src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwLJSw_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_ElecRFCut_VdwLJSw_GeomW4P1_VF_sse4_1_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_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     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  89     real             rswitch_scalar,d_scalar;
  90     __m128d          dummy_mask,cutoff_mask;
  91     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  92     __m128d          one     = _mm_set1_pd(1.0);
  93     __m128d          two     = _mm_set1_pd(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_pd(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     krf              = _mm_set1_pd(fr->ic->k_rf);
 108     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 109     crf              = _mm_set1_pd(fr->ic->c_rf);
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 117     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 118     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 119     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 120
 121     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 122     rcutoff_scalar   = fr->rcoulomb;
 123     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 124     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 125
 126     rswitch_scalar   = fr->rvdw_switch;
 127     rswitch          = _mm_set1_pd(rswitch_scalar);
 128     /* Setup switch parameters */
 129     d_scalar         = rcutoff_scalar-rswitch_scalar;
 130     d                = _mm_set1_pd(d_scalar);
 131     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 132     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 133     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 134     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 135     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 163
 164         fix0             = _mm_setzero_pd();
 165         fiy0             = _mm_setzero_pd();
 166         fiz0             = _mm_setzero_pd();
 167         fix1             = _mm_setzero_pd();
 168         fiy1             = _mm_setzero_pd();
 169         fiz1             = _mm_setzero_pd();
 170         fix2             = _mm_setzero_pd();
 171         fiy2             = _mm_setzero_pd();
 172         fiz2             = _mm_setzero_pd();
 173         fix3             = _mm_setzero_pd();
 174         fiy3             = _mm_setzero_pd();
 175         fiz3             = _mm_setzero_pd();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_pd();
 179         vvdwsum          = _mm_setzero_pd();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_pd(ix0,jx0);
 197             dy00             = _mm_sub_pd(iy0,jy0);
 198             dz00             = _mm_sub_pd(iz0,jz0);
 199             dx10             = _mm_sub_pd(ix1,jx0);
 200             dy10             = _mm_sub_pd(iy1,jy0);
 201             dz10             = _mm_sub_pd(iz1,jz0);
 202             dx20             = _mm_sub_pd(ix2,jx0);
 203             dy20             = _mm_sub_pd(iy2,jy0);
 204             dz20             = _mm_sub_pd(iz2,jz0);
 205             dx30             = _mm_sub_pd(ix3,jx0);
 206             dy30             = _mm_sub_pd(iy3,jy0);
 207             dz30             = _mm_sub_pd(iz3,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 213             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 214
 215             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 216             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 217             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 218             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 219
 220             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 221             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 223             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229
 230             fjx0             = _mm_setzero_pd();
 231             fjy0             = _mm_setzero_pd();
 232             fjz0             = _mm_setzero_pd();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             if (gmx_mm_any_lt(rsq00,rcutoff2))
 239             {
 240
 241             r00              = _mm_mul_pd(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 245                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 246
 247             /* LENNARD-JONES DISPERSION/REPULSION */
 248
 249             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 250             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 251             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 252             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 253             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 254
 255             d                = _mm_sub_pd(r00,rswitch);
 256             d                = _mm_max_pd(d,_mm_setzero_pd());
 257             d2               = _mm_mul_pd(d,d);
 258             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 259
 260             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 261
 262             /* Evaluate switch function */
 263             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 264             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 265             vvdw             = _mm_mul_pd(vvdw,sw);
 266             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 270             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 271
 272             fscal            = fvdw;
 273
 274             fscal            = _mm_and_pd(fscal,cutoff_mask);
 275
 276             /* Calculate temporary vectorial force */
 277             tx               = _mm_mul_pd(fscal,dx00);
 278             ty               = _mm_mul_pd(fscal,dy00);
 279             tz               = _mm_mul_pd(fscal,dz00);
 280
 281             /* Update vectorial force */
 282             fix0             = _mm_add_pd(fix0,tx);
 283             fiy0             = _mm_add_pd(fiy0,ty);
 284             fiz0             = _mm_add_pd(fiz0,tz);
 285
 286             fjx0             = _mm_add_pd(fjx0,tx);
 287             fjy0             = _mm_add_pd(fjy0,ty);
 288             fjz0             = _mm_add_pd(fjz0,tz);
 289
 290             }
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             if (gmx_mm_any_lt(rsq10,rcutoff2))
 297             {
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq10             = _mm_mul_pd(iq1,jq0);
 301
 302             /* REACTION-FIELD ELECTROSTATICS */
 303             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 304             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 305
 306             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velec            = _mm_and_pd(velec,cutoff_mask);
 310             velecsum         = _mm_add_pd(velecsum,velec);
 311
 312             fscal            = felec;
 313
 314             fscal            = _mm_and_pd(fscal,cutoff_mask);
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_pd(fscal,dx10);
 318             ty               = _mm_mul_pd(fscal,dy10);
 319             tz               = _mm_mul_pd(fscal,dz10);
 320
 321             /* Update vectorial force */
 322             fix1             = _mm_add_pd(fix1,tx);
 323             fiy1             = _mm_add_pd(fiy1,ty);
 324             fiz1             = _mm_add_pd(fiz1,tz);
 325
 326             fjx0             = _mm_add_pd(fjx0,tx);
 327             fjy0             = _mm_add_pd(fjy0,ty);
 328             fjz0             = _mm_add_pd(fjz0,tz);
 329
 330             }
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             if (gmx_mm_any_lt(rsq20,rcutoff2))
 337             {
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq20             = _mm_mul_pd(iq2,jq0);
 341
 342             /* REACTION-FIELD ELECTROSTATICS */
 343             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 344             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 345
 346             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velec            = _mm_and_pd(velec,cutoff_mask);
 350             velecsum         = _mm_add_pd(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354             fscal            = _mm_and_pd(fscal,cutoff_mask);
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm_mul_pd(fscal,dx20);
 358             ty               = _mm_mul_pd(fscal,dy20);
 359             tz               = _mm_mul_pd(fscal,dz20);
 360
 361             /* Update vectorial force */
 362             fix2             = _mm_add_pd(fix2,tx);
 363             fiy2             = _mm_add_pd(fiy2,ty);
 364             fiz2             = _mm_add_pd(fiz2,tz);
 365
 366             fjx0             = _mm_add_pd(fjx0,tx);
 367             fjy0             = _mm_add_pd(fjy0,ty);
 368             fjz0             = _mm_add_pd(fjz0,tz);
 369
 370             }
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             if (gmx_mm_any_lt(rsq30,rcutoff2))
 377             {
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq30             = _mm_mul_pd(iq3,jq0);
 381
 382             /* REACTION-FIELD ELECTROSTATICS */
 383             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 384             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 385
 386             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velec            = _mm_and_pd(velec,cutoff_mask);
 390             velecsum         = _mm_add_pd(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             fscal            = _mm_and_pd(fscal,cutoff_mask);
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm_mul_pd(fscal,dx30);
 398             ty               = _mm_mul_pd(fscal,dy30);
 399             tz               = _mm_mul_pd(fscal,dz30);
 400
 401             /* Update vectorial force */
 402             fix3             = _mm_add_pd(fix3,tx);
 403             fiy3             = _mm_add_pd(fiy3,ty);
 404             fiz3             = _mm_add_pd(fiz3,tz);
 405
 406             fjx0             = _mm_add_pd(fjx0,tx);
 407             fjy0             = _mm_add_pd(fjy0,ty);
 408             fjz0             = _mm_add_pd(fjz0,tz);
 409
 410             }
 411
 412             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 413
 414             /* Inner loop uses 170 flops */
 415         }
 416
 417         if(jidx<j_index_end)
 418         {
 419
 420             jnrA             = jjnr[jidx];
 421             j_coord_offsetA  = DIM*jnrA;
 422
 423             /* load j atom coordinates */
 424             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 425                                               &jx0,&jy0,&jz0);
 426
 427             /* Calculate displacement vector */
 428             dx00             = _mm_sub_pd(ix0,jx0);
 429             dy00             = _mm_sub_pd(iy0,jy0);
 430             dz00             = _mm_sub_pd(iz0,jz0);
 431             dx10             = _mm_sub_pd(ix1,jx0);
 432             dy10             = _mm_sub_pd(iy1,jy0);
 433             dz10             = _mm_sub_pd(iz1,jz0);
 434             dx20             = _mm_sub_pd(ix2,jx0);
 435             dy20             = _mm_sub_pd(iy2,jy0);
 436             dz20             = _mm_sub_pd(iz2,jz0);
 437             dx30             = _mm_sub_pd(ix3,jx0);
 438             dy30             = _mm_sub_pd(iy3,jy0);
 439             dz30             = _mm_sub_pd(iz3,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 443             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 444             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 445             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 446
 447             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 448             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 449             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 450             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 451
 452             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 453             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 454             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 455             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 456
 457             /* Load parameters for j particles */
 458             jq0              = _mm_load_sd(charge+jnrA+0);
 459             vdwjidx0A        = 2*vdwtype[jnrA+0];
 460
 461             fjx0             = _mm_setzero_pd();
 462             fjy0             = _mm_setzero_pd();
 463             fjz0             = _mm_setzero_pd();
 464
 465             /**************************
 466              * CALCULATE INTERACTIONS *
 467              **************************/
 468
 469             if (gmx_mm_any_lt(rsq00,rcutoff2))
 470             {
 471
 472             r00              = _mm_mul_pd(rsq00,rinv00);
 473
 474             /* Compute parameters for interactions between i and j atoms */
 475             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 476
 477             /* LENNARD-JONES DISPERSION/REPULSION */
 478
 479             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 480             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 481             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 482             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 483             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 484
 485             d                = _mm_sub_pd(r00,rswitch);
 486             d                = _mm_max_pd(d,_mm_setzero_pd());
 487             d2               = _mm_mul_pd(d,d);
 488             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 489
 490             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 491
 492             /* Evaluate switch function */
 493             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 494             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 495             vvdw             = _mm_mul_pd(vvdw,sw);
 496             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 497
 498             /* Update potential sum for this i atom from the interaction with this j atom. */
 499             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 500             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 501             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 502
 503             fscal            = fvdw;
 504
 505             fscal            = _mm_and_pd(fscal,cutoff_mask);
 506
 507             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm_mul_pd(fscal,dx00);
 511             ty               = _mm_mul_pd(fscal,dy00);
 512             tz               = _mm_mul_pd(fscal,dz00);
 513
 514             /* Update vectorial force */
 515             fix0             = _mm_add_pd(fix0,tx);
 516             fiy0             = _mm_add_pd(fiy0,ty);
 517             fiz0             = _mm_add_pd(fiz0,tz);
 518
 519             fjx0             = _mm_add_pd(fjx0,tx);
 520             fjy0             = _mm_add_pd(fjy0,ty);
 521             fjz0             = _mm_add_pd(fjz0,tz);
 522
 523             }
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             if (gmx_mm_any_lt(rsq10,rcutoff2))
 530             {
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq10             = _mm_mul_pd(iq1,jq0);
 534
 535             /* REACTION-FIELD ELECTROSTATICS */
 536             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 537             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 538
 539             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velec            = _mm_and_pd(velec,cutoff_mask);
 543             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 544             velecsum         = _mm_add_pd(velecsum,velec);
 545
 546             fscal            = felec;
 547
 548             fscal            = _mm_and_pd(fscal,cutoff_mask);
 549
 550             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm_mul_pd(fscal,dx10);
 554             ty               = _mm_mul_pd(fscal,dy10);
 555             tz               = _mm_mul_pd(fscal,dz10);
 556
 557             /* Update vectorial force */
 558             fix1             = _mm_add_pd(fix1,tx);
 559             fiy1             = _mm_add_pd(fiy1,ty);
 560             fiz1             = _mm_add_pd(fiz1,tz);
 561
 562             fjx0             = _mm_add_pd(fjx0,tx);
 563             fjy0             = _mm_add_pd(fjy0,ty);
 564             fjz0             = _mm_add_pd(fjz0,tz);
 565
 566             }
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm_any_lt(rsq20,rcutoff2))
 573             {
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq20             = _mm_mul_pd(iq2,jq0);
 577
 578             /* REACTION-FIELD ELECTROSTATICS */
 579             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 580             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 581
 582             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_and_pd(velec,cutoff_mask);
 586             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 587             velecsum         = _mm_add_pd(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_and_pd(fscal,cutoff_mask);
 592
 593             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_pd(fscal,dx20);
 597             ty               = _mm_mul_pd(fscal,dy20);
 598             tz               = _mm_mul_pd(fscal,dz20);
 599
 600             /* Update vectorial force */
 601             fix2             = _mm_add_pd(fix2,tx);
 602             fiy2             = _mm_add_pd(fiy2,ty);
 603             fiz2             = _mm_add_pd(fiz2,tz);
 604
 605             fjx0             = _mm_add_pd(fjx0,tx);
 606             fjy0             = _mm_add_pd(fjy0,ty);
 607             fjz0             = _mm_add_pd(fjz0,tz);
 608
 609             }
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             if (gmx_mm_any_lt(rsq30,rcutoff2))
 616             {
 617
 618             /* Compute parameters for interactions between i and j atoms */
 619             qq30             = _mm_mul_pd(iq3,jq0);
 620
 621             /* REACTION-FIELD ELECTROSTATICS */
 622             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 623             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 624
 625             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm_and_pd(velec,cutoff_mask);
 629             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 630             velecsum         = _mm_add_pd(velecsum,velec);
 631
 632             fscal            = felec;
 633
 634             fscal            = _mm_and_pd(fscal,cutoff_mask);
 635
 636             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 637
 638             /* Calculate temporary vectorial force */
 639             tx               = _mm_mul_pd(fscal,dx30);
 640             ty               = _mm_mul_pd(fscal,dy30);
 641             tz               = _mm_mul_pd(fscal,dz30);
 642
 643             /* Update vectorial force */
 644             fix3             = _mm_add_pd(fix3,tx);
 645             fiy3             = _mm_add_pd(fiy3,ty);
 646             fiz3             = _mm_add_pd(fiz3,tz);
 647
 648             fjx0             = _mm_add_pd(fjx0,tx);
 649             fjy0             = _mm_add_pd(fjy0,ty);
 650             fjz0             = _mm_add_pd(fjz0,tz);
 651
 652             }
 653
 654             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 655
 656             /* Inner loop uses 170 flops */
 657         }
 658
 659         /* End of innermost loop */
 660
 661         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 662                                               f+i_coord_offset,fshift+i_shift_offset);
 663
 664         ggid                        = gid[iidx];
 665         /* Update potential energies */
 666         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 667         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 668
 669         /* Increment number of inner iterations */
 670         inneriter                  += j_index_end - j_index_start;
 671
 672         /* Outer loop uses 26 flops */
 673     }
 674
 675     /* Increment number of outer iterations */
 676     outeriter        += nri;
 677
 678     /* Update outer/inner flops */
 679
 680     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*170);
 681 }
 682 /*
 683  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_double
 684  * Electrostatics interaction: ReactionField
 685  * VdW interaction:            LennardJones
 686  * Geometry:                   Water4-Particle
 687  * Calculate force/pot:        Force
 688  */
 689 void
 690 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse4_1_double
 691                     (t_nblist * gmx_restrict                nlist,
 692                      rvec * gmx_restrict                    xx,
 693                      rvec * gmx_restrict                    ff,
 694                      t_forcerec * gmx_restrict              fr,
 695                      t_mdatoms * gmx_restrict               mdatoms,
 696                      nb_kernel_data_t * gmx_restrict        kernel_data,
 697                      t_nrnb * gmx_restrict                  nrnb)
 698 {
 699     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 700      * just 0 for non-waters.
 701      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 702      * jnr indices corresponding to data put in the four positions in the SIMD register.
 703      */
 704     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 705     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 706     int              jnrA,jnrB;
 707     int              j_coord_offsetA,j_coord_offsetB;
 708     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 709     real             rcutoff_scalar;
 710     real             *shiftvec,*fshift,*x,*f;
 711     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 712     int              vdwioffset0;
 713     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 714     int              vdwioffset1;
 715     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 716     int              vdwioffset2;
 717     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 718     int              vdwioffset3;
 719     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 720     int              vdwjidx0A,vdwjidx0B;
 721     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 722     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 723     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 724     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 725     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 726     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 727     real             *charge;
 728     int              nvdwtype;
 729     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 730     int              *vdwtype;
 731     real             *vdwparam;
 732     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 733     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 734     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 735     real             rswitch_scalar,d_scalar;
 736     __m128d          dummy_mask,cutoff_mask;
 737     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 738     __m128d          one     = _mm_set1_pd(1.0);
 739     __m128d          two     = _mm_set1_pd(2.0);
 740     x                = xx[0];
 741     f                = ff[0];
 742
 743     nri              = nlist->nri;
 744     iinr             = nlist->iinr;
 745     jindex           = nlist->jindex;
 746     jjnr             = nlist->jjnr;
 747     shiftidx         = nlist->shift;
 748     gid              = nlist->gid;
 749     shiftvec         = fr->shift_vec[0];
 750     fshift           = fr->fshift[0];
 751     facel            = _mm_set1_pd(fr->epsfac);
 752     charge           = mdatoms->chargeA;
 753     krf              = _mm_set1_pd(fr->ic->k_rf);
 754     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 755     crf              = _mm_set1_pd(fr->ic->c_rf);
 756     nvdwtype         = fr->ntype;
 757     vdwparam         = fr->nbfp;
 758     vdwtype          = mdatoms->typeA;
 759
 760     /* Setup water-specific parameters */
 761     inr              = nlist->iinr[0];
 762     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 763     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 764     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 765     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 766
 767     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 768     rcutoff_scalar   = fr->rcoulomb;
 769     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 770     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 771
 772     rswitch_scalar   = fr->rvdw_switch;
 773     rswitch          = _mm_set1_pd(rswitch_scalar);
 774     /* Setup switch parameters */
 775     d_scalar         = rcutoff_scalar-rswitch_scalar;
 776     d                = _mm_set1_pd(d_scalar);
 777     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 778     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 779     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 780     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 781     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 782     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 783
 784     /* Avoid stupid compiler warnings */
 785     jnrA = jnrB = 0;
 786     j_coord_offsetA = 0;
 787     j_coord_offsetB = 0;
 788
 789     outeriter        = 0;
 790     inneriter        = 0;
 791
 792     /* Start outer loop over neighborlists */
 793     for(iidx=0; iidx<nri; iidx++)
 794     {
 795         /* Load shift vector for this list */
 796         i_shift_offset   = DIM*shiftidx[iidx];
 797
 798         /* Load limits for loop over neighbors */
 799         j_index_start    = jindex[iidx];
 800         j_index_end      = jindex[iidx+1];
 801
 802         /* Get outer coordinate index */
 803         inr              = iinr[iidx];
 804         i_coord_offset   = DIM*inr;
 805
 806         /* Load i particle coords and add shift vector */
 807         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 808                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 809
 810         fix0             = _mm_setzero_pd();
 811         fiy0             = _mm_setzero_pd();
 812         fiz0             = _mm_setzero_pd();
 813         fix1             = _mm_setzero_pd();
 814         fiy1             = _mm_setzero_pd();
 815         fiz1             = _mm_setzero_pd();
 816         fix2             = _mm_setzero_pd();
 817         fiy2             = _mm_setzero_pd();
 818         fiz2             = _mm_setzero_pd();
 819         fix3             = _mm_setzero_pd();
 820         fiy3             = _mm_setzero_pd();
 821         fiz3             = _mm_setzero_pd();
 822
 823         /* Start inner kernel loop */
 824         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 825         {
 826
 827             /* Get j neighbor index, and coordinate index */
 828             jnrA             = jjnr[jidx];
 829             jnrB             = jjnr[jidx+1];
 830             j_coord_offsetA  = DIM*jnrA;
 831             j_coord_offsetB  = DIM*jnrB;
 832
 833             /* load j atom coordinates */
 834             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 835                                               &jx0,&jy0,&jz0);
 836
 837             /* Calculate displacement vector */
 838             dx00             = _mm_sub_pd(ix0,jx0);
 839             dy00             = _mm_sub_pd(iy0,jy0);
 840             dz00             = _mm_sub_pd(iz0,jz0);
 841             dx10             = _mm_sub_pd(ix1,jx0);
 842             dy10             = _mm_sub_pd(iy1,jy0);
 843             dz10             = _mm_sub_pd(iz1,jz0);
 844             dx20             = _mm_sub_pd(ix2,jx0);
 845             dy20             = _mm_sub_pd(iy2,jy0);
 846             dz20             = _mm_sub_pd(iz2,jz0);
 847             dx30             = _mm_sub_pd(ix3,jx0);
 848             dy30             = _mm_sub_pd(iy3,jy0);
 849             dz30             = _mm_sub_pd(iz3,jz0);
 850
 851             /* Calculate squared distance and things based on it */
 852             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 853             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 854             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 855             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 856
 857             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 858             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 859             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 860             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 861
 862             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 863             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 864             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 865             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 866
 867             /* Load parameters for j particles */
 868             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 869             vdwjidx0A        = 2*vdwtype[jnrA+0];
 870             vdwjidx0B        = 2*vdwtype[jnrB+0];
 871
 872             fjx0             = _mm_setzero_pd();
 873             fjy0             = _mm_setzero_pd();
 874             fjz0             = _mm_setzero_pd();
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             if (gmx_mm_any_lt(rsq00,rcutoff2))
 881             {
 882
 883             r00              = _mm_mul_pd(rsq00,rinv00);
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 887                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 888
 889             /* LENNARD-JONES DISPERSION/REPULSION */
 890
 891             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 892             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 893             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 894             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 895             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 896
 897             d                = _mm_sub_pd(r00,rswitch);
 898             d                = _mm_max_pd(d,_mm_setzero_pd());
 899             d2               = _mm_mul_pd(d,d);
 900             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 901
 902             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 903
 904             /* Evaluate switch function */
 905             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 906             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 907             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 908
 909             fscal            = fvdw;
 910
 911             fscal            = _mm_and_pd(fscal,cutoff_mask);
 912
 913             /* Calculate temporary vectorial force */
 914             tx               = _mm_mul_pd(fscal,dx00);
 915             ty               = _mm_mul_pd(fscal,dy00);
 916             tz               = _mm_mul_pd(fscal,dz00);
 917
 918             /* Update vectorial force */
 919             fix0             = _mm_add_pd(fix0,tx);
 920             fiy0             = _mm_add_pd(fiy0,ty);
 921             fiz0             = _mm_add_pd(fiz0,tz);
 922
 923             fjx0             = _mm_add_pd(fjx0,tx);
 924             fjy0             = _mm_add_pd(fjy0,ty);
 925             fjz0             = _mm_add_pd(fjz0,tz);
 926
 927             }
 928
 929             /**************************
 930              * CALCULATE INTERACTIONS *
 931              **************************/
 932
 933             if (gmx_mm_any_lt(rsq10,rcutoff2))
 934             {
 935
 936             /* Compute parameters for interactions between i and j atoms */
 937             qq10             = _mm_mul_pd(iq1,jq0);
 938
 939             /* REACTION-FIELD ELECTROSTATICS */
 940             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 941
 942             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 943
 944             fscal            = felec;
 945
 946             fscal            = _mm_and_pd(fscal,cutoff_mask);
 947
 948             /* Calculate temporary vectorial force */
 949             tx               = _mm_mul_pd(fscal,dx10);
 950             ty               = _mm_mul_pd(fscal,dy10);
 951             tz               = _mm_mul_pd(fscal,dz10);
 952
 953             /* Update vectorial force */
 954             fix1             = _mm_add_pd(fix1,tx);
 955             fiy1             = _mm_add_pd(fiy1,ty);
 956             fiz1             = _mm_add_pd(fiz1,tz);
 957
 958             fjx0             = _mm_add_pd(fjx0,tx);
 959             fjy0             = _mm_add_pd(fjy0,ty);
 960             fjz0             = _mm_add_pd(fjz0,tz);
 961
 962             }
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             if (gmx_mm_any_lt(rsq20,rcutoff2))
 969             {
 970
 971             /* Compute parameters for interactions between i and j atoms */
 972             qq20             = _mm_mul_pd(iq2,jq0);
 973
 974             /* REACTION-FIELD ELECTROSTATICS */
 975             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 976
 977             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 978
 979             fscal            = felec;
 980
 981             fscal            = _mm_and_pd(fscal,cutoff_mask);
 982
 983             /* Calculate temporary vectorial force */
 984             tx               = _mm_mul_pd(fscal,dx20);
 985             ty               = _mm_mul_pd(fscal,dy20);
 986             tz               = _mm_mul_pd(fscal,dz20);
 987
 988             /* Update vectorial force */
 989             fix2             = _mm_add_pd(fix2,tx);
 990             fiy2             = _mm_add_pd(fiy2,ty);
 991             fiz2             = _mm_add_pd(fiz2,tz);
 992
 993             fjx0             = _mm_add_pd(fjx0,tx);
 994             fjy0             = _mm_add_pd(fjy0,ty);
 995             fjz0             = _mm_add_pd(fjz0,tz);
 996
 997             }
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             if (gmx_mm_any_lt(rsq30,rcutoff2))
1004             {
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq30             = _mm_mul_pd(iq3,jq0);
1008
1009             /* REACTION-FIELD ELECTROSTATICS */
1010             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1011
1012             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1013
1014             fscal            = felec;
1015
1016             fscal            = _mm_and_pd(fscal,cutoff_mask);
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm_mul_pd(fscal,dx30);
1020             ty               = _mm_mul_pd(fscal,dy30);
1021             tz               = _mm_mul_pd(fscal,dz30);
1022
1023             /* Update vectorial force */
1024             fix3             = _mm_add_pd(fix3,tx);
1025             fiy3             = _mm_add_pd(fiy3,ty);
1026             fiz3             = _mm_add_pd(fiz3,tz);
1027
1028             fjx0             = _mm_add_pd(fjx0,tx);
1029             fjy0             = _mm_add_pd(fjy0,ty);
1030             fjz0             = _mm_add_pd(fjz0,tz);
1031
1032             }
1033
1034             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1035
1036             /* Inner loop uses 149 flops */
1037         }
1038
1039         if(jidx<j_index_end)
1040         {
1041
1042             jnrA             = jjnr[jidx];
1043             j_coord_offsetA  = DIM*jnrA;
1044
1045             /* load j atom coordinates */
1046             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1047                                               &jx0,&jy0,&jz0);
1048
1049             /* Calculate displacement vector */
1050             dx00             = _mm_sub_pd(ix0,jx0);
1051             dy00             = _mm_sub_pd(iy0,jy0);
1052             dz00             = _mm_sub_pd(iz0,jz0);
1053             dx10             = _mm_sub_pd(ix1,jx0);
1054             dy10             = _mm_sub_pd(iy1,jy0);
1055             dz10             = _mm_sub_pd(iz1,jz0);
1056             dx20             = _mm_sub_pd(ix2,jx0);
1057             dy20             = _mm_sub_pd(iy2,jy0);
1058             dz20             = _mm_sub_pd(iz2,jz0);
1059             dx30             = _mm_sub_pd(ix3,jx0);
1060             dy30             = _mm_sub_pd(iy3,jy0);
1061             dz30             = _mm_sub_pd(iz3,jz0);
1062
1063             /* Calculate squared distance and things based on it */
1064             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1065             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1066             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1067             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1068
1069             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1070             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1071             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1072             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1073
1074             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1075             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1076             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1077             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1078
1079             /* Load parameters for j particles */
1080             jq0              = _mm_load_sd(charge+jnrA+0);
1081             vdwjidx0A        = 2*vdwtype[jnrA+0];
1082
1083             fjx0             = _mm_setzero_pd();
1084             fjy0             = _mm_setzero_pd();
1085             fjz0             = _mm_setzero_pd();
1086
1087             /**************************
1088              * CALCULATE INTERACTIONS *
1089              **************************/
1090
1091             if (gmx_mm_any_lt(rsq00,rcutoff2))
1092             {
1093
1094             r00              = _mm_mul_pd(rsq00,rinv00);
1095
1096             /* Compute parameters for interactions between i and j atoms */
1097             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1098
1099             /* LENNARD-JONES DISPERSION/REPULSION */
1100
1101             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1102             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1103             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1104             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
1105             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1106
1107             d                = _mm_sub_pd(r00,rswitch);
1108             d                = _mm_max_pd(d,_mm_setzero_pd());
1109             d2               = _mm_mul_pd(d,d);
1110             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
1111
1112             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
1113
1114             /* Evaluate switch function */
1115             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1116             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1117             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1118
1119             fscal            = fvdw;
1120
1121             fscal            = _mm_and_pd(fscal,cutoff_mask);
1122
1123             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1124
1125             /* Calculate temporary vectorial force */
1126             tx               = _mm_mul_pd(fscal,dx00);
1127             ty               = _mm_mul_pd(fscal,dy00);
1128             tz               = _mm_mul_pd(fscal,dz00);
1129
1130             /* Update vectorial force */
1131             fix0             = _mm_add_pd(fix0,tx);
1132             fiy0             = _mm_add_pd(fiy0,ty);
1133             fiz0             = _mm_add_pd(fiz0,tz);
1134
1135             fjx0             = _mm_add_pd(fjx0,tx);
1136             fjy0             = _mm_add_pd(fjy0,ty);
1137             fjz0             = _mm_add_pd(fjz0,tz);
1138
1139             }
1140
1141             /**************************
1142              * CALCULATE INTERACTIONS *
1143              **************************/
1144
1145             if (gmx_mm_any_lt(rsq10,rcutoff2))
1146             {
1147
1148             /* Compute parameters for interactions between i and j atoms */
1149             qq10             = _mm_mul_pd(iq1,jq0);
1150
1151             /* REACTION-FIELD ELECTROSTATICS */
1152             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1153
1154             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1155
1156             fscal            = felec;
1157
1158             fscal            = _mm_and_pd(fscal,cutoff_mask);
1159
1160             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1161
1162             /* Calculate temporary vectorial force */
1163             tx               = _mm_mul_pd(fscal,dx10);
1164             ty               = _mm_mul_pd(fscal,dy10);
1165             tz               = _mm_mul_pd(fscal,dz10);
1166
1167             /* Update vectorial force */
1168             fix1             = _mm_add_pd(fix1,tx);
1169             fiy1             = _mm_add_pd(fiy1,ty);
1170             fiz1             = _mm_add_pd(fiz1,tz);
1171
1172             fjx0             = _mm_add_pd(fjx0,tx);
1173             fjy0             = _mm_add_pd(fjy0,ty);
1174             fjz0             = _mm_add_pd(fjz0,tz);
1175
1176             }
1177
1178             /**************************
1179              * CALCULATE INTERACTIONS *
1180              **************************/
1181
1182             if (gmx_mm_any_lt(rsq20,rcutoff2))
1183             {
1184
1185             /* Compute parameters for interactions between i and j atoms */
1186             qq20             = _mm_mul_pd(iq2,jq0);
1187
1188             /* REACTION-FIELD ELECTROSTATICS */
1189             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1190
1191             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1192
1193             fscal            = felec;
1194
1195             fscal            = _mm_and_pd(fscal,cutoff_mask);
1196
1197             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1198
1199             /* Calculate temporary vectorial force */
1200             tx               = _mm_mul_pd(fscal,dx20);
1201             ty               = _mm_mul_pd(fscal,dy20);
1202             tz               = _mm_mul_pd(fscal,dz20);
1203
1204             /* Update vectorial force */
1205             fix2             = _mm_add_pd(fix2,tx);
1206             fiy2             = _mm_add_pd(fiy2,ty);
1207             fiz2             = _mm_add_pd(fiz2,tz);
1208
1209             fjx0             = _mm_add_pd(fjx0,tx);
1210             fjy0             = _mm_add_pd(fjy0,ty);
1211             fjz0             = _mm_add_pd(fjz0,tz);
1212
1213             }
1214
1215             /**************************
1216              * CALCULATE INTERACTIONS *
1217              **************************/
1218
1219             if (gmx_mm_any_lt(rsq30,rcutoff2))
1220             {
1221
1222             /* Compute parameters for interactions between i and j atoms */
1223             qq30             = _mm_mul_pd(iq3,jq0);
1224
1225             /* REACTION-FIELD ELECTROSTATICS */
1226             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1227
1228             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1229
1230             fscal            = felec;
1231
1232             fscal            = _mm_and_pd(fscal,cutoff_mask);
1233
1234             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1235
1236             /* Calculate temporary vectorial force */
1237             tx               = _mm_mul_pd(fscal,dx30);
1238             ty               = _mm_mul_pd(fscal,dy30);
1239             tz               = _mm_mul_pd(fscal,dz30);
1240
1241             /* Update vectorial force */
1242             fix3             = _mm_add_pd(fix3,tx);
1243             fiy3             = _mm_add_pd(fiy3,ty);
1244             fiz3             = _mm_add_pd(fiz3,tz);
1245
1246             fjx0             = _mm_add_pd(fjx0,tx);
1247             fjy0             = _mm_add_pd(fjy0,ty);
1248             fjz0             = _mm_add_pd(fjz0,tz);
1249
1250             }
1251
1252             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1253
1254             /* Inner loop uses 149 flops */
1255         }
1256
1257         /* End of innermost loop */
1258
1259         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1260                                               f+i_coord_offset,fshift+i_shift_offset);
1261
1262         /* Increment number of inner iterations */
1263         inneriter                  += j_index_end - j_index_start;
1264
1265         /* Outer loop uses 24 flops */
1266     }
1267
1268     /* Increment number of outer iterations */
1269     outeriter        += nri;
1270
1271     /* Update outer/inner flops */
1272
1273     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*149);
1274 }