src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_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_GeomW3P1_VF_sse2_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_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              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          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_elec->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+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     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 161
 162         fix0             = _mm_setzero_pd();
 163         fiy0             = _mm_setzero_pd();
 164         fiz0             = _mm_setzero_pd();
 165         fix1             = _mm_setzero_pd();
 166         fiy1             = _mm_setzero_pd();
 167         fiz1             = _mm_setzero_pd();
 168         fix2             = _mm_setzero_pd();
 169         fiy2             = _mm_setzero_pd();
 170         fiz2             = _mm_setzero_pd();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_pd();
 174         vvdwsum          = _mm_setzero_pd();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_pd(ix0,jx0);
 192             dy00             = _mm_sub_pd(iy0,jy0);
 193             dz00             = _mm_sub_pd(iz0,jz0);
 194             dx10             = _mm_sub_pd(ix1,jx0);
 195             dy10             = _mm_sub_pd(iy1,jy0);
 196             dz10             = _mm_sub_pd(iz1,jz0);
 197             dx20             = _mm_sub_pd(ix2,jx0);
 198             dy20             = _mm_sub_pd(iy2,jy0);
 199             dz20             = _mm_sub_pd(iz2,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 203             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 204             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 205
 206             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 207             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 208             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 209
 210             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216
 217             fjx0             = _mm_setzero_pd();
 218             fjy0             = _mm_setzero_pd();
 219             fjz0             = _mm_setzero_pd();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_pd(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_pd(iq0,jq0);
 229             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 231
 232             /* Calculate table index by multiplying r with table scale and truncate to integer */
 233             rt               = _mm_mul_pd(r00,vftabscale);
 234             vfitab           = _mm_cvttpd_epi32(rt);
 235             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 236             vfitab           = _mm_slli_epi32(vfitab,2);
 237
 238             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 239             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 240             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 241             GMX_MM_TRANSPOSE2_PD(Y,F);
 242             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 243             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 244             GMX_MM_TRANSPOSE2_PD(G,H);
 245             Heps             = _mm_mul_pd(vfeps,H);
 246             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 247             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 248             velec            = _mm_mul_pd(qq00,VV);
 249             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 250             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 251
 252             /* LENNARD-JONES DISPERSION/REPULSION */
 253
 254             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 255             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 256             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 257             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 258             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velecsum         = _mm_add_pd(velecsum,velec);
 262             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 263
 264             fscal            = _mm_add_pd(felec,fvdw);
 265
 266             /* Calculate temporary vectorial force */
 267             tx               = _mm_mul_pd(fscal,dx00);
 268             ty               = _mm_mul_pd(fscal,dy00);
 269             tz               = _mm_mul_pd(fscal,dz00);
 270
 271             /* Update vectorial force */
 272             fix0             = _mm_add_pd(fix0,tx);
 273             fiy0             = _mm_add_pd(fiy0,ty);
 274             fiz0             = _mm_add_pd(fiz0,tz);
 275
 276             fjx0             = _mm_add_pd(fjx0,tx);
 277             fjy0             = _mm_add_pd(fjy0,ty);
 278             fjz0             = _mm_add_pd(fjz0,tz);
 279
 280             /**************************
 281              * CALCULATE INTERACTIONS *
 282              **************************/
 283
 284             r10              = _mm_mul_pd(rsq10,rinv10);
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             qq10             = _mm_mul_pd(iq1,jq0);
 288
 289             /* Calculate table index by multiplying r with table scale and truncate to integer */
 290             rt               = _mm_mul_pd(r10,vftabscale);
 291             vfitab           = _mm_cvttpd_epi32(rt);
 292             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 293             vfitab           = _mm_slli_epi32(vfitab,2);
 294
 295             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 296             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 297             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 298             GMX_MM_TRANSPOSE2_PD(Y,F);
 299             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 300             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 301             GMX_MM_TRANSPOSE2_PD(G,H);
 302             Heps             = _mm_mul_pd(vfeps,H);
 303             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 304             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 305             velec            = _mm_mul_pd(qq10,VV);
 306             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 307             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velecsum         = _mm_add_pd(velecsum,velec);
 311
 312             fscal            = felec;
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm_mul_pd(fscal,dx10);
 316             ty               = _mm_mul_pd(fscal,dy10);
 317             tz               = _mm_mul_pd(fscal,dz10);
 318
 319             /* Update vectorial force */
 320             fix1             = _mm_add_pd(fix1,tx);
 321             fiy1             = _mm_add_pd(fiy1,ty);
 322             fiz1             = _mm_add_pd(fiz1,tz);
 323
 324             fjx0             = _mm_add_pd(fjx0,tx);
 325             fjy0             = _mm_add_pd(fjy0,ty);
 326             fjz0             = _mm_add_pd(fjz0,tz);
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             r20              = _mm_mul_pd(rsq20,rinv20);
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq20             = _mm_mul_pd(iq2,jq0);
 336
 337             /* Calculate table index by multiplying r with table scale and truncate to integer */
 338             rt               = _mm_mul_pd(r20,vftabscale);
 339             vfitab           = _mm_cvttpd_epi32(rt);
 340             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 341             vfitab           = _mm_slli_epi32(vfitab,2);
 342
 343             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 344             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 345             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 346             GMX_MM_TRANSPOSE2_PD(Y,F);
 347             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 348             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 349             GMX_MM_TRANSPOSE2_PD(G,H);
 350             Heps             = _mm_mul_pd(vfeps,H);
 351             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 352             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 353             velec            = _mm_mul_pd(qq20,VV);
 354             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 355             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velecsum         = _mm_add_pd(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm_mul_pd(fscal,dx20);
 364             ty               = _mm_mul_pd(fscal,dy20);
 365             tz               = _mm_mul_pd(fscal,dz20);
 366
 367             /* Update vectorial force */
 368             fix2             = _mm_add_pd(fix2,tx);
 369             fiy2             = _mm_add_pd(fiy2,ty);
 370             fiz2             = _mm_add_pd(fiz2,tz);
 371
 372             fjx0             = _mm_add_pd(fjx0,tx);
 373             fjy0             = _mm_add_pd(fjy0,ty);
 374             fjz0             = _mm_add_pd(fjz0,tz);
 375
 376             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 377
 378             /* Inner loop uses 145 flops */
 379         }
 380
 381         if(jidx<j_index_end)
 382         {
 383
 384             jnrA             = jjnr[jidx];
 385             j_coord_offsetA  = DIM*jnrA;
 386
 387             /* load j atom coordinates */
 388             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 389                                               &jx0,&jy0,&jz0);
 390
 391             /* Calculate displacement vector */
 392             dx00             = _mm_sub_pd(ix0,jx0);
 393             dy00             = _mm_sub_pd(iy0,jy0);
 394             dz00             = _mm_sub_pd(iz0,jz0);
 395             dx10             = _mm_sub_pd(ix1,jx0);
 396             dy10             = _mm_sub_pd(iy1,jy0);
 397             dz10             = _mm_sub_pd(iz1,jz0);
 398             dx20             = _mm_sub_pd(ix2,jx0);
 399             dy20             = _mm_sub_pd(iy2,jy0);
 400             dz20             = _mm_sub_pd(iz2,jz0);
 401
 402             /* Calculate squared distance and things based on it */
 403             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 404             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 405             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 406
 407             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 408             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 409             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 410
 411             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 412
 413             /* Load parameters for j particles */
 414             jq0              = _mm_load_sd(charge+jnrA+0);
 415             vdwjidx0A        = 2*vdwtype[jnrA+0];
 416
 417             fjx0             = _mm_setzero_pd();
 418             fjy0             = _mm_setzero_pd();
 419             fjz0             = _mm_setzero_pd();
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             r00              = _mm_mul_pd(rsq00,rinv00);
 426
 427             /* Compute parameters for interactions between i and j atoms */
 428             qq00             = _mm_mul_pd(iq0,jq0);
 429             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 430
 431             /* Calculate table index by multiplying r with table scale and truncate to integer */
 432             rt               = _mm_mul_pd(r00,vftabscale);
 433             vfitab           = _mm_cvttpd_epi32(rt);
 434             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 435             vfitab           = _mm_slli_epi32(vfitab,2);
 436
 437             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 438             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 439             F                = _mm_setzero_pd();
 440             GMX_MM_TRANSPOSE2_PD(Y,F);
 441             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 442             H                = _mm_setzero_pd();
 443             GMX_MM_TRANSPOSE2_PD(G,H);
 444             Heps             = _mm_mul_pd(vfeps,H);
 445             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 446             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 447             velec            = _mm_mul_pd(qq00,VV);
 448             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 449             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 450
 451             /* LENNARD-JONES DISPERSION/REPULSION */
 452
 453             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 454             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 455             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 456             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 457             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 461             velecsum         = _mm_add_pd(velecsum,velec);
 462             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 463             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 464
 465             fscal            = _mm_add_pd(felec,fvdw);
 466
 467             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 468
 469             /* Calculate temporary vectorial force */
 470             tx               = _mm_mul_pd(fscal,dx00);
 471             ty               = _mm_mul_pd(fscal,dy00);
 472             tz               = _mm_mul_pd(fscal,dz00);
 473
 474             /* Update vectorial force */
 475             fix0             = _mm_add_pd(fix0,tx);
 476             fiy0             = _mm_add_pd(fiy0,ty);
 477             fiz0             = _mm_add_pd(fiz0,tz);
 478
 479             fjx0             = _mm_add_pd(fjx0,tx);
 480             fjy0             = _mm_add_pd(fjy0,ty);
 481             fjz0             = _mm_add_pd(fjz0,tz);
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             r10              = _mm_mul_pd(rsq10,rinv10);
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             qq10             = _mm_mul_pd(iq1,jq0);
 491
 492             /* Calculate table index by multiplying r with table scale and truncate to integer */
 493             rt               = _mm_mul_pd(r10,vftabscale);
 494             vfitab           = _mm_cvttpd_epi32(rt);
 495             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 496             vfitab           = _mm_slli_epi32(vfitab,2);
 497
 498             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 499             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 500             F                = _mm_setzero_pd();
 501             GMX_MM_TRANSPOSE2_PD(Y,F);
 502             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 503             H                = _mm_setzero_pd();
 504             GMX_MM_TRANSPOSE2_PD(G,H);
 505             Heps             = _mm_mul_pd(vfeps,H);
 506             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 507             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 508             velec            = _mm_mul_pd(qq10,VV);
 509             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 510             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 514             velecsum         = _mm_add_pd(velecsum,velec);
 515
 516             fscal            = felec;
 517
 518             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 519
 520             /* Calculate temporary vectorial force */
 521             tx               = _mm_mul_pd(fscal,dx10);
 522             ty               = _mm_mul_pd(fscal,dy10);
 523             tz               = _mm_mul_pd(fscal,dz10);
 524
 525             /* Update vectorial force */
 526             fix1             = _mm_add_pd(fix1,tx);
 527             fiy1             = _mm_add_pd(fiy1,ty);
 528             fiz1             = _mm_add_pd(fiz1,tz);
 529
 530             fjx0             = _mm_add_pd(fjx0,tx);
 531             fjy0             = _mm_add_pd(fjy0,ty);
 532             fjz0             = _mm_add_pd(fjz0,tz);
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             r20              = _mm_mul_pd(rsq20,rinv20);
 539
 540             /* Compute parameters for interactions between i and j atoms */
 541             qq20             = _mm_mul_pd(iq2,jq0);
 542
 543             /* Calculate table index by multiplying r with table scale and truncate to integer */
 544             rt               = _mm_mul_pd(r20,vftabscale);
 545             vfitab           = _mm_cvttpd_epi32(rt);
 546             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 547             vfitab           = _mm_slli_epi32(vfitab,2);
 548
 549             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 550             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 551             F                = _mm_setzero_pd();
 552             GMX_MM_TRANSPOSE2_PD(Y,F);
 553             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 554             H                = _mm_setzero_pd();
 555             GMX_MM_TRANSPOSE2_PD(G,H);
 556             Heps             = _mm_mul_pd(vfeps,H);
 557             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 558             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 559             velec            = _mm_mul_pd(qq20,VV);
 560             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 561             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 565             velecsum         = _mm_add_pd(velecsum,velec);
 566
 567             fscal            = felec;
 568
 569             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm_mul_pd(fscal,dx20);
 573             ty               = _mm_mul_pd(fscal,dy20);
 574             tz               = _mm_mul_pd(fscal,dz20);
 575
 576             /* Update vectorial force */
 577             fix2             = _mm_add_pd(fix2,tx);
 578             fiy2             = _mm_add_pd(fiy2,ty);
 579             fiz2             = _mm_add_pd(fiz2,tz);
 580
 581             fjx0             = _mm_add_pd(fjx0,tx);
 582             fjy0             = _mm_add_pd(fjy0,ty);
 583             fjz0             = _mm_add_pd(fjz0,tz);
 584
 585             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 586
 587             /* Inner loop uses 145 flops */
 588         }
 589
 590         /* End of innermost loop */
 591
 592         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 593                                               f+i_coord_offset,fshift+i_shift_offset);
 594
 595         ggid                        = gid[iidx];
 596         /* Update potential energies */
 597         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 598         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 599
 600         /* Increment number of inner iterations */
 601         inneriter                  += j_index_end - j_index_start;
 602
 603         /* Outer loop uses 20 flops */
 604     }
 605
 606     /* Increment number of outer iterations */
 607     outeriter        += nri;
 608
 609     /* Update outer/inner flops */
 610
 611     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 612 }
 613 /*
 614  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_double
 615  * Electrostatics interaction: CubicSplineTable
 616  * VdW interaction:            LennardJones
 617  * Geometry:                   Water3-Particle
 618  * Calculate force/pot:        Force
 619  */
 620 void
 621 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_double
 622                     (t_nblist                    * gmx_restrict       nlist,
 623                      rvec                        * gmx_restrict          xx,
 624                      rvec                        * gmx_restrict          ff,
 625                      t_forcerec                  * gmx_restrict          fr,
 626                      t_mdatoms                   * gmx_restrict     mdatoms,
 627                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 628                      t_nrnb                      * gmx_restrict        nrnb)
 629 {
 630     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 631      * just 0 for non-waters.
 632      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 633      * jnr indices corresponding to data put in the four positions in the SIMD register.
 634      */
 635     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 636     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 637     int              jnrA,jnrB;
 638     int              j_coord_offsetA,j_coord_offsetB;
 639     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 640     real             rcutoff_scalar;
 641     real             *shiftvec,*fshift,*x,*f;
 642     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 643     int              vdwioffset0;
 644     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 645     int              vdwioffset1;
 646     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 647     int              vdwioffset2;
 648     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 649     int              vdwjidx0A,vdwjidx0B;
 650     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 651     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 652     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 653     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 654     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 655     real             *charge;
 656     int              nvdwtype;
 657     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 658     int              *vdwtype;
 659     real             *vdwparam;
 660     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 661     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 662     __m128i          vfitab;
 663     __m128i          ifour       = _mm_set1_epi32(4);
 664     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 665     real             *vftab;
 666     __m128d          dummy_mask,cutoff_mask;
 667     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 668     __m128d          one     = _mm_set1_pd(1.0);
 669     __m128d          two     = _mm_set1_pd(2.0);
 670     x                = xx[0];
 671     f                = ff[0];
 672
 673     nri              = nlist->nri;
 674     iinr             = nlist->iinr;
 675     jindex           = nlist->jindex;
 676     jjnr             = nlist->jjnr;
 677     shiftidx         = nlist->shift;
 678     gid              = nlist->gid;
 679     shiftvec         = fr->shift_vec[0];
 680     fshift           = fr->fshift[0];
 681     facel            = _mm_set1_pd(fr->epsfac);
 682     charge           = mdatoms->chargeA;
 683     nvdwtype         = fr->ntype;
 684     vdwparam         = fr->nbfp;
 685     vdwtype          = mdatoms->typeA;
 686
 687     vftab            = kernel_data->table_elec->data;
 688     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 689
 690     /* Setup water-specific parameters */
 691     inr              = nlist->iinr[0];
 692     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 693     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 694     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 695     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 696
 697     /* Avoid stupid compiler warnings */
 698     jnrA = jnrB = 0;
 699     j_coord_offsetA = 0;
 700     j_coord_offsetB = 0;
 701
 702     outeriter        = 0;
 703     inneriter        = 0;
 704
 705     /* Start outer loop over neighborlists */
 706     for(iidx=0; iidx<nri; iidx++)
 707     {
 708         /* Load shift vector for this list */
 709         i_shift_offset   = DIM*shiftidx[iidx];
 710
 711         /* Load limits for loop over neighbors */
 712         j_index_start    = jindex[iidx];
 713         j_index_end      = jindex[iidx+1];
 714
 715         /* Get outer coordinate index */
 716         inr              = iinr[iidx];
 717         i_coord_offset   = DIM*inr;
 718
 719         /* Load i particle coords and add shift vector */
 720         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 721                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 722
 723         fix0             = _mm_setzero_pd();
 724         fiy0             = _mm_setzero_pd();
 725         fiz0             = _mm_setzero_pd();
 726         fix1             = _mm_setzero_pd();
 727         fiy1             = _mm_setzero_pd();
 728         fiz1             = _mm_setzero_pd();
 729         fix2             = _mm_setzero_pd();
 730         fiy2             = _mm_setzero_pd();
 731         fiz2             = _mm_setzero_pd();
 732
 733         /* Start inner kernel loop */
 734         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 735         {
 736
 737             /* Get j neighbor index, and coordinate index */
 738             jnrA             = jjnr[jidx];
 739             jnrB             = jjnr[jidx+1];
 740             j_coord_offsetA  = DIM*jnrA;
 741             j_coord_offsetB  = DIM*jnrB;
 742
 743             /* load j atom coordinates */
 744             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 745                                               &jx0,&jy0,&jz0);
 746
 747             /* Calculate displacement vector */
 748             dx00             = _mm_sub_pd(ix0,jx0);
 749             dy00             = _mm_sub_pd(iy0,jy0);
 750             dz00             = _mm_sub_pd(iz0,jz0);
 751             dx10             = _mm_sub_pd(ix1,jx0);
 752             dy10             = _mm_sub_pd(iy1,jy0);
 753             dz10             = _mm_sub_pd(iz1,jz0);
 754             dx20             = _mm_sub_pd(ix2,jx0);
 755             dy20             = _mm_sub_pd(iy2,jy0);
 756             dz20             = _mm_sub_pd(iz2,jz0);
 757
 758             /* Calculate squared distance and things based on it */
 759             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 760             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 761             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 762
 763             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 764             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 765             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 766
 767             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 768
 769             /* Load parameters for j particles */
 770             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 771             vdwjidx0A        = 2*vdwtype[jnrA+0];
 772             vdwjidx0B        = 2*vdwtype[jnrB+0];
 773
 774             fjx0             = _mm_setzero_pd();
 775             fjy0             = _mm_setzero_pd();
 776             fjz0             = _mm_setzero_pd();
 777
 778             /**************************
 779              * CALCULATE INTERACTIONS *
 780              **************************/
 781
 782             r00              = _mm_mul_pd(rsq00,rinv00);
 783
 784             /* Compute parameters for interactions between i and j atoms */
 785             qq00             = _mm_mul_pd(iq0,jq0);
 786             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 787                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 788
 789             /* Calculate table index by multiplying r with table scale and truncate to integer */
 790             rt               = _mm_mul_pd(r00,vftabscale);
 791             vfitab           = _mm_cvttpd_epi32(rt);
 792             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 793             vfitab           = _mm_slli_epi32(vfitab,2);
 794
 795             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 796             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 797             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 798             GMX_MM_TRANSPOSE2_PD(Y,F);
 799             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 800             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 801             GMX_MM_TRANSPOSE2_PD(G,H);
 802             Heps             = _mm_mul_pd(vfeps,H);
 803             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 804             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 805             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 806
 807             /* LENNARD-JONES DISPERSION/REPULSION */
 808
 809             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 810             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 811
 812             fscal            = _mm_add_pd(felec,fvdw);
 813
 814             /* Calculate temporary vectorial force */
 815             tx               = _mm_mul_pd(fscal,dx00);
 816             ty               = _mm_mul_pd(fscal,dy00);
 817             tz               = _mm_mul_pd(fscal,dz00);
 818
 819             /* Update vectorial force */
 820             fix0             = _mm_add_pd(fix0,tx);
 821             fiy0             = _mm_add_pd(fiy0,ty);
 822             fiz0             = _mm_add_pd(fiz0,tz);
 823
 824             fjx0             = _mm_add_pd(fjx0,tx);
 825             fjy0             = _mm_add_pd(fjy0,ty);
 826             fjz0             = _mm_add_pd(fjz0,tz);
 827
 828             /**************************
 829              * CALCULATE INTERACTIONS *
 830              **************************/
 831
 832             r10              = _mm_mul_pd(rsq10,rinv10);
 833
 834             /* Compute parameters for interactions between i and j atoms */
 835             qq10             = _mm_mul_pd(iq1,jq0);
 836
 837             /* Calculate table index by multiplying r with table scale and truncate to integer */
 838             rt               = _mm_mul_pd(r10,vftabscale);
 839             vfitab           = _mm_cvttpd_epi32(rt);
 840             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 841             vfitab           = _mm_slli_epi32(vfitab,2);
 842
 843             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 844             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 845             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 846             GMX_MM_TRANSPOSE2_PD(Y,F);
 847             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 848             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 849             GMX_MM_TRANSPOSE2_PD(G,H);
 850             Heps             = _mm_mul_pd(vfeps,H);
 851             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 852             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 853             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 854
 855             fscal            = felec;
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm_mul_pd(fscal,dx10);
 859             ty               = _mm_mul_pd(fscal,dy10);
 860             tz               = _mm_mul_pd(fscal,dz10);
 861
 862             /* Update vectorial force */
 863             fix1             = _mm_add_pd(fix1,tx);
 864             fiy1             = _mm_add_pd(fiy1,ty);
 865             fiz1             = _mm_add_pd(fiz1,tz);
 866
 867             fjx0             = _mm_add_pd(fjx0,tx);
 868             fjy0             = _mm_add_pd(fjy0,ty);
 869             fjz0             = _mm_add_pd(fjz0,tz);
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             r20              = _mm_mul_pd(rsq20,rinv20);
 876
 877             /* Compute parameters for interactions between i and j atoms */
 878             qq20             = _mm_mul_pd(iq2,jq0);
 879
 880             /* Calculate table index by multiplying r with table scale and truncate to integer */
 881             rt               = _mm_mul_pd(r20,vftabscale);
 882             vfitab           = _mm_cvttpd_epi32(rt);
 883             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 884             vfitab           = _mm_slli_epi32(vfitab,2);
 885
 886             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 887             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 888             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 889             GMX_MM_TRANSPOSE2_PD(Y,F);
 890             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 891             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 892             GMX_MM_TRANSPOSE2_PD(G,H);
 893             Heps             = _mm_mul_pd(vfeps,H);
 894             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 895             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 896             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 897
 898             fscal            = felec;
 899
 900             /* Calculate temporary vectorial force */
 901             tx               = _mm_mul_pd(fscal,dx20);
 902             ty               = _mm_mul_pd(fscal,dy20);
 903             tz               = _mm_mul_pd(fscal,dz20);
 904
 905             /* Update vectorial force */
 906             fix2             = _mm_add_pd(fix2,tx);
 907             fiy2             = _mm_add_pd(fiy2,ty);
 908             fiz2             = _mm_add_pd(fiz2,tz);
 909
 910             fjx0             = _mm_add_pd(fjx0,tx);
 911             fjy0             = _mm_add_pd(fjy0,ty);
 912             fjz0             = _mm_add_pd(fjz0,tz);
 913
 914             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 915
 916             /* Inner loop uses 128 flops */
 917         }
 918
 919         if(jidx<j_index_end)
 920         {
 921
 922             jnrA             = jjnr[jidx];
 923             j_coord_offsetA  = DIM*jnrA;
 924
 925             /* load j atom coordinates */
 926             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 927                                               &jx0,&jy0,&jz0);
 928
 929             /* Calculate displacement vector */
 930             dx00             = _mm_sub_pd(ix0,jx0);
 931             dy00             = _mm_sub_pd(iy0,jy0);
 932             dz00             = _mm_sub_pd(iz0,jz0);
 933             dx10             = _mm_sub_pd(ix1,jx0);
 934             dy10             = _mm_sub_pd(iy1,jy0);
 935             dz10             = _mm_sub_pd(iz1,jz0);
 936             dx20             = _mm_sub_pd(ix2,jx0);
 937             dy20             = _mm_sub_pd(iy2,jy0);
 938             dz20             = _mm_sub_pd(iz2,jz0);
 939
 940             /* Calculate squared distance and things based on it */
 941             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 942             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 943             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 944
 945             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 946             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 947             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 948
 949             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 950
 951             /* Load parameters for j particles */
 952             jq0              = _mm_load_sd(charge+jnrA+0);
 953             vdwjidx0A        = 2*vdwtype[jnrA+0];
 954
 955             fjx0             = _mm_setzero_pd();
 956             fjy0             = _mm_setzero_pd();
 957             fjz0             = _mm_setzero_pd();
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             r00              = _mm_mul_pd(rsq00,rinv00);
 964
 965             /* Compute parameters for interactions between i and j atoms */
 966             qq00             = _mm_mul_pd(iq0,jq0);
 967             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 968
 969             /* Calculate table index by multiplying r with table scale and truncate to integer */
 970             rt               = _mm_mul_pd(r00,vftabscale);
 971             vfitab           = _mm_cvttpd_epi32(rt);
 972             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 973             vfitab           = _mm_slli_epi32(vfitab,2);
 974
 975             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 976             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 977             F                = _mm_setzero_pd();
 978             GMX_MM_TRANSPOSE2_PD(Y,F);
 979             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 980             H                = _mm_setzero_pd();
 981             GMX_MM_TRANSPOSE2_PD(G,H);
 982             Heps             = _mm_mul_pd(vfeps,H);
 983             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 984             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 985             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 986
 987             /* LENNARD-JONES DISPERSION/REPULSION */
 988
 989             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 990             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 991
 992             fscal            = _mm_add_pd(felec,fvdw);
 993
 994             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 995
 996             /* Calculate temporary vectorial force */
 997             tx               = _mm_mul_pd(fscal,dx00);
 998             ty               = _mm_mul_pd(fscal,dy00);
 999             tz               = _mm_mul_pd(fscal,dz00);
1000
1001             /* Update vectorial force */
1002             fix0             = _mm_add_pd(fix0,tx);
1003             fiy0             = _mm_add_pd(fiy0,ty);
1004             fiz0             = _mm_add_pd(fiz0,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             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             r10              = _mm_mul_pd(rsq10,rinv10);
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq10             = _mm_mul_pd(iq1,jq0);
1018
1019             /* Calculate table index by multiplying r with table scale and truncate to integer */
1020             rt               = _mm_mul_pd(r10,vftabscale);
1021             vfitab           = _mm_cvttpd_epi32(rt);
1022             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1023             vfitab           = _mm_slli_epi32(vfitab,2);
1024
1025             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1026             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1027             F                = _mm_setzero_pd();
1028             GMX_MM_TRANSPOSE2_PD(Y,F);
1029             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1030             H                = _mm_setzero_pd();
1031             GMX_MM_TRANSPOSE2_PD(G,H);
1032             Heps             = _mm_mul_pd(vfeps,H);
1033             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1034             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1035             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1036
1037             fscal            = felec;
1038
1039             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1040
1041             /* Calculate temporary vectorial force */
1042             tx               = _mm_mul_pd(fscal,dx10);
1043             ty               = _mm_mul_pd(fscal,dy10);
1044             tz               = _mm_mul_pd(fscal,dz10);
1045
1046             /* Update vectorial force */
1047             fix1             = _mm_add_pd(fix1,tx);
1048             fiy1             = _mm_add_pd(fiy1,ty);
1049             fiz1             = _mm_add_pd(fiz1,tz);
1050
1051             fjx0             = _mm_add_pd(fjx0,tx);
1052             fjy0             = _mm_add_pd(fjy0,ty);
1053             fjz0             = _mm_add_pd(fjz0,tz);
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             r20              = _mm_mul_pd(rsq20,rinv20);
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq20             = _mm_mul_pd(iq2,jq0);
1063
1064             /* Calculate table index by multiplying r with table scale and truncate to integer */
1065             rt               = _mm_mul_pd(r20,vftabscale);
1066             vfitab           = _mm_cvttpd_epi32(rt);
1067             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1068             vfitab           = _mm_slli_epi32(vfitab,2);
1069
1070             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1071             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1072             F                = _mm_setzero_pd();
1073             GMX_MM_TRANSPOSE2_PD(Y,F);
1074             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1075             H                = _mm_setzero_pd();
1076             GMX_MM_TRANSPOSE2_PD(G,H);
1077             Heps             = _mm_mul_pd(vfeps,H);
1078             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1079             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1080             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1081
1082             fscal            = felec;
1083
1084             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1085
1086             /* Calculate temporary vectorial force */
1087             tx               = _mm_mul_pd(fscal,dx20);
1088             ty               = _mm_mul_pd(fscal,dy20);
1089             tz               = _mm_mul_pd(fscal,dz20);
1090
1091             /* Update vectorial force */
1092             fix2             = _mm_add_pd(fix2,tx);
1093             fiy2             = _mm_add_pd(fiy2,ty);
1094             fiz2             = _mm_add_pd(fiz2,tz);
1095
1096             fjx0             = _mm_add_pd(fjx0,tx);
1097             fjy0             = _mm_add_pd(fjy0,ty);
1098             fjz0             = _mm_add_pd(fjz0,tz);
1099
1100             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1101
1102             /* Inner loop uses 128 flops */
1103         }
1104
1105         /* End of innermost loop */
1106
1107         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1108                                               f+i_coord_offset,fshift+i_shift_offset);
1109
1110         /* Increment number of inner iterations */
1111         inneriter                  += j_index_end - j_index_start;
1112
1113         /* Outer loop uses 18 flops */
1114     }
1115
1116     /* Increment number of outer iterations */
1117     outeriter        += nri;
1118
1119     /* Update outer/inner flops */
1120
1121     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1122 }