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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse2_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse2_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix0             = _mm_setzero_pd();
 166         fiy0             = _mm_setzero_pd();
 167         fiz0             = _mm_setzero_pd();
 168         fix1             = _mm_setzero_pd();
 169         fiy1             = _mm_setzero_pd();
 170         fiz1             = _mm_setzero_pd();
 171         fix2             = _mm_setzero_pd();
 172         fiy2             = _mm_setzero_pd();
 173         fiz2             = _mm_setzero_pd();
 174         fix3             = _mm_setzero_pd();
 175         fiy3             = _mm_setzero_pd();
 176         fiz3             = _mm_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_pd();
 180         vvdwsum          = _mm_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206             dx30             = _mm_sub_pd(ix3,jx0);
 207             dy30             = _mm_sub_pd(iy3,jy0);
 208             dz30             = _mm_sub_pd(iz3,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 212             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 213             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 214             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 215
 216             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 217             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 218             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 219
 220             rinvsq00         = gmx_mm_inv_pd(rsq00);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226
 227             fjx0             = _mm_setzero_pd();
 228             fjy0             = _mm_setzero_pd();
 229             fjz0             = _mm_setzero_pd();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 237                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 238
 239             /* LENNARD-JONES DISPERSION/REPULSION */
 240
 241             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 242             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 243             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 244             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 245             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 249
 250             fscal            = fvdw;
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm_mul_pd(fscal,dx00);
 254             ty               = _mm_mul_pd(fscal,dy00);
 255             tz               = _mm_mul_pd(fscal,dz00);
 256
 257             /* Update vectorial force */
 258             fix0             = _mm_add_pd(fix0,tx);
 259             fiy0             = _mm_add_pd(fiy0,ty);
 260             fiz0             = _mm_add_pd(fiz0,tz);
 261
 262             fjx0             = _mm_add_pd(fjx0,tx);
 263             fjy0             = _mm_add_pd(fjy0,ty);
 264             fjz0             = _mm_add_pd(fjz0,tz);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             r10              = _mm_mul_pd(rsq10,rinv10);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq10             = _mm_mul_pd(iq1,jq0);
 274
 275             /* Calculate table index by multiplying r with table scale and truncate to integer */
 276             rt               = _mm_mul_pd(r10,vftabscale);
 277             vfitab           = _mm_cvttpd_epi32(rt);
 278             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 279             vfitab           = _mm_slli_epi32(vfitab,2);
 280
 281             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 282             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 283             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 284             GMX_MM_TRANSPOSE2_PD(Y,F);
 285             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 286             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 287             GMX_MM_TRANSPOSE2_PD(G,H);
 288             Heps             = _mm_mul_pd(vfeps,H);
 289             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 290             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 291             velec            = _mm_mul_pd(qq10,VV);
 292             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 293             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_pd(velecsum,velec);
 297
 298             fscal            = felec;
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_pd(fscal,dx10);
 302             ty               = _mm_mul_pd(fscal,dy10);
 303             tz               = _mm_mul_pd(fscal,dz10);
 304
 305             /* Update vectorial force */
 306             fix1             = _mm_add_pd(fix1,tx);
 307             fiy1             = _mm_add_pd(fiy1,ty);
 308             fiz1             = _mm_add_pd(fiz1,tz);
 309
 310             fjx0             = _mm_add_pd(fjx0,tx);
 311             fjy0             = _mm_add_pd(fjy0,ty);
 312             fjz0             = _mm_add_pd(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r20              = _mm_mul_pd(rsq20,rinv20);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq20             = _mm_mul_pd(iq2,jq0);
 322
 323             /* Calculate table index by multiplying r with table scale and truncate to integer */
 324             rt               = _mm_mul_pd(r20,vftabscale);
 325             vfitab           = _mm_cvttpd_epi32(rt);
 326             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 327             vfitab           = _mm_slli_epi32(vfitab,2);
 328
 329             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 330             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 331             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 332             GMX_MM_TRANSPOSE2_PD(Y,F);
 333             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 334             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 335             GMX_MM_TRANSPOSE2_PD(G,H);
 336             Heps             = _mm_mul_pd(vfeps,H);
 337             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 338             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 339             velec            = _mm_mul_pd(qq20,VV);
 340             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 341             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_pd(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_pd(fscal,dx20);
 350             ty               = _mm_mul_pd(fscal,dy20);
 351             tz               = _mm_mul_pd(fscal,dz20);
 352
 353             /* Update vectorial force */
 354             fix2             = _mm_add_pd(fix2,tx);
 355             fiy2             = _mm_add_pd(fiy2,ty);
 356             fiz2             = _mm_add_pd(fiz2,tz);
 357
 358             fjx0             = _mm_add_pd(fjx0,tx);
 359             fjy0             = _mm_add_pd(fjy0,ty);
 360             fjz0             = _mm_add_pd(fjz0,tz);
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             r30              = _mm_mul_pd(rsq30,rinv30);
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq30             = _mm_mul_pd(iq3,jq0);
 370
 371             /* Calculate table index by multiplying r with table scale and truncate to integer */
 372             rt               = _mm_mul_pd(r30,vftabscale);
 373             vfitab           = _mm_cvttpd_epi32(rt);
 374             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 375             vfitab           = _mm_slli_epi32(vfitab,2);
 376
 377             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 378             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 379             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 380             GMX_MM_TRANSPOSE2_PD(Y,F);
 381             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 382             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 383             GMX_MM_TRANSPOSE2_PD(G,H);
 384             Heps             = _mm_mul_pd(vfeps,H);
 385             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 386             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 387             velec            = _mm_mul_pd(qq30,VV);
 388             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 389             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm_add_pd(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm_mul_pd(fscal,dx30);
 398             ty               = _mm_mul_pd(fscal,dy30);
 399             tz               = _mm_mul_pd(fscal,dz30);
 400
 401             /* Update vectorial force */
 402             fix3             = _mm_add_pd(fix3,tx);
 403             fiy3             = _mm_add_pd(fiy3,ty);
 404             fiz3             = _mm_add_pd(fiz3,tz);
 405
 406             fjx0             = _mm_add_pd(fjx0,tx);
 407             fjy0             = _mm_add_pd(fjy0,ty);
 408             fjz0             = _mm_add_pd(fjz0,tz);
 409
 410             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 411
 412             /* Inner loop uses 164 flops */
 413         }
 414
 415         if(jidx<j_index_end)
 416         {
 417
 418             jnrA             = jjnr[jidx];
 419             j_coord_offsetA  = DIM*jnrA;
 420
 421             /* load j atom coordinates */
 422             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 423                                               &jx0,&jy0,&jz0);
 424
 425             /* Calculate displacement vector */
 426             dx00             = _mm_sub_pd(ix0,jx0);
 427             dy00             = _mm_sub_pd(iy0,jy0);
 428             dz00             = _mm_sub_pd(iz0,jz0);
 429             dx10             = _mm_sub_pd(ix1,jx0);
 430             dy10             = _mm_sub_pd(iy1,jy0);
 431             dz10             = _mm_sub_pd(iz1,jz0);
 432             dx20             = _mm_sub_pd(ix2,jx0);
 433             dy20             = _mm_sub_pd(iy2,jy0);
 434             dz20             = _mm_sub_pd(iz2,jz0);
 435             dx30             = _mm_sub_pd(ix3,jx0);
 436             dy30             = _mm_sub_pd(iy3,jy0);
 437             dz30             = _mm_sub_pd(iz3,jz0);
 438
 439             /* Calculate squared distance and things based on it */
 440             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 441             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 442             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 443             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 444
 445             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 446             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 447             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 448
 449             rinvsq00         = gmx_mm_inv_pd(rsq00);
 450
 451             /* Load parameters for j particles */
 452             jq0              = _mm_load_sd(charge+jnrA+0);
 453             vdwjidx0A        = 2*vdwtype[jnrA+0];
 454
 455             fjx0             = _mm_setzero_pd();
 456             fjy0             = _mm_setzero_pd();
 457             fjz0             = _mm_setzero_pd();
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             /* Compute parameters for interactions between i and j atoms */
 464             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 465
 466             /* LENNARD-JONES DISPERSION/REPULSION */
 467
 468             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 469             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 470             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 471             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 472             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 476             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 477
 478             fscal            = fvdw;
 479
 480             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 481
 482             /* Calculate temporary vectorial force */
 483             tx               = _mm_mul_pd(fscal,dx00);
 484             ty               = _mm_mul_pd(fscal,dy00);
 485             tz               = _mm_mul_pd(fscal,dz00);
 486
 487             /* Update vectorial force */
 488             fix0             = _mm_add_pd(fix0,tx);
 489             fiy0             = _mm_add_pd(fiy0,ty);
 490             fiz0             = _mm_add_pd(fiz0,tz);
 491
 492             fjx0             = _mm_add_pd(fjx0,tx);
 493             fjy0             = _mm_add_pd(fjy0,ty);
 494             fjz0             = _mm_add_pd(fjz0,tz);
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             r10              = _mm_mul_pd(rsq10,rinv10);
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq10             = _mm_mul_pd(iq1,jq0);
 504
 505             /* Calculate table index by multiplying r with table scale and truncate to integer */
 506             rt               = _mm_mul_pd(r10,vftabscale);
 507             vfitab           = _mm_cvttpd_epi32(rt);
 508             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 509             vfitab           = _mm_slli_epi32(vfitab,2);
 510
 511             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 512             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 513             F                = _mm_setzero_pd();
 514             GMX_MM_TRANSPOSE2_PD(Y,F);
 515             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 516             H                = _mm_setzero_pd();
 517             GMX_MM_TRANSPOSE2_PD(G,H);
 518             Heps             = _mm_mul_pd(vfeps,H);
 519             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 520             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 521             velec            = _mm_mul_pd(qq10,VV);
 522             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 523             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 524
 525             /* Update potential sum for this i atom from the interaction with this j atom. */
 526             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 527             velecsum         = _mm_add_pd(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 532
 533             /* Calculate temporary vectorial force */
 534             tx               = _mm_mul_pd(fscal,dx10);
 535             ty               = _mm_mul_pd(fscal,dy10);
 536             tz               = _mm_mul_pd(fscal,dz10);
 537
 538             /* Update vectorial force */
 539             fix1             = _mm_add_pd(fix1,tx);
 540             fiy1             = _mm_add_pd(fiy1,ty);
 541             fiz1             = _mm_add_pd(fiz1,tz);
 542
 543             fjx0             = _mm_add_pd(fjx0,tx);
 544             fjy0             = _mm_add_pd(fjy0,ty);
 545             fjz0             = _mm_add_pd(fjz0,tz);
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             r20              = _mm_mul_pd(rsq20,rinv20);
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq20             = _mm_mul_pd(iq2,jq0);
 555
 556             /* Calculate table index by multiplying r with table scale and truncate to integer */
 557             rt               = _mm_mul_pd(r20,vftabscale);
 558             vfitab           = _mm_cvttpd_epi32(rt);
 559             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 560             vfitab           = _mm_slli_epi32(vfitab,2);
 561
 562             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 563             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 564             F                = _mm_setzero_pd();
 565             GMX_MM_TRANSPOSE2_PD(Y,F);
 566             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 567             H                = _mm_setzero_pd();
 568             GMX_MM_TRANSPOSE2_PD(G,H);
 569             Heps             = _mm_mul_pd(vfeps,H);
 570             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 571             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 572             velec            = _mm_mul_pd(qq20,VV);
 573             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 574             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 578             velecsum         = _mm_add_pd(velecsum,velec);
 579
 580             fscal            = felec;
 581
 582             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm_mul_pd(fscal,dx20);
 586             ty               = _mm_mul_pd(fscal,dy20);
 587             tz               = _mm_mul_pd(fscal,dz20);
 588
 589             /* Update vectorial force */
 590             fix2             = _mm_add_pd(fix2,tx);
 591             fiy2             = _mm_add_pd(fiy2,ty);
 592             fiz2             = _mm_add_pd(fiz2,tz);
 593
 594             fjx0             = _mm_add_pd(fjx0,tx);
 595             fjy0             = _mm_add_pd(fjy0,ty);
 596             fjz0             = _mm_add_pd(fjz0,tz);
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r30              = _mm_mul_pd(rsq30,rinv30);
 603
 604             /* Compute parameters for interactions between i and j atoms */
 605             qq30             = _mm_mul_pd(iq3,jq0);
 606
 607             /* Calculate table index by multiplying r with table scale and truncate to integer */
 608             rt               = _mm_mul_pd(r30,vftabscale);
 609             vfitab           = _mm_cvttpd_epi32(rt);
 610             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 611             vfitab           = _mm_slli_epi32(vfitab,2);
 612
 613             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 614             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 615             F                = _mm_setzero_pd();
 616             GMX_MM_TRANSPOSE2_PD(Y,F);
 617             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 618             H                = _mm_setzero_pd();
 619             GMX_MM_TRANSPOSE2_PD(G,H);
 620             Heps             = _mm_mul_pd(vfeps,H);
 621             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 622             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 623             velec            = _mm_mul_pd(qq30,VV);
 624             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 625             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 629             velecsum         = _mm_add_pd(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm_mul_pd(fscal,dx30);
 637             ty               = _mm_mul_pd(fscal,dy30);
 638             tz               = _mm_mul_pd(fscal,dz30);
 639
 640             /* Update vectorial force */
 641             fix3             = _mm_add_pd(fix3,tx);
 642             fiy3             = _mm_add_pd(fiy3,ty);
 643             fiz3             = _mm_add_pd(fiz3,tz);
 644
 645             fjx0             = _mm_add_pd(fjx0,tx);
 646             fjy0             = _mm_add_pd(fjy0,ty);
 647             fjz0             = _mm_add_pd(fjz0,tz);
 648
 649             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 650
 651             /* Inner loop uses 164 flops */
 652         }
 653
 654         /* End of innermost loop */
 655
 656         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 657                                               f+i_coord_offset,fshift+i_shift_offset);
 658
 659         ggid                        = gid[iidx];
 660         /* Update potential energies */
 661         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 662         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 663
 664         /* Increment number of inner iterations */
 665         inneriter                  += j_index_end - j_index_start;
 666
 667         /* Outer loop uses 26 flops */
 668     }
 669
 670     /* Increment number of outer iterations */
 671     outeriter        += nri;
 672
 673     /* Update outer/inner flops */
 674
 675     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
 676 }
 677 /*
 678  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_double
 679  * Electrostatics interaction: CubicSplineTable
 680  * VdW interaction:            LennardJones
 681  * Geometry:                   Water4-Particle
 682  * Calculate force/pot:        Force
 683  */
 684 void
 685 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse2_double
 686                     (t_nblist                    * gmx_restrict       nlist,
 687                      rvec                        * gmx_restrict          xx,
 688                      rvec                        * gmx_restrict          ff,
 689                      t_forcerec                  * gmx_restrict          fr,
 690                      t_mdatoms                   * gmx_restrict     mdatoms,
 691                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 692                      t_nrnb                      * gmx_restrict        nrnb)
 693 {
 694     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 695      * just 0 for non-waters.
 696      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 697      * jnr indices corresponding to data put in the four positions in the SIMD register.
 698      */
 699     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 700     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 701     int              jnrA,jnrB;
 702     int              j_coord_offsetA,j_coord_offsetB;
 703     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 704     real             rcutoff_scalar;
 705     real             *shiftvec,*fshift,*x,*f;
 706     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 707     int              vdwioffset0;
 708     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 709     int              vdwioffset1;
 710     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 711     int              vdwioffset2;
 712     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 713     int              vdwioffset3;
 714     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 715     int              vdwjidx0A,vdwjidx0B;
 716     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 717     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 718     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 719     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 720     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 721     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 722     real             *charge;
 723     int              nvdwtype;
 724     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 725     int              *vdwtype;
 726     real             *vdwparam;
 727     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 728     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 729     __m128i          vfitab;
 730     __m128i          ifour       = _mm_set1_epi32(4);
 731     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 732     real             *vftab;
 733     __m128d          dummy_mask,cutoff_mask;
 734     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 735     __m128d          one     = _mm_set1_pd(1.0);
 736     __m128d          two     = _mm_set1_pd(2.0);
 737     x                = xx[0];
 738     f                = ff[0];
 739
 740     nri              = nlist->nri;
 741     iinr             = nlist->iinr;
 742     jindex           = nlist->jindex;
 743     jjnr             = nlist->jjnr;
 744     shiftidx         = nlist->shift;
 745     gid              = nlist->gid;
 746     shiftvec         = fr->shift_vec[0];
 747     fshift           = fr->fshift[0];
 748     facel            = _mm_set1_pd(fr->epsfac);
 749     charge           = mdatoms->chargeA;
 750     nvdwtype         = fr->ntype;
 751     vdwparam         = fr->nbfp;
 752     vdwtype          = mdatoms->typeA;
 753
 754     vftab            = kernel_data->table_elec->data;
 755     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 756
 757     /* Setup water-specific parameters */
 758     inr              = nlist->iinr[0];
 759     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 760     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 761     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 762     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 763
 764     /* Avoid stupid compiler warnings */
 765     jnrA = jnrB = 0;
 766     j_coord_offsetA = 0;
 767     j_coord_offsetB = 0;
 768
 769     outeriter        = 0;
 770     inneriter        = 0;
 771
 772     /* Start outer loop over neighborlists */
 773     for(iidx=0; iidx<nri; iidx++)
 774     {
 775         /* Load shift vector for this list */
 776         i_shift_offset   = DIM*shiftidx[iidx];
 777
 778         /* Load limits for loop over neighbors */
 779         j_index_start    = jindex[iidx];
 780         j_index_end      = jindex[iidx+1];
 781
 782         /* Get outer coordinate index */
 783         inr              = iinr[iidx];
 784         i_coord_offset   = DIM*inr;
 785
 786         /* Load i particle coords and add shift vector */
 787         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 788                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 789
 790         fix0             = _mm_setzero_pd();
 791         fiy0             = _mm_setzero_pd();
 792         fiz0             = _mm_setzero_pd();
 793         fix1             = _mm_setzero_pd();
 794         fiy1             = _mm_setzero_pd();
 795         fiz1             = _mm_setzero_pd();
 796         fix2             = _mm_setzero_pd();
 797         fiy2             = _mm_setzero_pd();
 798         fiz2             = _mm_setzero_pd();
 799         fix3             = _mm_setzero_pd();
 800         fiy3             = _mm_setzero_pd();
 801         fiz3             = _mm_setzero_pd();
 802
 803         /* Start inner kernel loop */
 804         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 805         {
 806
 807             /* Get j neighbor index, and coordinate index */
 808             jnrA             = jjnr[jidx];
 809             jnrB             = jjnr[jidx+1];
 810             j_coord_offsetA  = DIM*jnrA;
 811             j_coord_offsetB  = DIM*jnrB;
 812
 813             /* load j atom coordinates */
 814             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 815                                               &jx0,&jy0,&jz0);
 816
 817             /* Calculate displacement vector */
 818             dx00             = _mm_sub_pd(ix0,jx0);
 819             dy00             = _mm_sub_pd(iy0,jy0);
 820             dz00             = _mm_sub_pd(iz0,jz0);
 821             dx10             = _mm_sub_pd(ix1,jx0);
 822             dy10             = _mm_sub_pd(iy1,jy0);
 823             dz10             = _mm_sub_pd(iz1,jz0);
 824             dx20             = _mm_sub_pd(ix2,jx0);
 825             dy20             = _mm_sub_pd(iy2,jy0);
 826             dz20             = _mm_sub_pd(iz2,jz0);
 827             dx30             = _mm_sub_pd(ix3,jx0);
 828             dy30             = _mm_sub_pd(iy3,jy0);
 829             dz30             = _mm_sub_pd(iz3,jz0);
 830
 831             /* Calculate squared distance and things based on it */
 832             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 833             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 834             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 835             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 836
 837             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 838             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 839             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 840
 841             rinvsq00         = gmx_mm_inv_pd(rsq00);
 842
 843             /* Load parameters for j particles */
 844             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 845             vdwjidx0A        = 2*vdwtype[jnrA+0];
 846             vdwjidx0B        = 2*vdwtype[jnrB+0];
 847
 848             fjx0             = _mm_setzero_pd();
 849             fjy0             = _mm_setzero_pd();
 850             fjz0             = _mm_setzero_pd();
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 858                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 859
 860             /* LENNARD-JONES DISPERSION/REPULSION */
 861
 862             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 863             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 864
 865             fscal            = fvdw;
 866
 867             /* Calculate temporary vectorial force */
 868             tx               = _mm_mul_pd(fscal,dx00);
 869             ty               = _mm_mul_pd(fscal,dy00);
 870             tz               = _mm_mul_pd(fscal,dz00);
 871
 872             /* Update vectorial force */
 873             fix0             = _mm_add_pd(fix0,tx);
 874             fiy0             = _mm_add_pd(fiy0,ty);
 875             fiz0             = _mm_add_pd(fiz0,tz);
 876
 877             fjx0             = _mm_add_pd(fjx0,tx);
 878             fjy0             = _mm_add_pd(fjy0,ty);
 879             fjz0             = _mm_add_pd(fjz0,tz);
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             r10              = _mm_mul_pd(rsq10,rinv10);
 886
 887             /* Compute parameters for interactions between i and j atoms */
 888             qq10             = _mm_mul_pd(iq1,jq0);
 889
 890             /* Calculate table index by multiplying r with table scale and truncate to integer */
 891             rt               = _mm_mul_pd(r10,vftabscale);
 892             vfitab           = _mm_cvttpd_epi32(rt);
 893             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 894             vfitab           = _mm_slli_epi32(vfitab,2);
 895
 896             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 897             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 898             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 899             GMX_MM_TRANSPOSE2_PD(Y,F);
 900             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 901             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 902             GMX_MM_TRANSPOSE2_PD(G,H);
 903             Heps             = _mm_mul_pd(vfeps,H);
 904             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 905             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 906             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 907
 908             fscal            = felec;
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm_mul_pd(fscal,dx10);
 912             ty               = _mm_mul_pd(fscal,dy10);
 913             tz               = _mm_mul_pd(fscal,dz10);
 914
 915             /* Update vectorial force */
 916             fix1             = _mm_add_pd(fix1,tx);
 917             fiy1             = _mm_add_pd(fiy1,ty);
 918             fiz1             = _mm_add_pd(fiz1,tz);
 919
 920             fjx0             = _mm_add_pd(fjx0,tx);
 921             fjy0             = _mm_add_pd(fjy0,ty);
 922             fjz0             = _mm_add_pd(fjz0,tz);
 923
 924             /**************************
 925              * CALCULATE INTERACTIONS *
 926              **************************/
 927
 928             r20              = _mm_mul_pd(rsq20,rinv20);
 929
 930             /* Compute parameters for interactions between i and j atoms */
 931             qq20             = _mm_mul_pd(iq2,jq0);
 932
 933             /* Calculate table index by multiplying r with table scale and truncate to integer */
 934             rt               = _mm_mul_pd(r20,vftabscale);
 935             vfitab           = _mm_cvttpd_epi32(rt);
 936             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 937             vfitab           = _mm_slli_epi32(vfitab,2);
 938
 939             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 940             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 941             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 942             GMX_MM_TRANSPOSE2_PD(Y,F);
 943             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 944             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 945             GMX_MM_TRANSPOSE2_PD(G,H);
 946             Heps             = _mm_mul_pd(vfeps,H);
 947             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 948             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 949             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 950
 951             fscal            = felec;
 952
 953             /* Calculate temporary vectorial force */
 954             tx               = _mm_mul_pd(fscal,dx20);
 955             ty               = _mm_mul_pd(fscal,dy20);
 956             tz               = _mm_mul_pd(fscal,dz20);
 957
 958             /* Update vectorial force */
 959             fix2             = _mm_add_pd(fix2,tx);
 960             fiy2             = _mm_add_pd(fiy2,ty);
 961             fiz2             = _mm_add_pd(fiz2,tz);
 962
 963             fjx0             = _mm_add_pd(fjx0,tx);
 964             fjy0             = _mm_add_pd(fjy0,ty);
 965             fjz0             = _mm_add_pd(fjz0,tz);
 966
 967             /**************************
 968              * CALCULATE INTERACTIONS *
 969              **************************/
 970
 971             r30              = _mm_mul_pd(rsq30,rinv30);
 972
 973             /* Compute parameters for interactions between i and j atoms */
 974             qq30             = _mm_mul_pd(iq3,jq0);
 975
 976             /* Calculate table index by multiplying r with table scale and truncate to integer */
 977             rt               = _mm_mul_pd(r30,vftabscale);
 978             vfitab           = _mm_cvttpd_epi32(rt);
 979             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 980             vfitab           = _mm_slli_epi32(vfitab,2);
 981
 982             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 983             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 984             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 985             GMX_MM_TRANSPOSE2_PD(Y,F);
 986             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 987             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 988             GMX_MM_TRANSPOSE2_PD(G,H);
 989             Heps             = _mm_mul_pd(vfeps,H);
 990             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 991             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 992             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 993
 994             fscal            = felec;
 995
 996             /* Calculate temporary vectorial force */
 997             tx               = _mm_mul_pd(fscal,dx30);
 998             ty               = _mm_mul_pd(fscal,dy30);
 999             tz               = _mm_mul_pd(fscal,dz30);
1000
1001             /* Update vectorial force */
1002             fix3             = _mm_add_pd(fix3,tx);
1003             fiy3             = _mm_add_pd(fiy3,ty);
1004             fiz3             = _mm_add_pd(fiz3,tz);
1005
1006             fjx0             = _mm_add_pd(fjx0,tx);
1007             fjy0             = _mm_add_pd(fjy0,ty);
1008             fjz0             = _mm_add_pd(fjz0,tz);
1009
1010             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1011
1012             /* Inner loop uses 147 flops */
1013         }
1014
1015         if(jidx<j_index_end)
1016         {
1017
1018             jnrA             = jjnr[jidx];
1019             j_coord_offsetA  = DIM*jnrA;
1020
1021             /* load j atom coordinates */
1022             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1023                                               &jx0,&jy0,&jz0);
1024
1025             /* Calculate displacement vector */
1026             dx00             = _mm_sub_pd(ix0,jx0);
1027             dy00             = _mm_sub_pd(iy0,jy0);
1028             dz00             = _mm_sub_pd(iz0,jz0);
1029             dx10             = _mm_sub_pd(ix1,jx0);
1030             dy10             = _mm_sub_pd(iy1,jy0);
1031             dz10             = _mm_sub_pd(iz1,jz0);
1032             dx20             = _mm_sub_pd(ix2,jx0);
1033             dy20             = _mm_sub_pd(iy2,jy0);
1034             dz20             = _mm_sub_pd(iz2,jz0);
1035             dx30             = _mm_sub_pd(ix3,jx0);
1036             dy30             = _mm_sub_pd(iy3,jy0);
1037             dz30             = _mm_sub_pd(iz3,jz0);
1038
1039             /* Calculate squared distance and things based on it */
1040             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1041             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1042             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1043             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1044
1045             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1046             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1047             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1048
1049             rinvsq00         = gmx_mm_inv_pd(rsq00);
1050
1051             /* Load parameters for j particles */
1052             jq0              = _mm_load_sd(charge+jnrA+0);
1053             vdwjidx0A        = 2*vdwtype[jnrA+0];
1054
1055             fjx0             = _mm_setzero_pd();
1056             fjy0             = _mm_setzero_pd();
1057             fjz0             = _mm_setzero_pd();
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1065
1066             /* LENNARD-JONES DISPERSION/REPULSION */
1067
1068             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1069             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1070
1071             fscal            = fvdw;
1072
1073             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1074
1075             /* Calculate temporary vectorial force */
1076             tx               = _mm_mul_pd(fscal,dx00);
1077             ty               = _mm_mul_pd(fscal,dy00);
1078             tz               = _mm_mul_pd(fscal,dz00);
1079
1080             /* Update vectorial force */
1081             fix0             = _mm_add_pd(fix0,tx);
1082             fiy0             = _mm_add_pd(fiy0,ty);
1083             fiz0             = _mm_add_pd(fiz0,tz);
1084
1085             fjx0             = _mm_add_pd(fjx0,tx);
1086             fjy0             = _mm_add_pd(fjy0,ty);
1087             fjz0             = _mm_add_pd(fjz0,tz);
1088
1089             /**************************
1090              * CALCULATE INTERACTIONS *
1091              **************************/
1092
1093             r10              = _mm_mul_pd(rsq10,rinv10);
1094
1095             /* Compute parameters for interactions between i and j atoms */
1096             qq10             = _mm_mul_pd(iq1,jq0);
1097
1098             /* Calculate table index by multiplying r with table scale and truncate to integer */
1099             rt               = _mm_mul_pd(r10,vftabscale);
1100             vfitab           = _mm_cvttpd_epi32(rt);
1101             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1102             vfitab           = _mm_slli_epi32(vfitab,2);
1103
1104             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1105             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1106             F                = _mm_setzero_pd();
1107             GMX_MM_TRANSPOSE2_PD(Y,F);
1108             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1109             H                = _mm_setzero_pd();
1110             GMX_MM_TRANSPOSE2_PD(G,H);
1111             Heps             = _mm_mul_pd(vfeps,H);
1112             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1113             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1114             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1115
1116             fscal            = felec;
1117
1118             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1119
1120             /* Calculate temporary vectorial force */
1121             tx               = _mm_mul_pd(fscal,dx10);
1122             ty               = _mm_mul_pd(fscal,dy10);
1123             tz               = _mm_mul_pd(fscal,dz10);
1124
1125             /* Update vectorial force */
1126             fix1             = _mm_add_pd(fix1,tx);
1127             fiy1             = _mm_add_pd(fiy1,ty);
1128             fiz1             = _mm_add_pd(fiz1,tz);
1129
1130             fjx0             = _mm_add_pd(fjx0,tx);
1131             fjy0             = _mm_add_pd(fjy0,ty);
1132             fjz0             = _mm_add_pd(fjz0,tz);
1133
1134             /**************************
1135              * CALCULATE INTERACTIONS *
1136              **************************/
1137
1138             r20              = _mm_mul_pd(rsq20,rinv20);
1139
1140             /* Compute parameters for interactions between i and j atoms */
1141             qq20             = _mm_mul_pd(iq2,jq0);
1142
1143             /* Calculate table index by multiplying r with table scale and truncate to integer */
1144             rt               = _mm_mul_pd(r20,vftabscale);
1145             vfitab           = _mm_cvttpd_epi32(rt);
1146             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1147             vfitab           = _mm_slli_epi32(vfitab,2);
1148
1149             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1150             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1151             F                = _mm_setzero_pd();
1152             GMX_MM_TRANSPOSE2_PD(Y,F);
1153             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1154             H                = _mm_setzero_pd();
1155             GMX_MM_TRANSPOSE2_PD(G,H);
1156             Heps             = _mm_mul_pd(vfeps,H);
1157             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1158             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1159             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1164
1165             /* Calculate temporary vectorial force */
1166             tx               = _mm_mul_pd(fscal,dx20);
1167             ty               = _mm_mul_pd(fscal,dy20);
1168             tz               = _mm_mul_pd(fscal,dz20);
1169
1170             /* Update vectorial force */
1171             fix2             = _mm_add_pd(fix2,tx);
1172             fiy2             = _mm_add_pd(fiy2,ty);
1173             fiz2             = _mm_add_pd(fiz2,tz);
1174
1175             fjx0             = _mm_add_pd(fjx0,tx);
1176             fjy0             = _mm_add_pd(fjy0,ty);
1177             fjz0             = _mm_add_pd(fjz0,tz);
1178
1179             /**************************
1180              * CALCULATE INTERACTIONS *
1181              **************************/
1182
1183             r30              = _mm_mul_pd(rsq30,rinv30);
1184
1185             /* Compute parameters for interactions between i and j atoms */
1186             qq30             = _mm_mul_pd(iq3,jq0);
1187
1188             /* Calculate table index by multiplying r with table scale and truncate to integer */
1189             rt               = _mm_mul_pd(r30,vftabscale);
1190             vfitab           = _mm_cvttpd_epi32(rt);
1191             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1192             vfitab           = _mm_slli_epi32(vfitab,2);
1193
1194             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1195             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1196             F                = _mm_setzero_pd();
1197             GMX_MM_TRANSPOSE2_PD(Y,F);
1198             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1199             H                = _mm_setzero_pd();
1200             GMX_MM_TRANSPOSE2_PD(G,H);
1201             Heps             = _mm_mul_pd(vfeps,H);
1202             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1203             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1204             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1205
1206             fscal            = felec;
1207
1208             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1209
1210             /* Calculate temporary vectorial force */
1211             tx               = _mm_mul_pd(fscal,dx30);
1212             ty               = _mm_mul_pd(fscal,dy30);
1213             tz               = _mm_mul_pd(fscal,dz30);
1214
1215             /* Update vectorial force */
1216             fix3             = _mm_add_pd(fix3,tx);
1217             fiy3             = _mm_add_pd(fiy3,ty);
1218             fiz3             = _mm_add_pd(fiz3,tz);
1219
1220             fjx0             = _mm_add_pd(fjx0,tx);
1221             fjy0             = _mm_add_pd(fjy0,ty);
1222             fjz0             = _mm_add_pd(fjz0,tz);
1223
1224             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1225
1226             /* Inner loop uses 147 flops */
1227         }
1228
1229         /* End of innermost loop */
1230
1231         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1232                                               f+i_coord_offset,fshift+i_shift_offset);
1233
1234         /* Increment number of inner iterations */
1235         inneriter                  += j_index_end - j_index_start;
1236
1237         /* Outer loop uses 24 flops */
1238     }
1239
1240     /* Increment number of outer iterations */
1241     outeriter        += nri;
1242
1243     /* Update outer/inner flops */
1244
1245     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1246 }