src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_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_GeomW3P1_VF_sse4_1_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_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              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  86     real             rswitch_scalar,d_scalar;
  87     __m128d          dummy_mask,cutoff_mask;
  88     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  89     __m128d          one     = _mm_set1_pd(1.0);
  90     __m128d          two     = _mm_set1_pd(2.0);
  91     x                = xx[0];
  92     f                = ff[0];
  93
  94     nri              = nlist->nri;
  95     iinr             = nlist->iinr;
  96     jindex           = nlist->jindex;
  97     jjnr             = nlist->jjnr;
  98     shiftidx         = nlist->shift;
  99     gid              = nlist->gid;
 100     shiftvec         = fr->shift_vec[0];
 101     fshift           = fr->fshift[0];
 102     facel            = _mm_set1_pd(fr->epsfac);
 103     charge           = mdatoms->chargeA;
 104     krf              = _mm_set1_pd(fr->ic->k_rf);
 105     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 106     crf              = _mm_set1_pd(fr->ic->c_rf);
 107     nvdwtype         = fr->ntype;
 108     vdwparam         = fr->nbfp;
 109     vdwtype          = mdatoms->typeA;
 110
 111     /* Setup water-specific parameters */
 112     inr              = nlist->iinr[0];
 113     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 114     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 115     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 116     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 117
 118     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 119     rcutoff_scalar   = fr->rcoulomb;
 120     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 121     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 122
 123     rswitch_scalar   = fr->rvdw_switch;
 124     rswitch          = _mm_set1_pd(rswitch_scalar);
 125     /* Setup switch parameters */
 126     d_scalar         = rcutoff_scalar-rswitch_scalar;
 127     d                = _mm_set1_pd(d_scalar);
 128     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 129     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 130     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 131     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 132     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 133     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 159                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 160
 161         fix0             = _mm_setzero_pd();
 162         fiy0             = _mm_setzero_pd();
 163         fiz0             = _mm_setzero_pd();
 164         fix1             = _mm_setzero_pd();
 165         fiy1             = _mm_setzero_pd();
 166         fiz1             = _mm_setzero_pd();
 167         fix2             = _mm_setzero_pd();
 168         fiy2             = _mm_setzero_pd();
 169         fiz2             = _mm_setzero_pd();
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_pd();
 173         vvdwsum          = _mm_setzero_pd();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_pd(ix0,jx0);
 191             dy00             = _mm_sub_pd(iy0,jy0);
 192             dz00             = _mm_sub_pd(iz0,jz0);
 193             dx10             = _mm_sub_pd(ix1,jx0);
 194             dy10             = _mm_sub_pd(iy1,jy0);
 195             dz10             = _mm_sub_pd(iz1,jz0);
 196             dx20             = _mm_sub_pd(ix2,jx0);
 197             dy20             = _mm_sub_pd(iy2,jy0);
 198             dz20             = _mm_sub_pd(iz2,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 202             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 204
 205             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 206             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 207             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 208
 209             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 210             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 211             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 212
 213             /* Load parameters for j particles */
 214             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 215             vdwjidx0A        = 2*vdwtype[jnrA+0];
 216             vdwjidx0B        = 2*vdwtype[jnrB+0];
 217
 218             fjx0             = _mm_setzero_pd();
 219             fjy0             = _mm_setzero_pd();
 220             fjz0             = _mm_setzero_pd();
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             if (gmx_mm_any_lt(rsq00,rcutoff2))
 227             {
 228
 229             r00              = _mm_mul_pd(rsq00,rinv00);
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm_mul_pd(iq0,jq0);
 233             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 235
 236             /* REACTION-FIELD ELECTROSTATICS */
 237             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 238             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 239
 240             /* LENNARD-JONES DISPERSION/REPULSION */
 241
 242             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 243             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 244             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 245             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 246             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 247
 248             d                = _mm_sub_pd(r00,rswitch);
 249             d                = _mm_max_pd(d,_mm_setzero_pd());
 250             d2               = _mm_mul_pd(d,d);
 251             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)))))));
 252
 253             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 254
 255             /* Evaluate switch function */
 256             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 257             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 258             vvdw             = _mm_mul_pd(vvdw,sw);
 259             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velec            = _mm_and_pd(velec,cutoff_mask);
 263             velecsum         = _mm_add_pd(velecsum,velec);
 264             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 265             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = _mm_add_pd(felec,fvdw);
 268
 269             fscal            = _mm_and_pd(fscal,cutoff_mask);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_pd(fscal,dx00);
 273             ty               = _mm_mul_pd(fscal,dy00);
 274             tz               = _mm_mul_pd(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_pd(fix0,tx);
 278             fiy0             = _mm_add_pd(fiy0,ty);
 279             fiz0             = _mm_add_pd(fiz0,tz);
 280
 281             fjx0             = _mm_add_pd(fjx0,tx);
 282             fjy0             = _mm_add_pd(fjy0,ty);
 283             fjz0             = _mm_add_pd(fjz0,tz);
 284
 285             }
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             if (gmx_mm_any_lt(rsq10,rcutoff2))
 292             {
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq10             = _mm_mul_pd(iq1,jq0);
 296
 297             /* REACTION-FIELD ELECTROSTATICS */
 298             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 299             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 300
 301             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velec            = _mm_and_pd(velec,cutoff_mask);
 305             velecsum         = _mm_add_pd(velecsum,velec);
 306
 307             fscal            = felec;
 308
 309             fscal            = _mm_and_pd(fscal,cutoff_mask);
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm_mul_pd(fscal,dx10);
 313             ty               = _mm_mul_pd(fscal,dy10);
 314             tz               = _mm_mul_pd(fscal,dz10);
 315
 316             /* Update vectorial force */
 317             fix1             = _mm_add_pd(fix1,tx);
 318             fiy1             = _mm_add_pd(fiy1,ty);
 319             fiz1             = _mm_add_pd(fiz1,tz);
 320
 321             fjx0             = _mm_add_pd(fjx0,tx);
 322             fjy0             = _mm_add_pd(fjy0,ty);
 323             fjz0             = _mm_add_pd(fjz0,tz);
 324
 325             }
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             if (gmx_mm_any_lt(rsq20,rcutoff2))
 332             {
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq20             = _mm_mul_pd(iq2,jq0);
 336
 337             /* REACTION-FIELD ELECTROSTATICS */
 338             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 339             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 340
 341             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _mm_and_pd(velec,cutoff_mask);
 345             velecsum         = _mm_add_pd(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             fscal            = _mm_and_pd(fscal,cutoff_mask);
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_pd(fscal,dx20);
 353             ty               = _mm_mul_pd(fscal,dy20);
 354             tz               = _mm_mul_pd(fscal,dz20);
 355
 356             /* Update vectorial force */
 357             fix2             = _mm_add_pd(fix2,tx);
 358             fiy2             = _mm_add_pd(fiy2,ty);
 359             fiz2             = _mm_add_pd(fiz2,tz);
 360
 361             fjx0             = _mm_add_pd(fjx0,tx);
 362             fjy0             = _mm_add_pd(fjy0,ty);
 363             fjz0             = _mm_add_pd(fjz0,tz);
 364
 365             }
 366
 367             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 368
 369             /* Inner loop uses 145 flops */
 370         }
 371
 372         if(jidx<j_index_end)
 373         {
 374
 375             jnrA             = jjnr[jidx];
 376             j_coord_offsetA  = DIM*jnrA;
 377
 378             /* load j atom coordinates */
 379             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 380                                               &jx0,&jy0,&jz0);
 381
 382             /* Calculate displacement vector */
 383             dx00             = _mm_sub_pd(ix0,jx0);
 384             dy00             = _mm_sub_pd(iy0,jy0);
 385             dz00             = _mm_sub_pd(iz0,jz0);
 386             dx10             = _mm_sub_pd(ix1,jx0);
 387             dy10             = _mm_sub_pd(iy1,jy0);
 388             dz10             = _mm_sub_pd(iz1,jz0);
 389             dx20             = _mm_sub_pd(ix2,jx0);
 390             dy20             = _mm_sub_pd(iy2,jy0);
 391             dz20             = _mm_sub_pd(iz2,jz0);
 392
 393             /* Calculate squared distance and things based on it */
 394             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 395             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 396             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 397
 398             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 399             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 400             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 401
 402             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 403             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 404             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 405
 406             /* Load parameters for j particles */
 407             jq0              = _mm_load_sd(charge+jnrA+0);
 408             vdwjidx0A        = 2*vdwtype[jnrA+0];
 409
 410             fjx0             = _mm_setzero_pd();
 411             fjy0             = _mm_setzero_pd();
 412             fjz0             = _mm_setzero_pd();
 413
 414             /**************************
 415              * CALCULATE INTERACTIONS *
 416              **************************/
 417
 418             if (gmx_mm_any_lt(rsq00,rcutoff2))
 419             {
 420
 421             r00              = _mm_mul_pd(rsq00,rinv00);
 422
 423             /* Compute parameters for interactions between i and j atoms */
 424             qq00             = _mm_mul_pd(iq0,jq0);
 425             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 426
 427             /* REACTION-FIELD ELECTROSTATICS */
 428             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 429             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 430
 431             /* LENNARD-JONES DISPERSION/REPULSION */
 432
 433             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 434             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 435             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 436             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 437             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 438
 439             d                = _mm_sub_pd(r00,rswitch);
 440             d                = _mm_max_pd(d,_mm_setzero_pd());
 441             d2               = _mm_mul_pd(d,d);
 442             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)))))));
 443
 444             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 445
 446             /* Evaluate switch function */
 447             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 448             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 449             vvdw             = _mm_mul_pd(vvdw,sw);
 450             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velec            = _mm_and_pd(velec,cutoff_mask);
 454             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 455             velecsum         = _mm_add_pd(velecsum,velec);
 456             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 457             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 458             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 459
 460             fscal            = _mm_add_pd(felec,fvdw);
 461
 462             fscal            = _mm_and_pd(fscal,cutoff_mask);
 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
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             if (gmx_mm_any_lt(rsq10,rcutoff2))
 487             {
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             qq10             = _mm_mul_pd(iq1,jq0);
 491
 492             /* REACTION-FIELD ELECTROSTATICS */
 493             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 494             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 495
 496             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 497
 498             /* Update potential sum for this i atom from the interaction with this j atom. */
 499             velec            = _mm_and_pd(velec,cutoff_mask);
 500             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 501             velecsum         = _mm_add_pd(velecsum,velec);
 502
 503             fscal            = felec;
 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,dx10);
 511             ty               = _mm_mul_pd(fscal,dy10);
 512             tz               = _mm_mul_pd(fscal,dz10);
 513
 514             /* Update vectorial force */
 515             fix1             = _mm_add_pd(fix1,tx);
 516             fiy1             = _mm_add_pd(fiy1,ty);
 517             fiz1             = _mm_add_pd(fiz1,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(rsq20,rcutoff2))
 530             {
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq20             = _mm_mul_pd(iq2,jq0);
 534
 535             /* REACTION-FIELD ELECTROSTATICS */
 536             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 537             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 538
 539             cutoff_mask      = _mm_cmplt_pd(rsq20,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,dx20);
 554             ty               = _mm_mul_pd(fscal,dy20);
 555             tz               = _mm_mul_pd(fscal,dz20);
 556
 557             /* Update vectorial force */
 558             fix2             = _mm_add_pd(fix2,tx);
 559             fiy2             = _mm_add_pd(fiy2,ty);
 560             fiz2             = _mm_add_pd(fiz2,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             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 569
 570             /* Inner loop uses 145 flops */
 571         }
 572
 573         /* End of innermost loop */
 574
 575         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 576                                               f+i_coord_offset,fshift+i_shift_offset);
 577
 578         ggid                        = gid[iidx];
 579         /* Update potential energies */
 580         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 581         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 582
 583         /* Increment number of inner iterations */
 584         inneriter                  += j_index_end - j_index_start;
 585
 586         /* Outer loop uses 20 flops */
 587     }
 588
 589     /* Increment number of outer iterations */
 590     outeriter        += nri;
 591
 592     /* Update outer/inner flops */
 593
 594     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 595 }
 596 /*
 597  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse4_1_double
 598  * Electrostatics interaction: ReactionField
 599  * VdW interaction:            LennardJones
 600  * Geometry:                   Water3-Particle
 601  * Calculate force/pot:        Force
 602  */
 603 void
 604 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse4_1_double
 605                     (t_nblist * gmx_restrict                nlist,
 606                      rvec * gmx_restrict                    xx,
 607                      rvec * gmx_restrict                    ff,
 608                      t_forcerec * gmx_restrict              fr,
 609                      t_mdatoms * gmx_restrict               mdatoms,
 610                      nb_kernel_data_t * gmx_restrict        kernel_data,
 611                      t_nrnb * gmx_restrict                  nrnb)
 612 {
 613     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 614      * just 0 for non-waters.
 615      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 616      * jnr indices corresponding to data put in the four positions in the SIMD register.
 617      */
 618     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 619     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 620     int              jnrA,jnrB;
 621     int              j_coord_offsetA,j_coord_offsetB;
 622     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 623     real             rcutoff_scalar;
 624     real             *shiftvec,*fshift,*x,*f;
 625     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 626     int              vdwioffset0;
 627     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 628     int              vdwioffset1;
 629     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 630     int              vdwioffset2;
 631     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 632     int              vdwjidx0A,vdwjidx0B;
 633     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 634     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 635     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 636     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 637     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 638     real             *charge;
 639     int              nvdwtype;
 640     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 641     int              *vdwtype;
 642     real             *vdwparam;
 643     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 644     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 645     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 646     real             rswitch_scalar,d_scalar;
 647     __m128d          dummy_mask,cutoff_mask;
 648     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 649     __m128d          one     = _mm_set1_pd(1.0);
 650     __m128d          two     = _mm_set1_pd(2.0);
 651     x                = xx[0];
 652     f                = ff[0];
 653
 654     nri              = nlist->nri;
 655     iinr             = nlist->iinr;
 656     jindex           = nlist->jindex;
 657     jjnr             = nlist->jjnr;
 658     shiftidx         = nlist->shift;
 659     gid              = nlist->gid;
 660     shiftvec         = fr->shift_vec[0];
 661     fshift           = fr->fshift[0];
 662     facel            = _mm_set1_pd(fr->epsfac);
 663     charge           = mdatoms->chargeA;
 664     krf              = _mm_set1_pd(fr->ic->k_rf);
 665     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 666     crf              = _mm_set1_pd(fr->ic->c_rf);
 667     nvdwtype         = fr->ntype;
 668     vdwparam         = fr->nbfp;
 669     vdwtype          = mdatoms->typeA;
 670
 671     /* Setup water-specific parameters */
 672     inr              = nlist->iinr[0];
 673     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 674     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 675     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 676     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 677
 678     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 679     rcutoff_scalar   = fr->rcoulomb;
 680     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 681     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 682
 683     rswitch_scalar   = fr->rvdw_switch;
 684     rswitch          = _mm_set1_pd(rswitch_scalar);
 685     /* Setup switch parameters */
 686     d_scalar         = rcutoff_scalar-rswitch_scalar;
 687     d                = _mm_set1_pd(d_scalar);
 688     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 689     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 690     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 691     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 692     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 693     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 694
 695     /* Avoid stupid compiler warnings */
 696     jnrA = jnrB = 0;
 697     j_coord_offsetA = 0;
 698     j_coord_offsetB = 0;
 699
 700     outeriter        = 0;
 701     inneriter        = 0;
 702
 703     /* Start outer loop over neighborlists */
 704     for(iidx=0; iidx<nri; iidx++)
 705     {
 706         /* Load shift vector for this list */
 707         i_shift_offset   = DIM*shiftidx[iidx];
 708
 709         /* Load limits for loop over neighbors */
 710         j_index_start    = jindex[iidx];
 711         j_index_end      = jindex[iidx+1];
 712
 713         /* Get outer coordinate index */
 714         inr              = iinr[iidx];
 715         i_coord_offset   = DIM*inr;
 716
 717         /* Load i particle coords and add shift vector */
 718         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 719                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 720
 721         fix0             = _mm_setzero_pd();
 722         fiy0             = _mm_setzero_pd();
 723         fiz0             = _mm_setzero_pd();
 724         fix1             = _mm_setzero_pd();
 725         fiy1             = _mm_setzero_pd();
 726         fiz1             = _mm_setzero_pd();
 727         fix2             = _mm_setzero_pd();
 728         fiy2             = _mm_setzero_pd();
 729         fiz2             = _mm_setzero_pd();
 730
 731         /* Start inner kernel loop */
 732         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 733         {
 734
 735             /* Get j neighbor index, and coordinate index */
 736             jnrA             = jjnr[jidx];
 737             jnrB             = jjnr[jidx+1];
 738             j_coord_offsetA  = DIM*jnrA;
 739             j_coord_offsetB  = DIM*jnrB;
 740
 741             /* load j atom coordinates */
 742             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 743                                               &jx0,&jy0,&jz0);
 744
 745             /* Calculate displacement vector */
 746             dx00             = _mm_sub_pd(ix0,jx0);
 747             dy00             = _mm_sub_pd(iy0,jy0);
 748             dz00             = _mm_sub_pd(iz0,jz0);
 749             dx10             = _mm_sub_pd(ix1,jx0);
 750             dy10             = _mm_sub_pd(iy1,jy0);
 751             dz10             = _mm_sub_pd(iz1,jz0);
 752             dx20             = _mm_sub_pd(ix2,jx0);
 753             dy20             = _mm_sub_pd(iy2,jy0);
 754             dz20             = _mm_sub_pd(iz2,jz0);
 755
 756             /* Calculate squared distance and things based on it */
 757             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 758             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 759             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 760
 761             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 762             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 763             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 764
 765             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 766             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 767             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 768
 769             /* Load parameters for j particles */
 770             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 771             vdwjidx0A        = 2*vdwtype[jnrA+0];
 772             vdwjidx0B        = 2*vdwtype[jnrB+0];
 773
 774             fjx0             = _mm_setzero_pd();
 775             fjy0             = _mm_setzero_pd();
 776             fjz0             = _mm_setzero_pd();
 777
 778             /**************************
 779              * CALCULATE INTERACTIONS *
 780              **************************/
 781
 782             if (gmx_mm_any_lt(rsq00,rcutoff2))
 783             {
 784
 785             r00              = _mm_mul_pd(rsq00,rinv00);
 786
 787             /* Compute parameters for interactions between i and j atoms */
 788             qq00             = _mm_mul_pd(iq0,jq0);
 789             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 790                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 791
 792             /* REACTION-FIELD ELECTROSTATICS */
 793             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 794
 795             /* LENNARD-JONES DISPERSION/REPULSION */
 796
 797             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 798             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 799             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 800             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 801             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 802
 803             d                = _mm_sub_pd(r00,rswitch);
 804             d                = _mm_max_pd(d,_mm_setzero_pd());
 805             d2               = _mm_mul_pd(d,d);
 806             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)))))));
 807
 808             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 809
 810             /* Evaluate switch function */
 811             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 812             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 813             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 814
 815             fscal            = _mm_add_pd(felec,fvdw);
 816
 817             fscal            = _mm_and_pd(fscal,cutoff_mask);
 818
 819             /* Calculate temporary vectorial force */
 820             tx               = _mm_mul_pd(fscal,dx00);
 821             ty               = _mm_mul_pd(fscal,dy00);
 822             tz               = _mm_mul_pd(fscal,dz00);
 823
 824             /* Update vectorial force */
 825             fix0             = _mm_add_pd(fix0,tx);
 826             fiy0             = _mm_add_pd(fiy0,ty);
 827             fiz0             = _mm_add_pd(fiz0,tz);
 828
 829             fjx0             = _mm_add_pd(fjx0,tx);
 830             fjy0             = _mm_add_pd(fjy0,ty);
 831             fjz0             = _mm_add_pd(fjz0,tz);
 832
 833             }
 834
 835             /**************************
 836              * CALCULATE INTERACTIONS *
 837              **************************/
 838
 839             if (gmx_mm_any_lt(rsq10,rcutoff2))
 840             {
 841
 842             /* Compute parameters for interactions between i and j atoms */
 843             qq10             = _mm_mul_pd(iq1,jq0);
 844
 845             /* REACTION-FIELD ELECTROSTATICS */
 846             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 847
 848             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 849
 850             fscal            = felec;
 851
 852             fscal            = _mm_and_pd(fscal,cutoff_mask);
 853
 854             /* Calculate temporary vectorial force */
 855             tx               = _mm_mul_pd(fscal,dx10);
 856             ty               = _mm_mul_pd(fscal,dy10);
 857             tz               = _mm_mul_pd(fscal,dz10);
 858
 859             /* Update vectorial force */
 860             fix1             = _mm_add_pd(fix1,tx);
 861             fiy1             = _mm_add_pd(fiy1,ty);
 862             fiz1             = _mm_add_pd(fiz1,tz);
 863
 864             fjx0             = _mm_add_pd(fjx0,tx);
 865             fjy0             = _mm_add_pd(fjy0,ty);
 866             fjz0             = _mm_add_pd(fjz0,tz);
 867
 868             }
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             if (gmx_mm_any_lt(rsq20,rcutoff2))
 875             {
 876
 877             /* Compute parameters for interactions between i and j atoms */
 878             qq20             = _mm_mul_pd(iq2,jq0);
 879
 880             /* REACTION-FIELD ELECTROSTATICS */
 881             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 882
 883             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 884
 885             fscal            = felec;
 886
 887             fscal            = _mm_and_pd(fscal,cutoff_mask);
 888
 889             /* Calculate temporary vectorial force */
 890             tx               = _mm_mul_pd(fscal,dx20);
 891             ty               = _mm_mul_pd(fscal,dy20);
 892             tz               = _mm_mul_pd(fscal,dz20);
 893
 894             /* Update vectorial force */
 895             fix2             = _mm_add_pd(fix2,tx);
 896             fiy2             = _mm_add_pd(fiy2,ty);
 897             fiz2             = _mm_add_pd(fiz2,tz);
 898
 899             fjx0             = _mm_add_pd(fjx0,tx);
 900             fjy0             = _mm_add_pd(fjy0,ty);
 901             fjz0             = _mm_add_pd(fjz0,tz);
 902
 903             }
 904
 905             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 906
 907             /* Inner loop uses 124 flops */
 908         }
 909
 910         if(jidx<j_index_end)
 911         {
 912
 913             jnrA             = jjnr[jidx];
 914             j_coord_offsetA  = DIM*jnrA;
 915
 916             /* load j atom coordinates */
 917             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 918                                               &jx0,&jy0,&jz0);
 919
 920             /* Calculate displacement vector */
 921             dx00             = _mm_sub_pd(ix0,jx0);
 922             dy00             = _mm_sub_pd(iy0,jy0);
 923             dz00             = _mm_sub_pd(iz0,jz0);
 924             dx10             = _mm_sub_pd(ix1,jx0);
 925             dy10             = _mm_sub_pd(iy1,jy0);
 926             dz10             = _mm_sub_pd(iz1,jz0);
 927             dx20             = _mm_sub_pd(ix2,jx0);
 928             dy20             = _mm_sub_pd(iy2,jy0);
 929             dz20             = _mm_sub_pd(iz2,jz0);
 930
 931             /* Calculate squared distance and things based on it */
 932             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 933             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 934             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 935
 936             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 937             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 938             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 939
 940             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 941             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 942             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 943
 944             /* Load parameters for j particles */
 945             jq0              = _mm_load_sd(charge+jnrA+0);
 946             vdwjidx0A        = 2*vdwtype[jnrA+0];
 947
 948             fjx0             = _mm_setzero_pd();
 949             fjy0             = _mm_setzero_pd();
 950             fjz0             = _mm_setzero_pd();
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             if (gmx_mm_any_lt(rsq00,rcutoff2))
 957             {
 958
 959             r00              = _mm_mul_pd(rsq00,rinv00);
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             qq00             = _mm_mul_pd(iq0,jq0);
 963             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 964
 965             /* REACTION-FIELD ELECTROSTATICS */
 966             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 967
 968             /* LENNARD-JONES DISPERSION/REPULSION */
 969
 970             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 971             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 972             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 973             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 974             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 975
 976             d                = _mm_sub_pd(r00,rswitch);
 977             d                = _mm_max_pd(d,_mm_setzero_pd());
 978             d2               = _mm_mul_pd(d,d);
 979             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)))))));
 980
 981             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 982
 983             /* Evaluate switch function */
 984             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 985             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 986             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 987
 988             fscal            = _mm_add_pd(felec,fvdw);
 989
 990             fscal            = _mm_and_pd(fscal,cutoff_mask);
 991
 992             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 993
 994             /* Calculate temporary vectorial force */
 995             tx               = _mm_mul_pd(fscal,dx00);
 996             ty               = _mm_mul_pd(fscal,dy00);
 997             tz               = _mm_mul_pd(fscal,dz00);
 998
 999             /* Update vectorial force */
1000             fix0             = _mm_add_pd(fix0,tx);
1001             fiy0             = _mm_add_pd(fiy0,ty);
1002             fiz0             = _mm_add_pd(fiz0,tz);
1003
1004             fjx0             = _mm_add_pd(fjx0,tx);
1005             fjy0             = _mm_add_pd(fjy0,ty);
1006             fjz0             = _mm_add_pd(fjz0,tz);
1007
1008             }
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             if (gmx_mm_any_lt(rsq10,rcutoff2))
1015             {
1016
1017             /* Compute parameters for interactions between i and j atoms */
1018             qq10             = _mm_mul_pd(iq1,jq0);
1019
1020             /* REACTION-FIELD ELECTROSTATICS */
1021             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1022
1023             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1024
1025             fscal            = felec;
1026
1027             fscal            = _mm_and_pd(fscal,cutoff_mask);
1028
1029             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1030
1031             /* Calculate temporary vectorial force */
1032             tx               = _mm_mul_pd(fscal,dx10);
1033             ty               = _mm_mul_pd(fscal,dy10);
1034             tz               = _mm_mul_pd(fscal,dz10);
1035
1036             /* Update vectorial force */
1037             fix1             = _mm_add_pd(fix1,tx);
1038             fiy1             = _mm_add_pd(fiy1,ty);
1039             fiz1             = _mm_add_pd(fiz1,tz);
1040
1041             fjx0             = _mm_add_pd(fjx0,tx);
1042             fjy0             = _mm_add_pd(fjy0,ty);
1043             fjz0             = _mm_add_pd(fjz0,tz);
1044
1045             }
1046
1047             /**************************
1048              * CALCULATE INTERACTIONS *
1049              **************************/
1050
1051             if (gmx_mm_any_lt(rsq20,rcutoff2))
1052             {
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq20             = _mm_mul_pd(iq2,jq0);
1056
1057             /* REACTION-FIELD ELECTROSTATICS */
1058             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1059
1060             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm_and_pd(fscal,cutoff_mask);
1065
1066             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1067
1068             /* Calculate temporary vectorial force */
1069             tx               = _mm_mul_pd(fscal,dx20);
1070             ty               = _mm_mul_pd(fscal,dy20);
1071             tz               = _mm_mul_pd(fscal,dz20);
1072
1073             /* Update vectorial force */
1074             fix2             = _mm_add_pd(fix2,tx);
1075             fiy2             = _mm_add_pd(fiy2,ty);
1076             fiz2             = _mm_add_pd(fiz2,tz);
1077
1078             fjx0             = _mm_add_pd(fjx0,tx);
1079             fjy0             = _mm_add_pd(fjy0,ty);
1080             fjz0             = _mm_add_pd(fjz0,tz);
1081
1082             }
1083
1084             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1085
1086             /* Inner loop uses 124 flops */
1087         }
1088
1089         /* End of innermost loop */
1090
1091         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1092                                               f+i_coord_offset,fshift+i_shift_offset);
1093
1094         /* Increment number of inner iterations */
1095         inneriter                  += j_index_end - j_index_start;
1096
1097         /* Outer loop uses 18 flops */
1098     }
1099
1100     /* Increment number of outer iterations */
1101     outeriter        += nri;
1102
1103     /* Update outer/inner flops */
1104
1105     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*124);
1106 }