src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     vftab            = kernel_data->table_elec_vdw->data;
 125     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm_setzero_pd();
 161         fiy0             = _mm_setzero_pd();
 162         fiz0             = _mm_setzero_pd();
 163         fix1             = _mm_setzero_pd();
 164         fiy1             = _mm_setzero_pd();
 165         fiz1             = _mm_setzero_pd();
 166         fix2             = _mm_setzero_pd();
 167         fiy2             = _mm_setzero_pd();
 168         fiz2             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172         vvdwsum          = _mm_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_pd(ix0,jx0);
 190             dy00             = _mm_sub_pd(iy0,jy0);
 191             dz00             = _mm_sub_pd(iz0,jz0);
 192             dx10             = _mm_sub_pd(ix1,jx0);
 193             dy10             = _mm_sub_pd(iy1,jy0);
 194             dz10             = _mm_sub_pd(iz1,jz0);
 195             dx20             = _mm_sub_pd(ix2,jx0);
 196             dy20             = _mm_sub_pd(iy2,jy0);
 197             dz20             = _mm_sub_pd(iz2,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 201             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 202             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 203
 204             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 205             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 206             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 210             vdwjidx0A        = 2*vdwtype[jnrA+0];
 211             vdwjidx0B        = 2*vdwtype[jnrB+0];
 212
 213             fjx0             = _mm_setzero_pd();
 214             fjy0             = _mm_setzero_pd();
 215             fjz0             = _mm_setzero_pd();
 216
 217             /**************************
 218              * CALCULATE INTERACTIONS *
 219              **************************/
 220
 221             r00              = _mm_mul_pd(rsq00,rinv00);
 222
 223             /* Compute parameters for interactions between i and j atoms */
 224             qq00             = _mm_mul_pd(iq0,jq0);
 225             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 226                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 227
 228             /* Calculate table index by multiplying r with table scale and truncate to integer */
 229             rt               = _mm_mul_pd(r00,vftabscale);
 230             vfitab           = _mm_cvttpd_epi32(rt);
 231             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 232             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 233
 234             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 235             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 236             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 237             GMX_MM_TRANSPOSE2_PD(Y,F);
 238             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 239             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 240             GMX_MM_TRANSPOSE2_PD(G,H);
 241             Heps             = _mm_mul_pd(vfeps,H);
 242             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 243             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 244             velec            = _mm_mul_pd(qq00,VV);
 245             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 246             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 247
 248             /* CUBIC SPLINE TABLE DISPERSION */
 249             vfitab           = _mm_add_epi32(vfitab,ifour);
 250             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 251             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 252             GMX_MM_TRANSPOSE2_PD(Y,F);
 253             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 254             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 255             GMX_MM_TRANSPOSE2_PD(G,H);
 256             Heps             = _mm_mul_pd(vfeps,H);
 257             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 258             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 259             vvdw6            = _mm_mul_pd(c6_00,VV);
 260             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 261             fvdw6            = _mm_mul_pd(c6_00,FF);
 262
 263             /* CUBIC SPLINE TABLE REPULSION */
 264             vfitab           = _mm_add_epi32(vfitab,ifour);
 265             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 266             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 267             GMX_MM_TRANSPOSE2_PD(Y,F);
 268             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 269             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 270             GMX_MM_TRANSPOSE2_PD(G,H);
 271             Heps             = _mm_mul_pd(vfeps,H);
 272             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 273             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 274             vvdw12           = _mm_mul_pd(c12_00,VV);
 275             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 276             fvdw12           = _mm_mul_pd(c12_00,FF);
 277             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 278             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm_add_pd(velecsum,velec);
 282             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 283
 284             fscal            = _mm_add_pd(felec,fvdw);
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_pd(fscal,dx00);
 288             ty               = _mm_mul_pd(fscal,dy00);
 289             tz               = _mm_mul_pd(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_add_pd(fix0,tx);
 293             fiy0             = _mm_add_pd(fiy0,ty);
 294             fiz0             = _mm_add_pd(fiz0,tz);
 295
 296             fjx0             = _mm_add_pd(fjx0,tx);
 297             fjy0             = _mm_add_pd(fjy0,ty);
 298             fjz0             = _mm_add_pd(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r10              = _mm_mul_pd(rsq10,rinv10);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq10             = _mm_mul_pd(iq1,jq0);
 308
 309             /* Calculate table index by multiplying r with table scale and truncate to integer */
 310             rt               = _mm_mul_pd(r10,vftabscale);
 311             vfitab           = _mm_cvttpd_epi32(rt);
 312             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 313             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 314
 315             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 316             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 317             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 318             GMX_MM_TRANSPOSE2_PD(Y,F);
 319             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 320             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 321             GMX_MM_TRANSPOSE2_PD(G,H);
 322             Heps             = _mm_mul_pd(vfeps,H);
 323             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 324             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 325             velec            = _mm_mul_pd(qq10,VV);
 326             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 327             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velecsum         = _mm_add_pd(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334             /* Calculate temporary vectorial force */
 335             tx               = _mm_mul_pd(fscal,dx10);
 336             ty               = _mm_mul_pd(fscal,dy10);
 337             tz               = _mm_mul_pd(fscal,dz10);
 338
 339             /* Update vectorial force */
 340             fix1             = _mm_add_pd(fix1,tx);
 341             fiy1             = _mm_add_pd(fiy1,ty);
 342             fiz1             = _mm_add_pd(fiz1,tz);
 343
 344             fjx0             = _mm_add_pd(fjx0,tx);
 345             fjy0             = _mm_add_pd(fjy0,ty);
 346             fjz0             = _mm_add_pd(fjz0,tz);
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             r20              = _mm_mul_pd(rsq20,rinv20);
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq20             = _mm_mul_pd(iq2,jq0);
 356
 357             /* Calculate table index by multiplying r with table scale and truncate to integer */
 358             rt               = _mm_mul_pd(r20,vftabscale);
 359             vfitab           = _mm_cvttpd_epi32(rt);
 360             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 361             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 362
 363             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 364             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 365             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 366             GMX_MM_TRANSPOSE2_PD(Y,F);
 367             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 368             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 369             GMX_MM_TRANSPOSE2_PD(G,H);
 370             Heps             = _mm_mul_pd(vfeps,H);
 371             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 372             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 373             velec            = _mm_mul_pd(qq20,VV);
 374             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 375             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm_add_pd(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm_mul_pd(fscal,dx20);
 384             ty               = _mm_mul_pd(fscal,dy20);
 385             tz               = _mm_mul_pd(fscal,dz20);
 386
 387             /* Update vectorial force */
 388             fix2             = _mm_add_pd(fix2,tx);
 389             fiy2             = _mm_add_pd(fiy2,ty);
 390             fiz2             = _mm_add_pd(fiz2,tz);
 391
 392             fjx0             = _mm_add_pd(fjx0,tx);
 393             fjy0             = _mm_add_pd(fjy0,ty);
 394             fjz0             = _mm_add_pd(fjz0,tz);
 395
 396             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 397
 398             /* Inner loop uses 162 flops */
 399         }
 400
 401         if(jidx<j_index_end)
 402         {
 403
 404             jnrA             = jjnr[jidx];
 405             j_coord_offsetA  = DIM*jnrA;
 406
 407             /* load j atom coordinates */
 408             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 409                                               &jx0,&jy0,&jz0);
 410
 411             /* Calculate displacement vector */
 412             dx00             = _mm_sub_pd(ix0,jx0);
 413             dy00             = _mm_sub_pd(iy0,jy0);
 414             dz00             = _mm_sub_pd(iz0,jz0);
 415             dx10             = _mm_sub_pd(ix1,jx0);
 416             dy10             = _mm_sub_pd(iy1,jy0);
 417             dz10             = _mm_sub_pd(iz1,jz0);
 418             dx20             = _mm_sub_pd(ix2,jx0);
 419             dy20             = _mm_sub_pd(iy2,jy0);
 420             dz20             = _mm_sub_pd(iz2,jz0);
 421
 422             /* Calculate squared distance and things based on it */
 423             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 424             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 425             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 426
 427             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 428             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 429             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 430
 431             /* Load parameters for j particles */
 432             jq0              = _mm_load_sd(charge+jnrA+0);
 433             vdwjidx0A        = 2*vdwtype[jnrA+0];
 434
 435             fjx0             = _mm_setzero_pd();
 436             fjy0             = _mm_setzero_pd();
 437             fjz0             = _mm_setzero_pd();
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             r00              = _mm_mul_pd(rsq00,rinv00);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq00             = _mm_mul_pd(iq0,jq0);
 447             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 448
 449             /* Calculate table index by multiplying r with table scale and truncate to integer */
 450             rt               = _mm_mul_pd(r00,vftabscale);
 451             vfitab           = _mm_cvttpd_epi32(rt);
 452             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 453             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 454
 455             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 456             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 457             F                = _mm_setzero_pd();
 458             GMX_MM_TRANSPOSE2_PD(Y,F);
 459             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 460             H                = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(G,H);
 462             Heps             = _mm_mul_pd(vfeps,H);
 463             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 464             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 465             velec            = _mm_mul_pd(qq00,VV);
 466             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 467             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 468
 469             /* CUBIC SPLINE TABLE DISPERSION */
 470             vfitab           = _mm_add_epi32(vfitab,ifour);
 471             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 472             F                = _mm_setzero_pd();
 473             GMX_MM_TRANSPOSE2_PD(Y,F);
 474             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 475             H                = _mm_setzero_pd();
 476             GMX_MM_TRANSPOSE2_PD(G,H);
 477             Heps             = _mm_mul_pd(vfeps,H);
 478             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 479             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 480             vvdw6            = _mm_mul_pd(c6_00,VV);
 481             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 482             fvdw6            = _mm_mul_pd(c6_00,FF);
 483
 484             /* CUBIC SPLINE TABLE REPULSION */
 485             vfitab           = _mm_add_epi32(vfitab,ifour);
 486             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 487             F                = _mm_setzero_pd();
 488             GMX_MM_TRANSPOSE2_PD(Y,F);
 489             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 490             H                = _mm_setzero_pd();
 491             GMX_MM_TRANSPOSE2_PD(G,H);
 492             Heps             = _mm_mul_pd(vfeps,H);
 493             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 494             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 495             vvdw12           = _mm_mul_pd(c12_00,VV);
 496             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 497             fvdw12           = _mm_mul_pd(c12_00,FF);
 498             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 499             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 503             velecsum         = _mm_add_pd(velecsum,velec);
 504             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 505             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 506
 507             fscal            = _mm_add_pd(felec,fvdw);
 508
 509             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm_mul_pd(fscal,dx00);
 513             ty               = _mm_mul_pd(fscal,dy00);
 514             tz               = _mm_mul_pd(fscal,dz00);
 515
 516             /* Update vectorial force */
 517             fix0             = _mm_add_pd(fix0,tx);
 518             fiy0             = _mm_add_pd(fiy0,ty);
 519             fiz0             = _mm_add_pd(fiz0,tz);
 520
 521             fjx0             = _mm_add_pd(fjx0,tx);
 522             fjy0             = _mm_add_pd(fjy0,ty);
 523             fjz0             = _mm_add_pd(fjz0,tz);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             r10              = _mm_mul_pd(rsq10,rinv10);
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             qq10             = _mm_mul_pd(iq1,jq0);
 533
 534             /* Calculate table index by multiplying r with table scale and truncate to integer */
 535             rt               = _mm_mul_pd(r10,vftabscale);
 536             vfitab           = _mm_cvttpd_epi32(rt);
 537             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 538             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 539
 540             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 541             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 542             F                = _mm_setzero_pd();
 543             GMX_MM_TRANSPOSE2_PD(Y,F);
 544             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 545             H                = _mm_setzero_pd();
 546             GMX_MM_TRANSPOSE2_PD(G,H);
 547             Heps             = _mm_mul_pd(vfeps,H);
 548             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 549             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 550             velec            = _mm_mul_pd(qq10,VV);
 551             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 552             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 556             velecsum         = _mm_add_pd(velecsum,velec);
 557
 558             fscal            = felec;
 559
 560             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 561
 562             /* Calculate temporary vectorial force */
 563             tx               = _mm_mul_pd(fscal,dx10);
 564             ty               = _mm_mul_pd(fscal,dy10);
 565             tz               = _mm_mul_pd(fscal,dz10);
 566
 567             /* Update vectorial force */
 568             fix1             = _mm_add_pd(fix1,tx);
 569             fiy1             = _mm_add_pd(fiy1,ty);
 570             fiz1             = _mm_add_pd(fiz1,tz);
 571
 572             fjx0             = _mm_add_pd(fjx0,tx);
 573             fjy0             = _mm_add_pd(fjy0,ty);
 574             fjz0             = _mm_add_pd(fjz0,tz);
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             r20              = _mm_mul_pd(rsq20,rinv20);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq20             = _mm_mul_pd(iq2,jq0);
 584
 585             /* Calculate table index by multiplying r with table scale and truncate to integer */
 586             rt               = _mm_mul_pd(r20,vftabscale);
 587             vfitab           = _mm_cvttpd_epi32(rt);
 588             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 589             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 590
 591             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 592             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 593             F                = _mm_setzero_pd();
 594             GMX_MM_TRANSPOSE2_PD(Y,F);
 595             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 596             H                = _mm_setzero_pd();
 597             GMX_MM_TRANSPOSE2_PD(G,H);
 598             Heps             = _mm_mul_pd(vfeps,H);
 599             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 600             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 601             velec            = _mm_mul_pd(qq20,VV);
 602             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 603             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 607             velecsum         = _mm_add_pd(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_pd(fscal,dx20);
 615             ty               = _mm_mul_pd(fscal,dy20);
 616             tz               = _mm_mul_pd(fscal,dz20);
 617
 618             /* Update vectorial force */
 619             fix2             = _mm_add_pd(fix2,tx);
 620             fiy2             = _mm_add_pd(fiy2,ty);
 621             fiz2             = _mm_add_pd(fiz2,tz);
 622
 623             fjx0             = _mm_add_pd(fjx0,tx);
 624             fjy0             = _mm_add_pd(fjy0,ty);
 625             fjz0             = _mm_add_pd(fjz0,tz);
 626
 627             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 628
 629             /* Inner loop uses 162 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 635                                               f+i_coord_offset,fshift+i_shift_offset);
 636
 637         ggid                        = gid[iidx];
 638         /* Update potential energies */
 639         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 640         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 641
 642         /* Increment number of inner iterations */
 643         inneriter                  += j_index_end - j_index_start;
 644
 645         /* Outer loop uses 20 flops */
 646     }
 647
 648     /* Increment number of outer iterations */
 649     outeriter        += nri;
 650
 651     /* Update outer/inner flops */
 652
 653     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*162);
 654 }
 655 /*
 656  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_double
 657  * Electrostatics interaction: CubicSplineTable
 658  * VdW interaction:            CubicSplineTable
 659  * Geometry:                   Water3-Particle
 660  * Calculate force/pot:        Force
 661  */
 662 void
 663 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_double
 664                     (t_nblist                    * gmx_restrict       nlist,
 665                      rvec                        * gmx_restrict          xx,
 666                      rvec                        * gmx_restrict          ff,
 667                      t_forcerec                  * gmx_restrict          fr,
 668                      t_mdatoms                   * gmx_restrict     mdatoms,
 669                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 670                      t_nrnb                      * gmx_restrict        nrnb)
 671 {
 672     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 673      * just 0 for non-waters.
 674      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 675      * jnr indices corresponding to data put in the four positions in the SIMD register.
 676      */
 677     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 678     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 679     int              jnrA,jnrB;
 680     int              j_coord_offsetA,j_coord_offsetB;
 681     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 682     real             rcutoff_scalar;
 683     real             *shiftvec,*fshift,*x,*f;
 684     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 685     int              vdwioffset0;
 686     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 687     int              vdwioffset1;
 688     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 689     int              vdwioffset2;
 690     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 691     int              vdwjidx0A,vdwjidx0B;
 692     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 693     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 694     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 695     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 696     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 697     real             *charge;
 698     int              nvdwtype;
 699     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 700     int              *vdwtype;
 701     real             *vdwparam;
 702     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 703     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 704     __m128i          vfitab;
 705     __m128i          ifour       = _mm_set1_epi32(4);
 706     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 707     real             *vftab;
 708     __m128d          dummy_mask,cutoff_mask;
 709     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 710     __m128d          one     = _mm_set1_pd(1.0);
 711     __m128d          two     = _mm_set1_pd(2.0);
 712     x                = xx[0];
 713     f                = ff[0];
 714
 715     nri              = nlist->nri;
 716     iinr             = nlist->iinr;
 717     jindex           = nlist->jindex;
 718     jjnr             = nlist->jjnr;
 719     shiftidx         = nlist->shift;
 720     gid              = nlist->gid;
 721     shiftvec         = fr->shift_vec[0];
 722     fshift           = fr->fshift[0];
 723     facel            = _mm_set1_pd(fr->epsfac);
 724     charge           = mdatoms->chargeA;
 725     nvdwtype         = fr->ntype;
 726     vdwparam         = fr->nbfp;
 727     vdwtype          = mdatoms->typeA;
 728
 729     vftab            = kernel_data->table_elec_vdw->data;
 730     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 731
 732     /* Setup water-specific parameters */
 733     inr              = nlist->iinr[0];
 734     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 735     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 736     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 737     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 738
 739     /* Avoid stupid compiler warnings */
 740     jnrA = jnrB = 0;
 741     j_coord_offsetA = 0;
 742     j_coord_offsetB = 0;
 743
 744     outeriter        = 0;
 745     inneriter        = 0;
 746
 747     /* Start outer loop over neighborlists */
 748     for(iidx=0; iidx<nri; iidx++)
 749     {
 750         /* Load shift vector for this list */
 751         i_shift_offset   = DIM*shiftidx[iidx];
 752
 753         /* Load limits for loop over neighbors */
 754         j_index_start    = jindex[iidx];
 755         j_index_end      = jindex[iidx+1];
 756
 757         /* Get outer coordinate index */
 758         inr              = iinr[iidx];
 759         i_coord_offset   = DIM*inr;
 760
 761         /* Load i particle coords and add shift vector */
 762         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 763                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 764
 765         fix0             = _mm_setzero_pd();
 766         fiy0             = _mm_setzero_pd();
 767         fiz0             = _mm_setzero_pd();
 768         fix1             = _mm_setzero_pd();
 769         fiy1             = _mm_setzero_pd();
 770         fiz1             = _mm_setzero_pd();
 771         fix2             = _mm_setzero_pd();
 772         fiy2             = _mm_setzero_pd();
 773         fiz2             = _mm_setzero_pd();
 774
 775         /* Start inner kernel loop */
 776         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 777         {
 778
 779             /* Get j neighbor index, and coordinate index */
 780             jnrA             = jjnr[jidx];
 781             jnrB             = jjnr[jidx+1];
 782             j_coord_offsetA  = DIM*jnrA;
 783             j_coord_offsetB  = DIM*jnrB;
 784
 785             /* load j atom coordinates */
 786             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 787                                               &jx0,&jy0,&jz0);
 788
 789             /* Calculate displacement vector */
 790             dx00             = _mm_sub_pd(ix0,jx0);
 791             dy00             = _mm_sub_pd(iy0,jy0);
 792             dz00             = _mm_sub_pd(iz0,jz0);
 793             dx10             = _mm_sub_pd(ix1,jx0);
 794             dy10             = _mm_sub_pd(iy1,jy0);
 795             dz10             = _mm_sub_pd(iz1,jz0);
 796             dx20             = _mm_sub_pd(ix2,jx0);
 797             dy20             = _mm_sub_pd(iy2,jy0);
 798             dz20             = _mm_sub_pd(iz2,jz0);
 799
 800             /* Calculate squared distance and things based on it */
 801             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 802             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 803             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 804
 805             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 806             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 807             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 808
 809             /* Load parameters for j particles */
 810             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 811             vdwjidx0A        = 2*vdwtype[jnrA+0];
 812             vdwjidx0B        = 2*vdwtype[jnrB+0];
 813
 814             fjx0             = _mm_setzero_pd();
 815             fjy0             = _mm_setzero_pd();
 816             fjz0             = _mm_setzero_pd();
 817
 818             /**************************
 819              * CALCULATE INTERACTIONS *
 820              **************************/
 821
 822             r00              = _mm_mul_pd(rsq00,rinv00);
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             qq00             = _mm_mul_pd(iq0,jq0);
 826             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 827                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 828
 829             /* Calculate table index by multiplying r with table scale and truncate to integer */
 830             rt               = _mm_mul_pd(r00,vftabscale);
 831             vfitab           = _mm_cvttpd_epi32(rt);
 832             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 833             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 834
 835             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 836             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 837             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 838             GMX_MM_TRANSPOSE2_PD(Y,F);
 839             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 840             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 841             GMX_MM_TRANSPOSE2_PD(G,H);
 842             Heps             = _mm_mul_pd(vfeps,H);
 843             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 844             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 845             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 846
 847             /* CUBIC SPLINE TABLE DISPERSION */
 848             vfitab           = _mm_add_epi32(vfitab,ifour);
 849             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 850             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 851             GMX_MM_TRANSPOSE2_PD(Y,F);
 852             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 853             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 854             GMX_MM_TRANSPOSE2_PD(G,H);
 855             Heps             = _mm_mul_pd(vfeps,H);
 856             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 857             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 858             fvdw6            = _mm_mul_pd(c6_00,FF);
 859
 860             /* CUBIC SPLINE TABLE REPULSION */
 861             vfitab           = _mm_add_epi32(vfitab,ifour);
 862             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 863             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 864             GMX_MM_TRANSPOSE2_PD(Y,F);
 865             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 866             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 867             GMX_MM_TRANSPOSE2_PD(G,H);
 868             Heps             = _mm_mul_pd(vfeps,H);
 869             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 870             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 871             fvdw12           = _mm_mul_pd(c12_00,FF);
 872             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 873
 874             fscal            = _mm_add_pd(felec,fvdw);
 875
 876             /* Calculate temporary vectorial force */
 877             tx               = _mm_mul_pd(fscal,dx00);
 878             ty               = _mm_mul_pd(fscal,dy00);
 879             tz               = _mm_mul_pd(fscal,dz00);
 880
 881             /* Update vectorial force */
 882             fix0             = _mm_add_pd(fix0,tx);
 883             fiy0             = _mm_add_pd(fiy0,ty);
 884             fiz0             = _mm_add_pd(fiz0,tz);
 885
 886             fjx0             = _mm_add_pd(fjx0,tx);
 887             fjy0             = _mm_add_pd(fjy0,ty);
 888             fjz0             = _mm_add_pd(fjz0,tz);
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             r10              = _mm_mul_pd(rsq10,rinv10);
 895
 896             /* Compute parameters for interactions between i and j atoms */
 897             qq10             = _mm_mul_pd(iq1,jq0);
 898
 899             /* Calculate table index by multiplying r with table scale and truncate to integer */
 900             rt               = _mm_mul_pd(r10,vftabscale);
 901             vfitab           = _mm_cvttpd_epi32(rt);
 902             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 903             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 904
 905             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 906             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 907             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 908             GMX_MM_TRANSPOSE2_PD(Y,F);
 909             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 910             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 911             GMX_MM_TRANSPOSE2_PD(G,H);
 912             Heps             = _mm_mul_pd(vfeps,H);
 913             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 914             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 915             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 916
 917             fscal            = felec;
 918
 919             /* Calculate temporary vectorial force */
 920             tx               = _mm_mul_pd(fscal,dx10);
 921             ty               = _mm_mul_pd(fscal,dy10);
 922             tz               = _mm_mul_pd(fscal,dz10);
 923
 924             /* Update vectorial force */
 925             fix1             = _mm_add_pd(fix1,tx);
 926             fiy1             = _mm_add_pd(fiy1,ty);
 927             fiz1             = _mm_add_pd(fiz1,tz);
 928
 929             fjx0             = _mm_add_pd(fjx0,tx);
 930             fjy0             = _mm_add_pd(fjy0,ty);
 931             fjz0             = _mm_add_pd(fjz0,tz);
 932
 933             /**************************
 934              * CALCULATE INTERACTIONS *
 935              **************************/
 936
 937             r20              = _mm_mul_pd(rsq20,rinv20);
 938
 939             /* Compute parameters for interactions between i and j atoms */
 940             qq20             = _mm_mul_pd(iq2,jq0);
 941
 942             /* Calculate table index by multiplying r with table scale and truncate to integer */
 943             rt               = _mm_mul_pd(r20,vftabscale);
 944             vfitab           = _mm_cvttpd_epi32(rt);
 945             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 946             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 947
 948             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 949             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 950             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 951             GMX_MM_TRANSPOSE2_PD(Y,F);
 952             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 953             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 954             GMX_MM_TRANSPOSE2_PD(G,H);
 955             Heps             = _mm_mul_pd(vfeps,H);
 956             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 957             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 958             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 959
 960             fscal            = felec;
 961
 962             /* Calculate temporary vectorial force */
 963             tx               = _mm_mul_pd(fscal,dx20);
 964             ty               = _mm_mul_pd(fscal,dy20);
 965             tz               = _mm_mul_pd(fscal,dz20);
 966
 967             /* Update vectorial force */
 968             fix2             = _mm_add_pd(fix2,tx);
 969             fiy2             = _mm_add_pd(fiy2,ty);
 970             fiz2             = _mm_add_pd(fiz2,tz);
 971
 972             fjx0             = _mm_add_pd(fjx0,tx);
 973             fjy0             = _mm_add_pd(fjy0,ty);
 974             fjz0             = _mm_add_pd(fjz0,tz);
 975
 976             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 977
 978             /* Inner loop uses 142 flops */
 979         }
 980
 981         if(jidx<j_index_end)
 982         {
 983
 984             jnrA             = jjnr[jidx];
 985             j_coord_offsetA  = DIM*jnrA;
 986
 987             /* load j atom coordinates */
 988             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 989                                               &jx0,&jy0,&jz0);
 990
 991             /* Calculate displacement vector */
 992             dx00             = _mm_sub_pd(ix0,jx0);
 993             dy00             = _mm_sub_pd(iy0,jy0);
 994             dz00             = _mm_sub_pd(iz0,jz0);
 995             dx10             = _mm_sub_pd(ix1,jx0);
 996             dy10             = _mm_sub_pd(iy1,jy0);
 997             dz10             = _mm_sub_pd(iz1,jz0);
 998             dx20             = _mm_sub_pd(ix2,jx0);
 999             dy20             = _mm_sub_pd(iy2,jy0);
1000             dz20             = _mm_sub_pd(iz2,jz0);
1001
1002             /* Calculate squared distance and things based on it */
1003             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1004             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1005             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1006
1007             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1008             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1009             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1010
1011             /* Load parameters for j particles */
1012             jq0              = _mm_load_sd(charge+jnrA+0);
1013             vdwjidx0A        = 2*vdwtype[jnrA+0];
1014
1015             fjx0             = _mm_setzero_pd();
1016             fjy0             = _mm_setzero_pd();
1017             fjz0             = _mm_setzero_pd();
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             r00              = _mm_mul_pd(rsq00,rinv00);
1024
1025             /* Compute parameters for interactions between i and j atoms */
1026             qq00             = _mm_mul_pd(iq0,jq0);
1027             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1028
1029             /* Calculate table index by multiplying r with table scale and truncate to integer */
1030             rt               = _mm_mul_pd(r00,vftabscale);
1031             vfitab           = _mm_cvttpd_epi32(rt);
1032             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1033             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1034
1035             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1036             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1037             F                = _mm_setzero_pd();
1038             GMX_MM_TRANSPOSE2_PD(Y,F);
1039             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1040             H                = _mm_setzero_pd();
1041             GMX_MM_TRANSPOSE2_PD(G,H);
1042             Heps             = _mm_mul_pd(vfeps,H);
1043             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1044             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1045             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1046
1047             /* CUBIC SPLINE TABLE DISPERSION */
1048             vfitab           = _mm_add_epi32(vfitab,ifour);
1049             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1050             F                = _mm_setzero_pd();
1051             GMX_MM_TRANSPOSE2_PD(Y,F);
1052             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1053             H                = _mm_setzero_pd();
1054             GMX_MM_TRANSPOSE2_PD(G,H);
1055             Heps             = _mm_mul_pd(vfeps,H);
1056             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1057             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1058             fvdw6            = _mm_mul_pd(c6_00,FF);
1059
1060             /* CUBIC SPLINE TABLE REPULSION */
1061             vfitab           = _mm_add_epi32(vfitab,ifour);
1062             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1063             F                = _mm_setzero_pd();
1064             GMX_MM_TRANSPOSE2_PD(Y,F);
1065             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1066             H                = _mm_setzero_pd();
1067             GMX_MM_TRANSPOSE2_PD(G,H);
1068             Heps             = _mm_mul_pd(vfeps,H);
1069             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1070             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1071             fvdw12           = _mm_mul_pd(c12_00,FF);
1072             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1073
1074             fscal            = _mm_add_pd(felec,fvdw);
1075
1076             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm_mul_pd(fscal,dx00);
1080             ty               = _mm_mul_pd(fscal,dy00);
1081             tz               = _mm_mul_pd(fscal,dz00);
1082
1083             /* Update vectorial force */
1084             fix0             = _mm_add_pd(fix0,tx);
1085             fiy0             = _mm_add_pd(fiy0,ty);
1086             fiz0             = _mm_add_pd(fiz0,tz);
1087
1088             fjx0             = _mm_add_pd(fjx0,tx);
1089             fjy0             = _mm_add_pd(fjy0,ty);
1090             fjz0             = _mm_add_pd(fjz0,tz);
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             r10              = _mm_mul_pd(rsq10,rinv10);
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq10             = _mm_mul_pd(iq1,jq0);
1100
1101             /* Calculate table index by multiplying r with table scale and truncate to integer */
1102             rt               = _mm_mul_pd(r10,vftabscale);
1103             vfitab           = _mm_cvttpd_epi32(rt);
1104             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1105             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1106
1107             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1108             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1109             F                = _mm_setzero_pd();
1110             GMX_MM_TRANSPOSE2_PD(Y,F);
1111             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1112             H                = _mm_setzero_pd();
1113             GMX_MM_TRANSPOSE2_PD(G,H);
1114             Heps             = _mm_mul_pd(vfeps,H);
1115             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1116             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1117             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1118
1119             fscal            = felec;
1120
1121             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1122
1123             /* Calculate temporary vectorial force */
1124             tx               = _mm_mul_pd(fscal,dx10);
1125             ty               = _mm_mul_pd(fscal,dy10);
1126             tz               = _mm_mul_pd(fscal,dz10);
1127
1128             /* Update vectorial force */
1129             fix1             = _mm_add_pd(fix1,tx);
1130             fiy1             = _mm_add_pd(fiy1,ty);
1131             fiz1             = _mm_add_pd(fiz1,tz);
1132
1133             fjx0             = _mm_add_pd(fjx0,tx);
1134             fjy0             = _mm_add_pd(fjy0,ty);
1135             fjz0             = _mm_add_pd(fjz0,tz);
1136
1137             /**************************
1138              * CALCULATE INTERACTIONS *
1139              **************************/
1140
1141             r20              = _mm_mul_pd(rsq20,rinv20);
1142
1143             /* Compute parameters for interactions between i and j atoms */
1144             qq20             = _mm_mul_pd(iq2,jq0);
1145
1146             /* Calculate table index by multiplying r with table scale and truncate to integer */
1147             rt               = _mm_mul_pd(r20,vftabscale);
1148             vfitab           = _mm_cvttpd_epi32(rt);
1149             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1150             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1151
1152             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1153             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1154             F                = _mm_setzero_pd();
1155             GMX_MM_TRANSPOSE2_PD(Y,F);
1156             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1157             H                = _mm_setzero_pd();
1158             GMX_MM_TRANSPOSE2_PD(G,H);
1159             Heps             = _mm_mul_pd(vfeps,H);
1160             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1161             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1162             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1167
1168             /* Calculate temporary vectorial force */
1169             tx               = _mm_mul_pd(fscal,dx20);
1170             ty               = _mm_mul_pd(fscal,dy20);
1171             tz               = _mm_mul_pd(fscal,dz20);
1172
1173             /* Update vectorial force */
1174             fix2             = _mm_add_pd(fix2,tx);
1175             fiy2             = _mm_add_pd(fiy2,ty);
1176             fiz2             = _mm_add_pd(fiz2,tz);
1177
1178             fjx0             = _mm_add_pd(fjx0,tx);
1179             fjy0             = _mm_add_pd(fjy0,ty);
1180             fjz0             = _mm_add_pd(fjz0,tz);
1181
1182             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1183
1184             /* Inner loop uses 142 flops */
1185         }
1186
1187         /* End of innermost loop */
1188
1189         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1190                                               f+i_coord_offset,fshift+i_shift_offset);
1191
1192         /* Increment number of inner iterations */
1193         inneriter                  += j_index_end - j_index_start;
1194
1195         /* Outer loop uses 18 flops */
1196     }
1197
1198     /* Increment number of outer iterations */
1199     outeriter        += nri;
1200
1201     /* Update outer/inner flops */
1202
1203     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1204 }