src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_sse2_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_double.h"
  34 #include "kernelutil_x86_sse2_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_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_elec->data;
 105     vftabscale       = _mm_set1_pd(kernel_data->table_elec->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_cvtepi32_pd(vfitab));
 190             vfitab           = _mm_slli_epi32(vfitab,2);
 191
 192             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 193             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 194             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 195             GMX_MM_TRANSPOSE2_PD(Y,F);
 196             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 197             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 198             GMX_MM_TRANSPOSE2_PD(G,H);
 199             Heps             = _mm_mul_pd(vfeps,H);
 200             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 201             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 202             velec            = _mm_mul_pd(qq00,VV);
 203             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 204             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 205
 206             /* LENNARD-JONES DISPERSION/REPULSION */
 207
 208             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 209             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 210             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 211             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 212             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 213
 214             /* Update potential sum for this i atom from the interaction with this j atom. */
 215             velecsum         = _mm_add_pd(velecsum,velec);
 216             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 217
 218             fscal            = _mm_add_pd(felec,fvdw);
 219
 220             /* Calculate temporary vectorial force */
 221             tx               = _mm_mul_pd(fscal,dx00);
 222             ty               = _mm_mul_pd(fscal,dy00);
 223             tz               = _mm_mul_pd(fscal,dz00);
 224
 225             /* Update vectorial force */
 226             fix0             = _mm_add_pd(fix0,tx);
 227             fiy0             = _mm_add_pd(fiy0,ty);
 228             fiz0             = _mm_add_pd(fiz0,tz);
 229
 230             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 231
 232             /* Inner loop uses 56 flops */
 233         }
 234
 235         if(jidx<j_index_end)
 236         {
 237
 238             jnrA             = jjnr[jidx];
 239             j_coord_offsetA  = DIM*jnrA;
 240
 241             /* load j atom coordinates */
 242             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 243                                               &jx0,&jy0,&jz0);
 244
 245             /* Calculate displacement vector */
 246             dx00             = _mm_sub_pd(ix0,jx0);
 247             dy00             = _mm_sub_pd(iy0,jy0);
 248             dz00             = _mm_sub_pd(iz0,jz0);
 249
 250             /* Calculate squared distance and things based on it */
 251             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 252
 253             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 254
 255             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 256
 257             /* Load parameters for j particles */
 258             jq0              = _mm_load_sd(charge+jnrA+0);
 259             vdwjidx0A        = 2*vdwtype[jnrA+0];
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             r00              = _mm_mul_pd(rsq00,rinv00);
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq00             = _mm_mul_pd(iq0,jq0);
 269             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 270
 271             /* Calculate table index by multiplying r with table scale and truncate to integer */
 272             rt               = _mm_mul_pd(r00,vftabscale);
 273             vfitab           = _mm_cvttpd_epi32(rt);
 274             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 275             vfitab           = _mm_slli_epi32(vfitab,2);
 276
 277             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 278             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 279             F                = _mm_setzero_pd();
 280             GMX_MM_TRANSPOSE2_PD(Y,F);
 281             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 282             H                = _mm_setzero_pd();
 283             GMX_MM_TRANSPOSE2_PD(G,H);
 284             Heps             = _mm_mul_pd(vfeps,H);
 285             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 286             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 287             velec            = _mm_mul_pd(qq00,VV);
 288             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 289             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 290
 291             /* LENNARD-JONES DISPERSION/REPULSION */
 292
 293             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 294             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 295             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 296             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 297             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 301             velecsum         = _mm_add_pd(velecsum,velec);
 302             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 303             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 304
 305             fscal            = _mm_add_pd(felec,fvdw);
 306
 307             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm_mul_pd(fscal,dx00);
 311             ty               = _mm_mul_pd(fscal,dy00);
 312             tz               = _mm_mul_pd(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm_add_pd(fix0,tx);
 316             fiy0             = _mm_add_pd(fiy0,ty);
 317             fiz0             = _mm_add_pd(fiz0,tz);
 318
 319             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 320
 321             /* Inner loop uses 56 flops */
 322         }
 323
 324         /* End of innermost loop */
 325
 326         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 327                                               f+i_coord_offset,fshift+i_shift_offset);
 328
 329         ggid                        = gid[iidx];
 330         /* Update potential energies */
 331         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 332         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 333
 334         /* Increment number of inner iterations */
 335         inneriter                  += j_index_end - j_index_start;
 336
 337         /* Outer loop uses 9 flops */
 338     }
 339
 340     /* Increment number of outer iterations */
 341     outeriter        += nri;
 342
 343     /* Update outer/inner flops */
 344
 345     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*56);
 346 }
 347 /*
 348  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_double
 349  * Electrostatics interaction: CubicSplineTable
 350  * VdW interaction:            LennardJones
 351  * Geometry:                   Particle-Particle
 352  * Calculate force/pot:        Force
 353  */
 354 void
 355 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_double
 356                     (t_nblist * gmx_restrict                nlist,
 357                      rvec * gmx_restrict                    xx,
 358                      rvec * gmx_restrict                    ff,
 359                      t_forcerec * gmx_restrict              fr,
 360                      t_mdatoms * gmx_restrict               mdatoms,
 361                      nb_kernel_data_t * gmx_restrict        kernel_data,
 362                      t_nrnb * gmx_restrict                  nrnb)
 363 {
 364     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 365      * just 0 for non-waters.
 366      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 367      * jnr indices corresponding to data put in the four positions in the SIMD register.
 368      */
 369     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 370     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 371     int              jnrA,jnrB;
 372     int              j_coord_offsetA,j_coord_offsetB;
 373     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 374     real             rcutoff_scalar;
 375     real             *shiftvec,*fshift,*x,*f;
 376     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 377     int              vdwioffset0;
 378     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 379     int              vdwjidx0A,vdwjidx0B;
 380     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 381     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 382     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 383     real             *charge;
 384     int              nvdwtype;
 385     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 386     int              *vdwtype;
 387     real             *vdwparam;
 388     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 389     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 390     __m128i          vfitab;
 391     __m128i          ifour       = _mm_set1_epi32(4);
 392     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 393     real             *vftab;
 394     __m128d          dummy_mask,cutoff_mask;
 395     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 396     __m128d          one     = _mm_set1_pd(1.0);
 397     __m128d          two     = _mm_set1_pd(2.0);
 398     x                = xx[0];
 399     f                = ff[0];
 400
 401     nri              = nlist->nri;
 402     iinr             = nlist->iinr;
 403     jindex           = nlist->jindex;
 404     jjnr             = nlist->jjnr;
 405     shiftidx         = nlist->shift;
 406     gid              = nlist->gid;
 407     shiftvec         = fr->shift_vec[0];
 408     fshift           = fr->fshift[0];
 409     facel            = _mm_set1_pd(fr->epsfac);
 410     charge           = mdatoms->chargeA;
 411     nvdwtype         = fr->ntype;
 412     vdwparam         = fr->nbfp;
 413     vdwtype          = mdatoms->typeA;
 414
 415     vftab            = kernel_data->table_elec->data;
 416     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 417
 418     /* Avoid stupid compiler warnings */
 419     jnrA = jnrB = 0;
 420     j_coord_offsetA = 0;
 421     j_coord_offsetB = 0;
 422
 423     outeriter        = 0;
 424     inneriter        = 0;
 425
 426     /* Start outer loop over neighborlists */
 427     for(iidx=0; iidx<nri; iidx++)
 428     {
 429         /* Load shift vector for this list */
 430         i_shift_offset   = DIM*shiftidx[iidx];
 431
 432         /* Load limits for loop over neighbors */
 433         j_index_start    = jindex[iidx];
 434         j_index_end      = jindex[iidx+1];
 435
 436         /* Get outer coordinate index */
 437         inr              = iinr[iidx];
 438         i_coord_offset   = DIM*inr;
 439
 440         /* Load i particle coords and add shift vector */
 441         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 442
 443         fix0             = _mm_setzero_pd();
 444         fiy0             = _mm_setzero_pd();
 445         fiz0             = _mm_setzero_pd();
 446
 447         /* Load parameters for i particles */
 448         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 449         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 450
 451         /* Start inner kernel loop */
 452         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 453         {
 454
 455             /* Get j neighbor index, and coordinate index */
 456             jnrA             = jjnr[jidx];
 457             jnrB             = jjnr[jidx+1];
 458             j_coord_offsetA  = DIM*jnrA;
 459             j_coord_offsetB  = DIM*jnrB;
 460
 461             /* load j atom coordinates */
 462             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_pd(ix0,jx0);
 467             dy00             = _mm_sub_pd(iy0,jy0);
 468             dz00             = _mm_sub_pd(iz0,jz0);
 469
 470             /* Calculate squared distance and things based on it */
 471             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 472
 473             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 474
 475             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 476
 477             /* Load parameters for j particles */
 478             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 479             vdwjidx0A        = 2*vdwtype[jnrA+0];
 480             vdwjidx0B        = 2*vdwtype[jnrB+0];
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             r00              = _mm_mul_pd(rsq00,rinv00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq00             = _mm_mul_pd(iq0,jq0);
 490             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 491                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 492
 493             /* Calculate table index by multiplying r with table scale and truncate to integer */
 494             rt               = _mm_mul_pd(r00,vftabscale);
 495             vfitab           = _mm_cvttpd_epi32(rt);
 496             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 497             vfitab           = _mm_slli_epi32(vfitab,2);
 498
 499             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 500             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 501             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 502             GMX_MM_TRANSPOSE2_PD(Y,F);
 503             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 504             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 505             GMX_MM_TRANSPOSE2_PD(G,H);
 506             Heps             = _mm_mul_pd(vfeps,H);
 507             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 508             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 509             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 510
 511             /* LENNARD-JONES DISPERSION/REPULSION */
 512
 513             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 514             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 515
 516             fscal            = _mm_add_pd(felec,fvdw);
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm_mul_pd(fscal,dx00);
 520             ty               = _mm_mul_pd(fscal,dy00);
 521             tz               = _mm_mul_pd(fscal,dz00);
 522
 523             /* Update vectorial force */
 524             fix0             = _mm_add_pd(fix0,tx);
 525             fiy0             = _mm_add_pd(fiy0,ty);
 526             fiz0             = _mm_add_pd(fiz0,tz);
 527
 528             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 529
 530             /* Inner loop uses 47 flops */
 531         }
 532
 533         if(jidx<j_index_end)
 534         {
 535
 536             jnrA             = jjnr[jidx];
 537             j_coord_offsetA  = DIM*jnrA;
 538
 539             /* load j atom coordinates */
 540             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 541                                               &jx0,&jy0,&jz0);
 542
 543             /* Calculate displacement vector */
 544             dx00             = _mm_sub_pd(ix0,jx0);
 545             dy00             = _mm_sub_pd(iy0,jy0);
 546             dz00             = _mm_sub_pd(iz0,jz0);
 547
 548             /* Calculate squared distance and things based on it */
 549             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 550
 551             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 552
 553             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 554
 555             /* Load parameters for j particles */
 556             jq0              = _mm_load_sd(charge+jnrA+0);
 557             vdwjidx0A        = 2*vdwtype[jnrA+0];
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r00              = _mm_mul_pd(rsq00,rinv00);
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             qq00             = _mm_mul_pd(iq0,jq0);
 567             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 568
 569             /* Calculate table index by multiplying r with table scale and truncate to integer */
 570             rt               = _mm_mul_pd(r00,vftabscale);
 571             vfitab           = _mm_cvttpd_epi32(rt);
 572             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 573             vfitab           = _mm_slli_epi32(vfitab,2);
 574
 575             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 576             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 577             F                = _mm_setzero_pd();
 578             GMX_MM_TRANSPOSE2_PD(Y,F);
 579             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 580             H                = _mm_setzero_pd();
 581             GMX_MM_TRANSPOSE2_PD(G,H);
 582             Heps             = _mm_mul_pd(vfeps,H);
 583             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 584             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 585             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 586
 587             /* LENNARD-JONES DISPERSION/REPULSION */
 588
 589             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 590             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 591
 592             fscal            = _mm_add_pd(felec,fvdw);
 593
 594             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm_mul_pd(fscal,dx00);
 598             ty               = _mm_mul_pd(fscal,dy00);
 599             tz               = _mm_mul_pd(fscal,dz00);
 600
 601             /* Update vectorial force */
 602             fix0             = _mm_add_pd(fix0,tx);
 603             fiy0             = _mm_add_pd(fiy0,ty);
 604             fiz0             = _mm_add_pd(fiz0,tz);
 605
 606             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 607
 608             /* Inner loop uses 47 flops */
 609         }
 610
 611         /* End of innermost loop */
 612
 613         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 614                                               f+i_coord_offset,fshift+i_shift_offset);
 615
 616         /* Increment number of inner iterations */
 617         inneriter                  += j_index_end - j_index_start;
 618
 619         /* Outer loop uses 7 flops */
 620     }
 621
 622     /* Increment number of outer iterations */
 623     outeriter        += nri;
 624
 625     /* Update outer/inner flops */
 626
 627     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*47);
 628 }