src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_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_ElecCSTab_VdwLJ_GeomW4P1_VF_sse4_1_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_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              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_elec->data;
 114     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 119     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 120     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 147                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 148
 149         fix0             = _mm_setzero_pd();
 150         fiy0             = _mm_setzero_pd();
 151         fiz0             = _mm_setzero_pd();
 152         fix1             = _mm_setzero_pd();
 153         fiy1             = _mm_setzero_pd();
 154         fiz1             = _mm_setzero_pd();
 155         fix2             = _mm_setzero_pd();
 156         fiy2             = _mm_setzero_pd();
 157         fiz2             = _mm_setzero_pd();
 158         fix3             = _mm_setzero_pd();
 159         fiy3             = _mm_setzero_pd();
 160         fiz3             = _mm_setzero_pd();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184             dx10             = _mm_sub_pd(ix1,jx0);
 185             dy10             = _mm_sub_pd(iy1,jy0);
 186             dz10             = _mm_sub_pd(iz1,jz0);
 187             dx20             = _mm_sub_pd(ix2,jx0);
 188             dy20             = _mm_sub_pd(iy2,jy0);
 189             dz20             = _mm_sub_pd(iz2,jz0);
 190             dx30             = _mm_sub_pd(ix3,jx0);
 191             dy30             = _mm_sub_pd(iy3,jy0);
 192             dz30             = _mm_sub_pd(iz3,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 196             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 197             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 198             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 199
 200             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 201             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 202             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 203
 204             rinvsq00         = gmx_mm_inv_pd(rsq00);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 208             vdwjidx0A        = 2*vdwtype[jnrA+0];
 209             vdwjidx0B        = 2*vdwtype[jnrB+0];
 210
 211             fjx0             = _mm_setzero_pd();
 212             fjy0             = _mm_setzero_pd();
 213             fjz0             = _mm_setzero_pd();
 214
 215             /**************************
 216              * CALCULATE INTERACTIONS *
 217              **************************/
 218
 219             /* Compute parameters for interactions between i and j atoms */
 220             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 221                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 222
 223             /* LENNARD-JONES DISPERSION/REPULSION */
 224
 225             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 226             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 227             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 228             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 229             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 233
 234             fscal            = fvdw;
 235
 236             /* Calculate temporary vectorial force */
 237             tx               = _mm_mul_pd(fscal,dx00);
 238             ty               = _mm_mul_pd(fscal,dy00);
 239             tz               = _mm_mul_pd(fscal,dz00);
 240
 241             /* Update vectorial force */
 242             fix0             = _mm_add_pd(fix0,tx);
 243             fiy0             = _mm_add_pd(fiy0,ty);
 244             fiz0             = _mm_add_pd(fiz0,tz);
 245
 246             fjx0             = _mm_add_pd(fjx0,tx);
 247             fjy0             = _mm_add_pd(fjy0,ty);
 248             fjz0             = _mm_add_pd(fjz0,tz);
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             r10              = _mm_mul_pd(rsq10,rinv10);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq10             = _mm_mul_pd(iq1,jq0);
 258
 259             /* Calculate table index by multiplying r with table scale and truncate to integer */
 260             rt               = _mm_mul_pd(r10,vftabscale);
 261             vfitab           = _mm_cvttpd_epi32(rt);
 262             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 263             vfitab           = _mm_slli_epi32(vfitab,2);
 264
 265             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 266             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 267             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 268             GMX_MM_TRANSPOSE2_PD(Y,F);
 269             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 270             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 271             GMX_MM_TRANSPOSE2_PD(G,H);
 272             Heps             = _mm_mul_pd(vfeps,H);
 273             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 274             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 275             velec            = _mm_mul_pd(qq10,VV);
 276             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 277             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             velecsum         = _mm_add_pd(velecsum,velec);
 281
 282             fscal            = felec;
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_pd(fscal,dx10);
 286             ty               = _mm_mul_pd(fscal,dy10);
 287             tz               = _mm_mul_pd(fscal,dz10);
 288
 289             /* Update vectorial force */
 290             fix1             = _mm_add_pd(fix1,tx);
 291             fiy1             = _mm_add_pd(fiy1,ty);
 292             fiz1             = _mm_add_pd(fiz1,tz);
 293
 294             fjx0             = _mm_add_pd(fjx0,tx);
 295             fjy0             = _mm_add_pd(fjy0,ty);
 296             fjz0             = _mm_add_pd(fjz0,tz);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r20              = _mm_mul_pd(rsq20,rinv20);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq20             = _mm_mul_pd(iq2,jq0);
 306
 307             /* Calculate table index by multiplying r with table scale and truncate to integer */
 308             rt               = _mm_mul_pd(r20,vftabscale);
 309             vfitab           = _mm_cvttpd_epi32(rt);
 310             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 311             vfitab           = _mm_slli_epi32(vfitab,2);
 312
 313             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 314             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 315             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 316             GMX_MM_TRANSPOSE2_PD(Y,F);
 317             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 318             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 319             GMX_MM_TRANSPOSE2_PD(G,H);
 320             Heps             = _mm_mul_pd(vfeps,H);
 321             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 322             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 323             velec            = _mm_mul_pd(qq20,VV);
 324             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 325             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm_add_pd(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_pd(fscal,dx20);
 334             ty               = _mm_mul_pd(fscal,dy20);
 335             tz               = _mm_mul_pd(fscal,dz20);
 336
 337             /* Update vectorial force */
 338             fix2             = _mm_add_pd(fix2,tx);
 339             fiy2             = _mm_add_pd(fiy2,ty);
 340             fiz2             = _mm_add_pd(fiz2,tz);
 341
 342             fjx0             = _mm_add_pd(fjx0,tx);
 343             fjy0             = _mm_add_pd(fjy0,ty);
 344             fjz0             = _mm_add_pd(fjz0,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             r30              = _mm_mul_pd(rsq30,rinv30);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq30             = _mm_mul_pd(iq3,jq0);
 354
 355             /* Calculate table index by multiplying r with table scale and truncate to integer */
 356             rt               = _mm_mul_pd(r30,vftabscale);
 357             vfitab           = _mm_cvttpd_epi32(rt);
 358             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 359             vfitab           = _mm_slli_epi32(vfitab,2);
 360
 361             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 362             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 363             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 364             GMX_MM_TRANSPOSE2_PD(Y,F);
 365             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 366             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 367             GMX_MM_TRANSPOSE2_PD(G,H);
 368             Heps             = _mm_mul_pd(vfeps,H);
 369             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 370             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 371             velec            = _mm_mul_pd(qq30,VV);
 372             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 373             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velecsum         = _mm_add_pd(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm_mul_pd(fscal,dx30);
 382             ty               = _mm_mul_pd(fscal,dy30);
 383             tz               = _mm_mul_pd(fscal,dz30);
 384
 385             /* Update vectorial force */
 386             fix3             = _mm_add_pd(fix3,tx);
 387             fiy3             = _mm_add_pd(fiy3,ty);
 388             fiz3             = _mm_add_pd(fiz3,tz);
 389
 390             fjx0             = _mm_add_pd(fjx0,tx);
 391             fjy0             = _mm_add_pd(fjy0,ty);
 392             fjz0             = _mm_add_pd(fjz0,tz);
 393
 394             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 395
 396             /* Inner loop uses 164 flops */
 397         }
 398
 399         if(jidx<j_index_end)
 400         {
 401
 402             jnrA             = jjnr[jidx];
 403             j_coord_offsetA  = DIM*jnrA;
 404
 405             /* load j atom coordinates */
 406             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 407                                               &jx0,&jy0,&jz0);
 408
 409             /* Calculate displacement vector */
 410             dx00             = _mm_sub_pd(ix0,jx0);
 411             dy00             = _mm_sub_pd(iy0,jy0);
 412             dz00             = _mm_sub_pd(iz0,jz0);
 413             dx10             = _mm_sub_pd(ix1,jx0);
 414             dy10             = _mm_sub_pd(iy1,jy0);
 415             dz10             = _mm_sub_pd(iz1,jz0);
 416             dx20             = _mm_sub_pd(ix2,jx0);
 417             dy20             = _mm_sub_pd(iy2,jy0);
 418             dz20             = _mm_sub_pd(iz2,jz0);
 419             dx30             = _mm_sub_pd(ix3,jx0);
 420             dy30             = _mm_sub_pd(iy3,jy0);
 421             dz30             = _mm_sub_pd(iz3,jz0);
 422
 423             /* Calculate squared distance and things based on it */
 424             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 425             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 426             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 427             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 428
 429             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 430             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 431             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 432
 433             rinvsq00         = gmx_mm_inv_pd(rsq00);
 434
 435             /* Load parameters for j particles */
 436             jq0              = _mm_load_sd(charge+jnrA+0);
 437             vdwjidx0A        = 2*vdwtype[jnrA+0];
 438
 439             fjx0             = _mm_setzero_pd();
 440             fjy0             = _mm_setzero_pd();
 441             fjz0             = _mm_setzero_pd();
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             /* Compute parameters for interactions between i and j atoms */
 448             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 449
 450             /* LENNARD-JONES DISPERSION/REPULSION */
 451
 452             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 453             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 454             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 455             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 456             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 460             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 461
 462             fscal            = fvdw;
 463
 464             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm_mul_pd(fscal,dx00);
 468             ty               = _mm_mul_pd(fscal,dy00);
 469             tz               = _mm_mul_pd(fscal,dz00);
 470
 471             /* Update vectorial force */
 472             fix0             = _mm_add_pd(fix0,tx);
 473             fiy0             = _mm_add_pd(fiy0,ty);
 474             fiz0             = _mm_add_pd(fiz0,tz);
 475
 476             fjx0             = _mm_add_pd(fjx0,tx);
 477             fjy0             = _mm_add_pd(fjy0,ty);
 478             fjz0             = _mm_add_pd(fjz0,tz);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r10              = _mm_mul_pd(rsq10,rinv10);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq10             = _mm_mul_pd(iq1,jq0);
 488
 489             /* Calculate table index by multiplying r with table scale and truncate to integer */
 490             rt               = _mm_mul_pd(r10,vftabscale);
 491             vfitab           = _mm_cvttpd_epi32(rt);
 492             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 493             vfitab           = _mm_slli_epi32(vfitab,2);
 494
 495             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 496             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 497             F                = _mm_setzero_pd();
 498             GMX_MM_TRANSPOSE2_PD(Y,F);
 499             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 500             H                = _mm_setzero_pd();
 501             GMX_MM_TRANSPOSE2_PD(G,H);
 502             Heps             = _mm_mul_pd(vfeps,H);
 503             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 504             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 505             velec            = _mm_mul_pd(qq10,VV);
 506             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 507             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 511             velecsum         = _mm_add_pd(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 516
 517             /* Calculate temporary vectorial force */
 518             tx               = _mm_mul_pd(fscal,dx10);
 519             ty               = _mm_mul_pd(fscal,dy10);
 520             tz               = _mm_mul_pd(fscal,dz10);
 521
 522             /* Update vectorial force */
 523             fix1             = _mm_add_pd(fix1,tx);
 524             fiy1             = _mm_add_pd(fiy1,ty);
 525             fiz1             = _mm_add_pd(fiz1,tz);
 526
 527             fjx0             = _mm_add_pd(fjx0,tx);
 528             fjy0             = _mm_add_pd(fjy0,ty);
 529             fjz0             = _mm_add_pd(fjz0,tz);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r20              = _mm_mul_pd(rsq20,rinv20);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq20             = _mm_mul_pd(iq2,jq0);
 539
 540             /* Calculate table index by multiplying r with table scale and truncate to integer */
 541             rt               = _mm_mul_pd(r20,vftabscale);
 542             vfitab           = _mm_cvttpd_epi32(rt);
 543             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 544             vfitab           = _mm_slli_epi32(vfitab,2);
 545
 546             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 547             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 548             F                = _mm_setzero_pd();
 549             GMX_MM_TRANSPOSE2_PD(Y,F);
 550             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 551             H                = _mm_setzero_pd();
 552             GMX_MM_TRANSPOSE2_PD(G,H);
 553             Heps             = _mm_mul_pd(vfeps,H);
 554             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 555             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 556             velec            = _mm_mul_pd(qq20,VV);
 557             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 558             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 562             velecsum         = _mm_add_pd(velecsum,velec);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm_mul_pd(fscal,dx20);
 570             ty               = _mm_mul_pd(fscal,dy20);
 571             tz               = _mm_mul_pd(fscal,dz20);
 572
 573             /* Update vectorial force */
 574             fix2             = _mm_add_pd(fix2,tx);
 575             fiy2             = _mm_add_pd(fiy2,ty);
 576             fiz2             = _mm_add_pd(fiz2,tz);
 577
 578             fjx0             = _mm_add_pd(fjx0,tx);
 579             fjy0             = _mm_add_pd(fjy0,ty);
 580             fjz0             = _mm_add_pd(fjz0,tz);
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             r30              = _mm_mul_pd(rsq30,rinv30);
 587
 588             /* Compute parameters for interactions between i and j atoms */
 589             qq30             = _mm_mul_pd(iq3,jq0);
 590
 591             /* Calculate table index by multiplying r with table scale and truncate to integer */
 592             rt               = _mm_mul_pd(r30,vftabscale);
 593             vfitab           = _mm_cvttpd_epi32(rt);
 594             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 595             vfitab           = _mm_slli_epi32(vfitab,2);
 596
 597             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 598             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 599             F                = _mm_setzero_pd();
 600             GMX_MM_TRANSPOSE2_PD(Y,F);
 601             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 602             H                = _mm_setzero_pd();
 603             GMX_MM_TRANSPOSE2_PD(G,H);
 604             Heps             = _mm_mul_pd(vfeps,H);
 605             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 606             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 607             velec            = _mm_mul_pd(qq30,VV);
 608             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 609             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 610
 611             /* Update potential sum for this i atom from the interaction with this j atom. */
 612             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 613             velecsum         = _mm_add_pd(velecsum,velec);
 614
 615             fscal            = felec;
 616
 617             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_pd(fscal,dx30);
 621             ty               = _mm_mul_pd(fscal,dy30);
 622             tz               = _mm_mul_pd(fscal,dz30);
 623
 624             /* Update vectorial force */
 625             fix3             = _mm_add_pd(fix3,tx);
 626             fiy3             = _mm_add_pd(fiy3,ty);
 627             fiz3             = _mm_add_pd(fiz3,tz);
 628
 629             fjx0             = _mm_add_pd(fjx0,tx);
 630             fjy0             = _mm_add_pd(fjy0,ty);
 631             fjz0             = _mm_add_pd(fjz0,tz);
 632
 633             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 634
 635             /* Inner loop uses 164 flops */
 636         }
 637
 638         /* End of innermost loop */
 639
 640         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 641                                               f+i_coord_offset,fshift+i_shift_offset);
 642
 643         ggid                        = gid[iidx];
 644         /* Update potential energies */
 645         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 646         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 647
 648         /* Increment number of inner iterations */
 649         inneriter                  += j_index_end - j_index_start;
 650
 651         /* Outer loop uses 26 flops */
 652     }
 653
 654     /* Increment number of outer iterations */
 655     outeriter        += nri;
 656
 657     /* Update outer/inner flops */
 658
 659     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
 660 }
 661 /*
 662  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_double
 663  * Electrostatics interaction: CubicSplineTable
 664  * VdW interaction:            LennardJones
 665  * Geometry:                   Water4-Particle
 666  * Calculate force/pot:        Force
 667  */
 668 void
 669 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_double
 670                     (t_nblist * gmx_restrict                nlist,
 671                      rvec * gmx_restrict                    xx,
 672                      rvec * gmx_restrict                    ff,
 673                      t_forcerec * gmx_restrict              fr,
 674                      t_mdatoms * gmx_restrict               mdatoms,
 675                      nb_kernel_data_t * gmx_restrict        kernel_data,
 676                      t_nrnb * gmx_restrict                  nrnb)
 677 {
 678     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 679      * just 0 for non-waters.
 680      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 681      * jnr indices corresponding to data put in the four positions in the SIMD register.
 682      */
 683     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 684     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 685     int              jnrA,jnrB;
 686     int              j_coord_offsetA,j_coord_offsetB;
 687     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 688     real             rcutoff_scalar;
 689     real             *shiftvec,*fshift,*x,*f;
 690     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 691     int              vdwioffset0;
 692     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 693     int              vdwioffset1;
 694     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 695     int              vdwioffset2;
 696     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 697     int              vdwioffset3;
 698     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 699     int              vdwjidx0A,vdwjidx0B;
 700     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 701     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 702     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 703     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 704     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 705     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 706     real             *charge;
 707     int              nvdwtype;
 708     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 709     int              *vdwtype;
 710     real             *vdwparam;
 711     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 712     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 713     __m128i          vfitab;
 714     __m128i          ifour       = _mm_set1_epi32(4);
 715     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 716     real             *vftab;
 717     __m128d          dummy_mask,cutoff_mask;
 718     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 719     __m128d          one     = _mm_set1_pd(1.0);
 720     __m128d          two     = _mm_set1_pd(2.0);
 721     x                = xx[0];
 722     f                = ff[0];
 723
 724     nri              = nlist->nri;
 725     iinr             = nlist->iinr;
 726     jindex           = nlist->jindex;
 727     jjnr             = nlist->jjnr;
 728     shiftidx         = nlist->shift;
 729     gid              = nlist->gid;
 730     shiftvec         = fr->shift_vec[0];
 731     fshift           = fr->fshift[0];
 732     facel            = _mm_set1_pd(fr->epsfac);
 733     charge           = mdatoms->chargeA;
 734     nvdwtype         = fr->ntype;
 735     vdwparam         = fr->nbfp;
 736     vdwtype          = mdatoms->typeA;
 737
 738     vftab            = kernel_data->table_elec->data;
 739     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 740
 741     /* Setup water-specific parameters */
 742     inr              = nlist->iinr[0];
 743     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 744     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 745     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 746     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 747
 748     /* Avoid stupid compiler warnings */
 749     jnrA = jnrB = 0;
 750     j_coord_offsetA = 0;
 751     j_coord_offsetB = 0;
 752
 753     outeriter        = 0;
 754     inneriter        = 0;
 755
 756     /* Start outer loop over neighborlists */
 757     for(iidx=0; iidx<nri; iidx++)
 758     {
 759         /* Load shift vector for this list */
 760         i_shift_offset   = DIM*shiftidx[iidx];
 761
 762         /* Load limits for loop over neighbors */
 763         j_index_start    = jindex[iidx];
 764         j_index_end      = jindex[iidx+1];
 765
 766         /* Get outer coordinate index */
 767         inr              = iinr[iidx];
 768         i_coord_offset   = DIM*inr;
 769
 770         /* Load i particle coords and add shift vector */
 771         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 772                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 773
 774         fix0             = _mm_setzero_pd();
 775         fiy0             = _mm_setzero_pd();
 776         fiz0             = _mm_setzero_pd();
 777         fix1             = _mm_setzero_pd();
 778         fiy1             = _mm_setzero_pd();
 779         fiz1             = _mm_setzero_pd();
 780         fix2             = _mm_setzero_pd();
 781         fiy2             = _mm_setzero_pd();
 782         fiz2             = _mm_setzero_pd();
 783         fix3             = _mm_setzero_pd();
 784         fiy3             = _mm_setzero_pd();
 785         fiz3             = _mm_setzero_pd();
 786
 787         /* Start inner kernel loop */
 788         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 789         {
 790
 791             /* Get j neighbor index, and coordinate index */
 792             jnrA             = jjnr[jidx];
 793             jnrB             = jjnr[jidx+1];
 794             j_coord_offsetA  = DIM*jnrA;
 795             j_coord_offsetB  = DIM*jnrB;
 796
 797             /* load j atom coordinates */
 798             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 799                                               &jx0,&jy0,&jz0);
 800
 801             /* Calculate displacement vector */
 802             dx00             = _mm_sub_pd(ix0,jx0);
 803             dy00             = _mm_sub_pd(iy0,jy0);
 804             dz00             = _mm_sub_pd(iz0,jz0);
 805             dx10             = _mm_sub_pd(ix1,jx0);
 806             dy10             = _mm_sub_pd(iy1,jy0);
 807             dz10             = _mm_sub_pd(iz1,jz0);
 808             dx20             = _mm_sub_pd(ix2,jx0);
 809             dy20             = _mm_sub_pd(iy2,jy0);
 810             dz20             = _mm_sub_pd(iz2,jz0);
 811             dx30             = _mm_sub_pd(ix3,jx0);
 812             dy30             = _mm_sub_pd(iy3,jy0);
 813             dz30             = _mm_sub_pd(iz3,jz0);
 814
 815             /* Calculate squared distance and things based on it */
 816             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 817             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 818             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 819             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 820
 821             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 822             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 823             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 824
 825             rinvsq00         = gmx_mm_inv_pd(rsq00);
 826
 827             /* Load parameters for j particles */
 828             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 829             vdwjidx0A        = 2*vdwtype[jnrA+0];
 830             vdwjidx0B        = 2*vdwtype[jnrB+0];
 831
 832             fjx0             = _mm_setzero_pd();
 833             fjy0             = _mm_setzero_pd();
 834             fjz0             = _mm_setzero_pd();
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             /* Compute parameters for interactions between i and j atoms */
 841             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 842                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 843
 844             /* LENNARD-JONES DISPERSION/REPULSION */
 845
 846             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 847             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 848
 849             fscal            = fvdw;
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm_mul_pd(fscal,dx00);
 853             ty               = _mm_mul_pd(fscal,dy00);
 854             tz               = _mm_mul_pd(fscal,dz00);
 855
 856             /* Update vectorial force */
 857             fix0             = _mm_add_pd(fix0,tx);
 858             fiy0             = _mm_add_pd(fiy0,ty);
 859             fiz0             = _mm_add_pd(fiz0,tz);
 860
 861             fjx0             = _mm_add_pd(fjx0,tx);
 862             fjy0             = _mm_add_pd(fjy0,ty);
 863             fjz0             = _mm_add_pd(fjz0,tz);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             r10              = _mm_mul_pd(rsq10,rinv10);
 870
 871             /* Compute parameters for interactions between i and j atoms */
 872             qq10             = _mm_mul_pd(iq1,jq0);
 873
 874             /* Calculate table index by multiplying r with table scale and truncate to integer */
 875             rt               = _mm_mul_pd(r10,vftabscale);
 876             vfitab           = _mm_cvttpd_epi32(rt);
 877             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 878             vfitab           = _mm_slli_epi32(vfitab,2);
 879
 880             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 881             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 882             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 883             GMX_MM_TRANSPOSE2_PD(Y,F);
 884             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 885             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 886             GMX_MM_TRANSPOSE2_PD(G,H);
 887             Heps             = _mm_mul_pd(vfeps,H);
 888             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 889             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 890             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 891
 892             fscal            = felec;
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm_mul_pd(fscal,dx10);
 896             ty               = _mm_mul_pd(fscal,dy10);
 897             tz               = _mm_mul_pd(fscal,dz10);
 898
 899             /* Update vectorial force */
 900             fix1             = _mm_add_pd(fix1,tx);
 901             fiy1             = _mm_add_pd(fiy1,ty);
 902             fiz1             = _mm_add_pd(fiz1,tz);
 903
 904             fjx0             = _mm_add_pd(fjx0,tx);
 905             fjy0             = _mm_add_pd(fjy0,ty);
 906             fjz0             = _mm_add_pd(fjz0,tz);
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             r20              = _mm_mul_pd(rsq20,rinv20);
 913
 914             /* Compute parameters for interactions between i and j atoms */
 915             qq20             = _mm_mul_pd(iq2,jq0);
 916
 917             /* Calculate table index by multiplying r with table scale and truncate to integer */
 918             rt               = _mm_mul_pd(r20,vftabscale);
 919             vfitab           = _mm_cvttpd_epi32(rt);
 920             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 921             vfitab           = _mm_slli_epi32(vfitab,2);
 922
 923             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 924             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 925             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 926             GMX_MM_TRANSPOSE2_PD(Y,F);
 927             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 928             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 929             GMX_MM_TRANSPOSE2_PD(G,H);
 930             Heps             = _mm_mul_pd(vfeps,H);
 931             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 932             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 933             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 934
 935             fscal            = felec;
 936
 937             /* Calculate temporary vectorial force */
 938             tx               = _mm_mul_pd(fscal,dx20);
 939             ty               = _mm_mul_pd(fscal,dy20);
 940             tz               = _mm_mul_pd(fscal,dz20);
 941
 942             /* Update vectorial force */
 943             fix2             = _mm_add_pd(fix2,tx);
 944             fiy2             = _mm_add_pd(fiy2,ty);
 945             fiz2             = _mm_add_pd(fiz2,tz);
 946
 947             fjx0             = _mm_add_pd(fjx0,tx);
 948             fjy0             = _mm_add_pd(fjy0,ty);
 949             fjz0             = _mm_add_pd(fjz0,tz);
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             r30              = _mm_mul_pd(rsq30,rinv30);
 956
 957             /* Compute parameters for interactions between i and j atoms */
 958             qq30             = _mm_mul_pd(iq3,jq0);
 959
 960             /* Calculate table index by multiplying r with table scale and truncate to integer */
 961             rt               = _mm_mul_pd(r30,vftabscale);
 962             vfitab           = _mm_cvttpd_epi32(rt);
 963             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 964             vfitab           = _mm_slli_epi32(vfitab,2);
 965
 966             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 967             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 968             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 969             GMX_MM_TRANSPOSE2_PD(Y,F);
 970             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 971             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 972             GMX_MM_TRANSPOSE2_PD(G,H);
 973             Heps             = _mm_mul_pd(vfeps,H);
 974             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 975             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 976             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 977
 978             fscal            = felec;
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm_mul_pd(fscal,dx30);
 982             ty               = _mm_mul_pd(fscal,dy30);
 983             tz               = _mm_mul_pd(fscal,dz30);
 984
 985             /* Update vectorial force */
 986             fix3             = _mm_add_pd(fix3,tx);
 987             fiy3             = _mm_add_pd(fiy3,ty);
 988             fiz3             = _mm_add_pd(fiz3,tz);
 989
 990             fjx0             = _mm_add_pd(fjx0,tx);
 991             fjy0             = _mm_add_pd(fjy0,ty);
 992             fjz0             = _mm_add_pd(fjz0,tz);
 993
 994             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 995
 996             /* Inner loop uses 147 flops */
 997         }
 998
 999         if(jidx<j_index_end)
1000         {
1001
1002             jnrA             = jjnr[jidx];
1003             j_coord_offsetA  = DIM*jnrA;
1004
1005             /* load j atom coordinates */
1006             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1007                                               &jx0,&jy0,&jz0);
1008
1009             /* Calculate displacement vector */
1010             dx00             = _mm_sub_pd(ix0,jx0);
1011             dy00             = _mm_sub_pd(iy0,jy0);
1012             dz00             = _mm_sub_pd(iz0,jz0);
1013             dx10             = _mm_sub_pd(ix1,jx0);
1014             dy10             = _mm_sub_pd(iy1,jy0);
1015             dz10             = _mm_sub_pd(iz1,jz0);
1016             dx20             = _mm_sub_pd(ix2,jx0);
1017             dy20             = _mm_sub_pd(iy2,jy0);
1018             dz20             = _mm_sub_pd(iz2,jz0);
1019             dx30             = _mm_sub_pd(ix3,jx0);
1020             dy30             = _mm_sub_pd(iy3,jy0);
1021             dz30             = _mm_sub_pd(iz3,jz0);
1022
1023             /* Calculate squared distance and things based on it */
1024             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1025             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1026             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1027             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1028
1029             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1030             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1031             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1032
1033             rinvsq00         = gmx_mm_inv_pd(rsq00);
1034
1035             /* Load parameters for j particles */
1036             jq0              = _mm_load_sd(charge+jnrA+0);
1037             vdwjidx0A        = 2*vdwtype[jnrA+0];
1038
1039             fjx0             = _mm_setzero_pd();
1040             fjy0             = _mm_setzero_pd();
1041             fjz0             = _mm_setzero_pd();
1042
1043             /**************************
1044              * CALCULATE INTERACTIONS *
1045              **************************/
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1049
1050             /* LENNARD-JONES DISPERSION/REPULSION */
1051
1052             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1053             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1054
1055             fscal            = fvdw;
1056
1057             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1058
1059             /* Calculate temporary vectorial force */
1060             tx               = _mm_mul_pd(fscal,dx00);
1061             ty               = _mm_mul_pd(fscal,dy00);
1062             tz               = _mm_mul_pd(fscal,dz00);
1063
1064             /* Update vectorial force */
1065             fix0             = _mm_add_pd(fix0,tx);
1066             fiy0             = _mm_add_pd(fiy0,ty);
1067             fiz0             = _mm_add_pd(fiz0,tz);
1068
1069             fjx0             = _mm_add_pd(fjx0,tx);
1070             fjy0             = _mm_add_pd(fjy0,ty);
1071             fjz0             = _mm_add_pd(fjz0,tz);
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             r10              = _mm_mul_pd(rsq10,rinv10);
1078
1079             /* Compute parameters for interactions between i and j atoms */
1080             qq10             = _mm_mul_pd(iq1,jq0);
1081
1082             /* Calculate table index by multiplying r with table scale and truncate to integer */
1083             rt               = _mm_mul_pd(r10,vftabscale);
1084             vfitab           = _mm_cvttpd_epi32(rt);
1085             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1086             vfitab           = _mm_slli_epi32(vfitab,2);
1087
1088             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1089             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1090             F                = _mm_setzero_pd();
1091             GMX_MM_TRANSPOSE2_PD(Y,F);
1092             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1093             H                = _mm_setzero_pd();
1094             GMX_MM_TRANSPOSE2_PD(G,H);
1095             Heps             = _mm_mul_pd(vfeps,H);
1096             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1097             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1098             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1099
1100             fscal            = felec;
1101
1102             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1103
1104             /* Calculate temporary vectorial force */
1105             tx               = _mm_mul_pd(fscal,dx10);
1106             ty               = _mm_mul_pd(fscal,dy10);
1107             tz               = _mm_mul_pd(fscal,dz10);
1108
1109             /* Update vectorial force */
1110             fix1             = _mm_add_pd(fix1,tx);
1111             fiy1             = _mm_add_pd(fiy1,ty);
1112             fiz1             = _mm_add_pd(fiz1,tz);
1113
1114             fjx0             = _mm_add_pd(fjx0,tx);
1115             fjy0             = _mm_add_pd(fjy0,ty);
1116             fjz0             = _mm_add_pd(fjz0,tz);
1117
1118             /**************************
1119              * CALCULATE INTERACTIONS *
1120              **************************/
1121
1122             r20              = _mm_mul_pd(rsq20,rinv20);
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             qq20             = _mm_mul_pd(iq2,jq0);
1126
1127             /* Calculate table index by multiplying r with table scale and truncate to integer */
1128             rt               = _mm_mul_pd(r20,vftabscale);
1129             vfitab           = _mm_cvttpd_epi32(rt);
1130             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1131             vfitab           = _mm_slli_epi32(vfitab,2);
1132
1133             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1134             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1135             F                = _mm_setzero_pd();
1136             GMX_MM_TRANSPOSE2_PD(Y,F);
1137             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1138             H                = _mm_setzero_pd();
1139             GMX_MM_TRANSPOSE2_PD(G,H);
1140             Heps             = _mm_mul_pd(vfeps,H);
1141             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1142             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1143             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1144
1145             fscal            = felec;
1146
1147             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1148
1149             /* Calculate temporary vectorial force */
1150             tx               = _mm_mul_pd(fscal,dx20);
1151             ty               = _mm_mul_pd(fscal,dy20);
1152             tz               = _mm_mul_pd(fscal,dz20);
1153
1154             /* Update vectorial force */
1155             fix2             = _mm_add_pd(fix2,tx);
1156             fiy2             = _mm_add_pd(fiy2,ty);
1157             fiz2             = _mm_add_pd(fiz2,tz);
1158
1159             fjx0             = _mm_add_pd(fjx0,tx);
1160             fjy0             = _mm_add_pd(fjy0,ty);
1161             fjz0             = _mm_add_pd(fjz0,tz);
1162
1163             /**************************
1164              * CALCULATE INTERACTIONS *
1165              **************************/
1166
1167             r30              = _mm_mul_pd(rsq30,rinv30);
1168
1169             /* Compute parameters for interactions between i and j atoms */
1170             qq30             = _mm_mul_pd(iq3,jq0);
1171
1172             /* Calculate table index by multiplying r with table scale and truncate to integer */
1173             rt               = _mm_mul_pd(r30,vftabscale);
1174             vfitab           = _mm_cvttpd_epi32(rt);
1175             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1176             vfitab           = _mm_slli_epi32(vfitab,2);
1177
1178             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1179             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1180             F                = _mm_setzero_pd();
1181             GMX_MM_TRANSPOSE2_PD(Y,F);
1182             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1183             H                = _mm_setzero_pd();
1184             GMX_MM_TRANSPOSE2_PD(G,H);
1185             Heps             = _mm_mul_pd(vfeps,H);
1186             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1187             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1188             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1189
1190             fscal            = felec;
1191
1192             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1193
1194             /* Calculate temporary vectorial force */
1195             tx               = _mm_mul_pd(fscal,dx30);
1196             ty               = _mm_mul_pd(fscal,dy30);
1197             tz               = _mm_mul_pd(fscal,dz30);
1198
1199             /* Update vectorial force */
1200             fix3             = _mm_add_pd(fix3,tx);
1201             fiy3             = _mm_add_pd(fiy3,ty);
1202             fiz3             = _mm_add_pd(fiz3,tz);
1203
1204             fjx0             = _mm_add_pd(fjx0,tx);
1205             fjy0             = _mm_add_pd(fjy0,ty);
1206             fjz0             = _mm_add_pd(fjz0,tz);
1207
1208             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1209
1210             /* Inner loop uses 147 flops */
1211         }
1212
1213         /* End of innermost loop */
1214
1215         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1216                                               f+i_coord_offset,fshift+i_shift_offset);
1217
1218         /* Increment number of inner iterations */
1219         inneriter                  += j_index_end - j_index_start;
1220
1221         /* Outer loop uses 24 flops */
1222     }
1223
1224     /* Increment number of outer iterations */
1225     outeriter        += nri;
1226
1227     /* Update outer/inner flops */
1228
1229     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1230 }