src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse4_1_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  78     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  79     __m128i          vfitab;
  80     __m128i          ifour       = _mm_set1_epi32(4);
  81     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  82     real             *vftab;
  83     __m128d          dummy_mask,cutoff_mask;
  84     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  85     __m128d          one     = _mm_set1_pd(1.0);
  86     __m128d          two     = _mm_set1_pd(2.0);
  87     x                = xx[0];
  88     f                = ff[0];
  89
  90     nri              = nlist->nri;
  91     iinr             = nlist->iinr;
  92     jindex           = nlist->jindex;
  93     jjnr             = nlist->jjnr;
  94     shiftidx         = nlist->shift;
  95     gid              = nlist->gid;
  96     shiftvec         = fr->shift_vec[0];
  97     fshift           = fr->fshift[0];
  98     facel            = _mm_set1_pd(fr->epsfac);
  99     charge           = mdatoms->chargeA;
 100     nvdwtype         = fr->ntype;
 101     vdwparam         = fr->nbfp;
 102     vdwtype          = mdatoms->typeA;
 103
 104     vftab            = kernel_data->table_vdw->data;
 105     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 106
 107     /* Avoid stupid compiler warnings */
 108     jnrA = jnrB = 0;
 109     j_coord_offsetA = 0;
 110     j_coord_offsetB = 0;
 111
 112     outeriter        = 0;
 113     inneriter        = 0;
 114
 115     /* Start outer loop over neighborlists */
 116     for(iidx=0; iidx<nri; iidx++)
 117     {
 118         /* Load shift vector for this list */
 119         i_shift_offset   = DIM*shiftidx[iidx];
 120
 121         /* Load limits for loop over neighbors */
 122         j_index_start    = jindex[iidx];
 123         j_index_end      = jindex[iidx+1];
 124
 125         /* Get outer coordinate index */
 126         inr              = iinr[iidx];
 127         i_coord_offset   = DIM*inr;
 128
 129         /* Load i particle coords and add shift vector */
 130         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 131
 132         fix0             = _mm_setzero_pd();
 133         fiy0             = _mm_setzero_pd();
 134         fiz0             = _mm_setzero_pd();
 135
 136         /* Load parameters for i particles */
 137         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 138         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140         /* Reset potential sums */
 141         velecsum         = _mm_setzero_pd();
 142         vvdwsum          = _mm_setzero_pd();
 143
 144         /* Start inner kernel loop */
 145         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 146         {
 147
 148             /* Get j neighbor index, and coordinate index */
 149             jnrA             = jjnr[jidx];
 150             jnrB             = jjnr[jidx+1];
 151             j_coord_offsetA  = DIM*jnrA;
 152             j_coord_offsetB  = DIM*jnrB;
 153
 154             /* load j atom coordinates */
 155             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 156                                               &jx0,&jy0,&jz0);
 157
 158             /* Calculate displacement vector */
 159             dx00             = _mm_sub_pd(ix0,jx0);
 160             dy00             = _mm_sub_pd(iy0,jy0);
 161             dz00             = _mm_sub_pd(iz0,jz0);
 162
 163             /* Calculate squared distance and things based on it */
 164             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 165
 166             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 167
 168             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 169
 170             /* Load parameters for j particles */
 171             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 172             vdwjidx0A        = 2*vdwtype[jnrA+0];
 173             vdwjidx0B        = 2*vdwtype[jnrB+0];
 174
 175             /**************************
 176              * CALCULATE INTERACTIONS *
 177              **************************/
 178
 179             r00              = _mm_mul_pd(rsq00,rinv00);
 180
 181             /* Compute parameters for interactions between i and j atoms */
 182             qq00             = _mm_mul_pd(iq0,jq0);
 183             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 184                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 185
 186             /* Calculate table index by multiplying r with table scale and truncate to integer */
 187             rt               = _mm_mul_pd(r00,vftabscale);
 188             vfitab           = _mm_cvttpd_epi32(rt);
 189             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 190             vfitab           = _mm_slli_epi32(vfitab,3);
 191
 192             /* COULOMB ELECTROSTATICS */
 193             velec            = _mm_mul_pd(qq00,rinv00);
 194             felec            = _mm_mul_pd(velec,rinvsq00);
 195
 196             /* CUBIC SPLINE TABLE DISPERSION */
 197             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 198             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 199             GMX_MM_TRANSPOSE2_PD(Y,F);
 200             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 201             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 202             GMX_MM_TRANSPOSE2_PD(G,H);
 203             Heps             = _mm_mul_pd(vfeps,H);
 204             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 205             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 206             vvdw6            = _mm_mul_pd(c6_00,VV);
 207             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 208             fvdw6            = _mm_mul_pd(c6_00,FF);
 209
 210             /* CUBIC SPLINE TABLE REPULSION */
 211             vfitab           = _mm_add_epi32(vfitab,ifour);
 212             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 213             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 214             GMX_MM_TRANSPOSE2_PD(Y,F);
 215             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 216             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 217             GMX_MM_TRANSPOSE2_PD(G,H);
 218             Heps             = _mm_mul_pd(vfeps,H);
 219             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 220             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 221             vvdw12           = _mm_mul_pd(c12_00,VV);
 222             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 223             fvdw12           = _mm_mul_pd(c12_00,FF);
 224             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 225             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             velecsum         = _mm_add_pd(velecsum,velec);
 229             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 230
 231             fscal            = _mm_add_pd(felec,fvdw);
 232
 233             /* Calculate temporary vectorial force */
 234             tx               = _mm_mul_pd(fscal,dx00);
 235             ty               = _mm_mul_pd(fscal,dy00);
 236             tz               = _mm_mul_pd(fscal,dz00);
 237
 238             /* Update vectorial force */
 239             fix0             = _mm_add_pd(fix0,tx);
 240             fiy0             = _mm_add_pd(fiy0,ty);
 241             fiz0             = _mm_add_pd(fiz0,tz);
 242
 243             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 244
 245             /* Inner loop uses 63 flops */
 246         }
 247
 248         if(jidx<j_index_end)
 249         {
 250
 251             jnrA             = jjnr[jidx];
 252             j_coord_offsetA  = DIM*jnrA;
 253
 254             /* load j atom coordinates */
 255             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 256                                               &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm_sub_pd(ix0,jx0);
 260             dy00             = _mm_sub_pd(iy0,jy0);
 261             dz00             = _mm_sub_pd(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 265
 266             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 267
 268             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 269
 270             /* Load parameters for j particles */
 271             jq0              = _mm_load_sd(charge+jnrA+0);
 272             vdwjidx0A        = 2*vdwtype[jnrA+0];
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             r00              = _mm_mul_pd(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq00             = _mm_mul_pd(iq0,jq0);
 282             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 283
 284             /* Calculate table index by multiplying r with table scale and truncate to integer */
 285             rt               = _mm_mul_pd(r00,vftabscale);
 286             vfitab           = _mm_cvttpd_epi32(rt);
 287             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 288             vfitab           = _mm_slli_epi32(vfitab,3);
 289
 290             /* COULOMB ELECTROSTATICS */
 291             velec            = _mm_mul_pd(qq00,rinv00);
 292             felec            = _mm_mul_pd(velec,rinvsq00);
 293
 294             /* CUBIC SPLINE TABLE DISPERSION */
 295             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 296             F                = _mm_setzero_pd();
 297             GMX_MM_TRANSPOSE2_PD(Y,F);
 298             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 299             H                = _mm_setzero_pd();
 300             GMX_MM_TRANSPOSE2_PD(G,H);
 301             Heps             = _mm_mul_pd(vfeps,H);
 302             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 303             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 304             vvdw6            = _mm_mul_pd(c6_00,VV);
 305             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 306             fvdw6            = _mm_mul_pd(c6_00,FF);
 307
 308             /* CUBIC SPLINE TABLE REPULSION */
 309             vfitab           = _mm_add_epi32(vfitab,ifour);
 310             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 311             F                = _mm_setzero_pd();
 312             GMX_MM_TRANSPOSE2_PD(Y,F);
 313             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 314             H                = _mm_setzero_pd();
 315             GMX_MM_TRANSPOSE2_PD(G,H);
 316             Heps             = _mm_mul_pd(vfeps,H);
 317             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 318             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 319             vvdw12           = _mm_mul_pd(c12_00,VV);
 320             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 321             fvdw12           = _mm_mul_pd(c12_00,FF);
 322             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 323             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 327             velecsum         = _mm_add_pd(velecsum,velec);
 328             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 329             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 330
 331             fscal            = _mm_add_pd(felec,fvdw);
 332
 333             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm_mul_pd(fscal,dx00);
 337             ty               = _mm_mul_pd(fscal,dy00);
 338             tz               = _mm_mul_pd(fscal,dz00);
 339
 340             /* Update vectorial force */
 341             fix0             = _mm_add_pd(fix0,tx);
 342             fiy0             = _mm_add_pd(fiy0,ty);
 343             fiz0             = _mm_add_pd(fiz0,tz);
 344
 345             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 346
 347             /* Inner loop uses 63 flops */
 348         }
 349
 350         /* End of innermost loop */
 351
 352         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 353                                               f+i_coord_offset,fshift+i_shift_offset);
 354
 355         ggid                        = gid[iidx];
 356         /* Update potential energies */
 357         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 358         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 359
 360         /* Increment number of inner iterations */
 361         inneriter                  += j_index_end - j_index_start;
 362
 363         /* Outer loop uses 9 flops */
 364     }
 365
 366     /* Increment number of outer iterations */
 367     outeriter        += nri;
 368
 369     /* Update outer/inner flops */
 370
 371     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*63);
 372 }
 373 /*
 374  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse4_1_double
 375  * Electrostatics interaction: Coulomb
 376  * VdW interaction:            CubicSplineTable
 377  * Geometry:                   Particle-Particle
 378  * Calculate force/pot:        Force
 379  */
 380 void
 381 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse4_1_double
 382                     (t_nblist * gmx_restrict                nlist,
 383                      rvec * gmx_restrict                    xx,
 384                      rvec * gmx_restrict                    ff,
 385                      t_forcerec * gmx_restrict              fr,
 386                      t_mdatoms * gmx_restrict               mdatoms,
 387                      nb_kernel_data_t * gmx_restrict        kernel_data,
 388                      t_nrnb * gmx_restrict                  nrnb)
 389 {
 390     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 391      * just 0 for non-waters.
 392      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 393      * jnr indices corresponding to data put in the four positions in the SIMD register.
 394      */
 395     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 396     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 397     int              jnrA,jnrB;
 398     int              j_coord_offsetA,j_coord_offsetB;
 399     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 400     real             rcutoff_scalar;
 401     real             *shiftvec,*fshift,*x,*f;
 402     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 403     int              vdwioffset0;
 404     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 405     int              vdwjidx0A,vdwjidx0B;
 406     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 407     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 408     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 409     real             *charge;
 410     int              nvdwtype;
 411     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 412     int              *vdwtype;
 413     real             *vdwparam;
 414     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 415     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 416     __m128i          vfitab;
 417     __m128i          ifour       = _mm_set1_epi32(4);
 418     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 419     real             *vftab;
 420     __m128d          dummy_mask,cutoff_mask;
 421     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 422     __m128d          one     = _mm_set1_pd(1.0);
 423     __m128d          two     = _mm_set1_pd(2.0);
 424     x                = xx[0];
 425     f                = ff[0];
 426
 427     nri              = nlist->nri;
 428     iinr             = nlist->iinr;
 429     jindex           = nlist->jindex;
 430     jjnr             = nlist->jjnr;
 431     shiftidx         = nlist->shift;
 432     gid              = nlist->gid;
 433     shiftvec         = fr->shift_vec[0];
 434     fshift           = fr->fshift[0];
 435     facel            = _mm_set1_pd(fr->epsfac);
 436     charge           = mdatoms->chargeA;
 437     nvdwtype         = fr->ntype;
 438     vdwparam         = fr->nbfp;
 439     vdwtype          = mdatoms->typeA;
 440
 441     vftab            = kernel_data->table_vdw->data;
 442     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 443
 444     /* Avoid stupid compiler warnings */
 445     jnrA = jnrB = 0;
 446     j_coord_offsetA = 0;
 447     j_coord_offsetB = 0;
 448
 449     outeriter        = 0;
 450     inneriter        = 0;
 451
 452     /* Start outer loop over neighborlists */
 453     for(iidx=0; iidx<nri; iidx++)
 454     {
 455         /* Load shift vector for this list */
 456         i_shift_offset   = DIM*shiftidx[iidx];
 457
 458         /* Load limits for loop over neighbors */
 459         j_index_start    = jindex[iidx];
 460         j_index_end      = jindex[iidx+1];
 461
 462         /* Get outer coordinate index */
 463         inr              = iinr[iidx];
 464         i_coord_offset   = DIM*inr;
 465
 466         /* Load i particle coords and add shift vector */
 467         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 468
 469         fix0             = _mm_setzero_pd();
 470         fiy0             = _mm_setzero_pd();
 471         fiz0             = _mm_setzero_pd();
 472
 473         /* Load parameters for i particles */
 474         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 475         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 476
 477         /* Start inner kernel loop */
 478         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 479         {
 480
 481             /* Get j neighbor index, and coordinate index */
 482             jnrA             = jjnr[jidx];
 483             jnrB             = jjnr[jidx+1];
 484             j_coord_offsetA  = DIM*jnrA;
 485             j_coord_offsetB  = DIM*jnrB;
 486
 487             /* load j atom coordinates */
 488             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 489                                               &jx0,&jy0,&jz0);
 490
 491             /* Calculate displacement vector */
 492             dx00             = _mm_sub_pd(ix0,jx0);
 493             dy00             = _mm_sub_pd(iy0,jy0);
 494             dz00             = _mm_sub_pd(iz0,jz0);
 495
 496             /* Calculate squared distance and things based on it */
 497             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 498
 499             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 500
 501             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 502
 503             /* Load parameters for j particles */
 504             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 505             vdwjidx0A        = 2*vdwtype[jnrA+0];
 506             vdwjidx0B        = 2*vdwtype[jnrB+0];
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             r00              = _mm_mul_pd(rsq00,rinv00);
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq00             = _mm_mul_pd(iq0,jq0);
 516             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 517                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 518
 519             /* Calculate table index by multiplying r with table scale and truncate to integer */
 520             rt               = _mm_mul_pd(r00,vftabscale);
 521             vfitab           = _mm_cvttpd_epi32(rt);
 522             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 523             vfitab           = _mm_slli_epi32(vfitab,3);
 524
 525             /* COULOMB ELECTROSTATICS */
 526             velec            = _mm_mul_pd(qq00,rinv00);
 527             felec            = _mm_mul_pd(velec,rinvsq00);
 528
 529             /* CUBIC SPLINE TABLE DISPERSION */
 530             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 531             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 532             GMX_MM_TRANSPOSE2_PD(Y,F);
 533             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 534             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 535             GMX_MM_TRANSPOSE2_PD(G,H);
 536             Heps             = _mm_mul_pd(vfeps,H);
 537             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 538             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 539             fvdw6            = _mm_mul_pd(c6_00,FF);
 540
 541             /* CUBIC SPLINE TABLE REPULSION */
 542             vfitab           = _mm_add_epi32(vfitab,ifour);
 543             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 544             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 545             GMX_MM_TRANSPOSE2_PD(Y,F);
 546             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 547             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 548             GMX_MM_TRANSPOSE2_PD(G,H);
 549             Heps             = _mm_mul_pd(vfeps,H);
 550             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 551             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 552             fvdw12           = _mm_mul_pd(c12_00,FF);
 553             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 554
 555             fscal            = _mm_add_pd(felec,fvdw);
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm_mul_pd(fscal,dx00);
 559             ty               = _mm_mul_pd(fscal,dy00);
 560             tz               = _mm_mul_pd(fscal,dz00);
 561
 562             /* Update vectorial force */
 563             fix0             = _mm_add_pd(fix0,tx);
 564             fiy0             = _mm_add_pd(fiy0,ty);
 565             fiz0             = _mm_add_pd(fiz0,tz);
 566
 567             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 568
 569             /* Inner loop uses 54 flops */
 570         }
 571
 572         if(jidx<j_index_end)
 573         {
 574
 575             jnrA             = jjnr[jidx];
 576             j_coord_offsetA  = DIM*jnrA;
 577
 578             /* load j atom coordinates */
 579             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 580                                               &jx0,&jy0,&jz0);
 581
 582             /* Calculate displacement vector */
 583             dx00             = _mm_sub_pd(ix0,jx0);
 584             dy00             = _mm_sub_pd(iy0,jy0);
 585             dz00             = _mm_sub_pd(iz0,jz0);
 586
 587             /* Calculate squared distance and things based on it */
 588             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 589
 590             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 591
 592             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 593
 594             /* Load parameters for j particles */
 595             jq0              = _mm_load_sd(charge+jnrA+0);
 596             vdwjidx0A        = 2*vdwtype[jnrA+0];
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r00              = _mm_mul_pd(rsq00,rinv00);
 603
 604             /* Compute parameters for interactions between i and j atoms */
 605             qq00             = _mm_mul_pd(iq0,jq0);
 606             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 607
 608             /* Calculate table index by multiplying r with table scale and truncate to integer */
 609             rt               = _mm_mul_pd(r00,vftabscale);
 610             vfitab           = _mm_cvttpd_epi32(rt);
 611             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 612             vfitab           = _mm_slli_epi32(vfitab,3);
 613
 614             /* COULOMB ELECTROSTATICS */
 615             velec            = _mm_mul_pd(qq00,rinv00);
 616             felec            = _mm_mul_pd(velec,rinvsq00);
 617
 618             /* CUBIC SPLINE TABLE DISPERSION */
 619             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 620             F                = _mm_setzero_pd();
 621             GMX_MM_TRANSPOSE2_PD(Y,F);
 622             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 623             H                = _mm_setzero_pd();
 624             GMX_MM_TRANSPOSE2_PD(G,H);
 625             Heps             = _mm_mul_pd(vfeps,H);
 626             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 627             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 628             fvdw6            = _mm_mul_pd(c6_00,FF);
 629
 630             /* CUBIC SPLINE TABLE REPULSION */
 631             vfitab           = _mm_add_epi32(vfitab,ifour);
 632             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 633             F                = _mm_setzero_pd();
 634             GMX_MM_TRANSPOSE2_PD(Y,F);
 635             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 636             H                = _mm_setzero_pd();
 637             GMX_MM_TRANSPOSE2_PD(G,H);
 638             Heps             = _mm_mul_pd(vfeps,H);
 639             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 640             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 641             fvdw12           = _mm_mul_pd(c12_00,FF);
 642             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 643
 644             fscal            = _mm_add_pd(felec,fvdw);
 645
 646             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm_mul_pd(fscal,dx00);
 650             ty               = _mm_mul_pd(fscal,dy00);
 651             tz               = _mm_mul_pd(fscal,dz00);
 652
 653             /* Update vectorial force */
 654             fix0             = _mm_add_pd(fix0,tx);
 655             fiy0             = _mm_add_pd(fiy0,ty);
 656             fiz0             = _mm_add_pd(fiz0,tz);
 657
 658             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 659
 660             /* Inner loop uses 54 flops */
 661         }
 662
 663         /* End of innermost loop */
 664
 665         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 666                                               f+i_coord_offset,fshift+i_shift_offset);
 667
 668         /* Increment number of inner iterations */
 669         inneriter                  += j_index_end - j_index_start;
 670
 671         /* Outer loop uses 7 flops */
 672     }
 673
 674     /* Increment number of outer iterations */
 675     outeriter        += nri;
 676
 677     /* Update outer/inner flops */
 678
 679     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*54);
 680 }