /* Calculate table index by multiplying r with table scale and truncate to integer */
1097             rt               = _mm_mul_pd(r10,vftabscale);
1098             vfitab           = _mm_cvttpd_epi32(rt);
1099             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1100             vfitab           = _mm_slli_epi32(vfitab,2);
1101
1102             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1103             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1104             F                = _mm_setzero_pd();
1105             GMX_MM_TRANSPOSE2_PD(Y,F);
1106             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1107             H                = _mm_setzero_pd();
1108             GMX_MM_TRANSPOSE2_PD(G,H);
1109             Heps             = _mm_mul_pd(vfeps,H);
1110             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1111             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1112             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1113
1114             fscal            = felec;
1115
1116             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm_mul_pd(fscal,dx10);
1120             ty               = _mm_mul_pd(fscal,dy10);
1121             tz               = _mm_mul_pd(fscal,dz10);
1122
1123             /* Update vectorial force */
1124             fix1             = _mm_add_pd(fix1,tx);
1125             fiy1             = _mm_add_pd(fiy1,ty);
1126             fiz1             = _mm_add_pd(fiz1,tz);
1127
1128             fjx0             = _mm_add_pd(fjx0,tx);
1129             fjy0             = _mm_add_pd(fjy0,ty);
1130             fjz0             = _mm_add_pd(fjz0,tz);
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             r20              = _mm_mul_pd(rsq20,rinv20);
1137
1138             /* Compute parameters for interactions between i and j atoms */
1139             qq20             = _mm_mul_pd(iq2,jq0);
1140
1141             /* Calculate table index by multiplying r with table scale and truncate to integer */
1142             rt               = _mm_mul_pd(r20,vftabscale);
1143             vfitab           = _mm_cvttpd_epi32(rt);
1144             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1145             vfitab           = _mm_slli_epi32(vfitab,2);
1146
1147             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1148             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1149             F                = _mm_setzero_pd();
1150             GMX_MM_TRANSPOSE2_PD(Y,F);
1151             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1152             H                = _mm_setzero_pd();
1153             GMX_MM_TRANSPOSE2_PD(G,H);
1154             Heps             = _mm_mul_pd(vfeps,H);
1155             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1156             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1157             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1158
1159             fscal            = felec;
1160
1161             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1162
1163             /* Calculate temporary vectorial force */
1164             tx               = _mm_mul_pd(fscal,dx20);
1165             ty               = _mm_mul_pd(fscal,dy20);
1166             tz               = _mm_mul_pd(fscal,dz20);
1167
1168             /* Update vectorial force */
1169             fix2             = _mm_add_pd(fix2,tx);
1170             fiy2             = _mm_add_pd(fiy2,ty);
1171             fiz2             = _mm_add_pd(fiz2,tz);
1172
1173             fjx0             = _mm_add_pd(fjx0,tx);
1174             fjy0             = _mm_add_pd(fjy0,ty);
1175             fjz0             = _mm_add_pd(fjz0,tz);
1176
1177             /**************************
1178              * CALCULATE INTERACTIONS *
1179              **************************/
1180
1181             r30              = _mm_mul_pd(rsq30,rinv30);
1182
1183             /* Compute parameters for interactions between i and j atoms */
1184             qq30             = _mm_mul_pd(iq3,jq0);
1185
1186             /* Calculate table index by multiplying r with table scale and truncate to integer */
1187             rt               = _mm_mul_pd(r30,vftabscale);
1188             vfitab           = _mm_cvttpd_epi32(rt);
1189             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1190             vfitab           = _mm_slli_epi32(vfitab,2);
1191
1192             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1193             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1194             F                = _mm_setzero_pd();
1195             GMX_MM_TRANSPOSE2_PD(Y,F);
1196             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1197             H                = _mm_setzero_pd();
1198             GMX_MM_TRANSPOSE2_PD(G,H);
1199             Heps             = _mm_mul_pd(vfeps,H);
1200             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1201             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1202             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1203
1204             fscal            = felec;
1205
1206             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1207
1208             /* Calculate temporary vectorial force */
1209             tx               = _mm_mul_pd(fscal,dx30);
1210             ty               = _mm_mul_pd(fscal,dy30);
1211             tz               = _mm_mul_pd(fscal,dz30);
1212
1213             /* Update vectorial force */
1214             fix3             = _mm_add_pd(fix3,tx);
1215             fiy3             = _mm_add_pd(fiy3,ty);
1216             fiz3             = _mm_add_pd(fiz3,tz);
1217
1218             fjx0             = _mm_add_pd(fjx0,tx);
1219             fjy0             = _mm_add_pd(fjy0,ty);
1220             fjz0             = _mm_add_pd(fjz0,tz);
1221
1222             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1223
1224             /* Inner loop uses 147 flops */
1225         }
1226
1227         /* End of innermost loop */
1228
1229         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1230                                               f+i_coord_offset,fshift+i_shift_offset);
1231
1232         /* Increment number of inner iterations */
1233         inneriter                  += j_index_end - j_index_start;
1234
1235         /* Outer loop uses 24 flops */
1236     }
1237
1238     /* Increment number of outer iterations */
1239     outeriter        += nri;
1240
1241     /* Update outer/inner flops */
1242
1243     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1244 }