src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwCSTab_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_VdwCSTab_GeomW4P1_VF_sse4_1_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_pd(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 122     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 123     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 155                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 156
 157         fix0             = _mm_setzero_pd();
 158         fiy0             = _mm_setzero_pd();
 159         fiz0             = _mm_setzero_pd();
 160         fix1             = _mm_setzero_pd();
 161         fiy1             = _mm_setzero_pd();
 162         fiz1             = _mm_setzero_pd();
 163         fix2             = _mm_setzero_pd();
 164         fiy2             = _mm_setzero_pd();
 165         fiz2             = _mm_setzero_pd();
 166         fix3             = _mm_setzero_pd();
 167         fiy3             = _mm_setzero_pd();
 168         fiz3             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172         vvdwsum          = _mm_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_pd(ix0,jx0);
 190             dy00             = _mm_sub_pd(iy0,jy0);
 191             dz00             = _mm_sub_pd(iz0,jz0);
 192             dx10             = _mm_sub_pd(ix1,jx0);
 193             dy10             = _mm_sub_pd(iy1,jy0);
 194             dz10             = _mm_sub_pd(iz1,jz0);
 195             dx20             = _mm_sub_pd(ix2,jx0);
 196             dy20             = _mm_sub_pd(iy2,jy0);
 197             dz20             = _mm_sub_pd(iz2,jz0);
 198             dx30             = _mm_sub_pd(ix3,jx0);
 199             dy30             = _mm_sub_pd(iy3,jy0);
 200             dz30             = _mm_sub_pd(iz3,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 206             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 207
 208             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 209             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 210             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 211             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 212
 213             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 215             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 219             vdwjidx0A        = 2*vdwtype[jnrA+0];
 220             vdwjidx0B        = 2*vdwtype[jnrB+0];
 221
 222             fjx0             = _mm_setzero_pd();
 223             fjy0             = _mm_setzero_pd();
 224             fjz0             = _mm_setzero_pd();
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             r00              = _mm_mul_pd(rsq00,rinv00);
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 235
 236             /* Calculate table index by multiplying r with table scale and truncate to integer */
 237             rt               = _mm_mul_pd(r00,vftabscale);
 238             vfitab           = _mm_cvttpd_epi32(rt);
 239             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 240             vfitab           = _mm_slli_epi32(vfitab,3);
 241
 242             /* CUBIC SPLINE TABLE DISPERSION */
 243             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 244             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 245             GMX_MM_TRANSPOSE2_PD(Y,F);
 246             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 247             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 248             GMX_MM_TRANSPOSE2_PD(G,H);
 249             Heps             = _mm_mul_pd(vfeps,H);
 250             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 251             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 252             vvdw6            = _mm_mul_pd(c6_00,VV);
 253             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 254             fvdw6            = _mm_mul_pd(c6_00,FF);
 255
 256             /* CUBIC SPLINE TABLE REPULSION */
 257             vfitab           = _mm_add_epi32(vfitab,ifour);
 258             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 259             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 260             GMX_MM_TRANSPOSE2_PD(Y,F);
 261             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 262             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 263             GMX_MM_TRANSPOSE2_PD(G,H);
 264             Heps             = _mm_mul_pd(vfeps,H);
 265             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 266             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 267             vvdw12           = _mm_mul_pd(c12_00,VV);
 268             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 269             fvdw12           = _mm_mul_pd(c12_00,FF);
 270             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 271             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 275
 276             fscal            = fvdw;
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm_mul_pd(fscal,dx00);
 280             ty               = _mm_mul_pd(fscal,dy00);
 281             tz               = _mm_mul_pd(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm_add_pd(fix0,tx);
 285             fiy0             = _mm_add_pd(fiy0,ty);
 286             fiz0             = _mm_add_pd(fiz0,tz);
 287
 288             fjx0             = _mm_add_pd(fjx0,tx);
 289             fjy0             = _mm_add_pd(fjy0,ty);
 290             fjz0             = _mm_add_pd(fjz0,tz);
 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 167 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             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 453             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 454             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 455
 456             /* Load parameters for j particles */
 457             jq0              = _mm_load_sd(charge+jnrA+0);
 458             vdwjidx0A        = 2*vdwtype[jnrA+0];
 459
 460             fjx0             = _mm_setzero_pd();
 461             fjy0             = _mm_setzero_pd();
 462             fjz0             = _mm_setzero_pd();
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r00              = _mm_mul_pd(rsq00,rinv00);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 472
 473             /* Calculate table index by multiplying r with table scale and truncate to integer */
 474             rt               = _mm_mul_pd(r00,vftabscale);
 475             vfitab           = _mm_cvttpd_epi32(rt);
 476             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 477             vfitab           = _mm_slli_epi32(vfitab,3);
 478
 479             /* CUBIC SPLINE TABLE DISPERSION */
 480             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 481             F                = _mm_setzero_pd();
 482             GMX_MM_TRANSPOSE2_PD(Y,F);
 483             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 484             H                = _mm_setzero_pd();
 485             GMX_MM_TRANSPOSE2_PD(G,H);
 486             Heps             = _mm_mul_pd(vfeps,H);
 487             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 488             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 489             vvdw6            = _mm_mul_pd(c6_00,VV);
 490             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 491             fvdw6            = _mm_mul_pd(c6_00,FF);
 492
 493             /* CUBIC SPLINE TABLE REPULSION */
 494             vfitab           = _mm_add_epi32(vfitab,ifour);
 495             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 496             F                = _mm_setzero_pd();
 497             GMX_MM_TRANSPOSE2_PD(Y,F);
 498             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 499             H                = _mm_setzero_pd();
 500             GMX_MM_TRANSPOSE2_PD(G,H);
 501             Heps             = _mm_mul_pd(vfeps,H);
 502             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 503             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 504             vvdw12           = _mm_mul_pd(c12_00,VV);
 505             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 506             fvdw12           = _mm_mul_pd(c12_00,FF);
 507             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 508             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 512             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 513
 514             fscal            = fvdw;
 515
 516             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm_mul_pd(fscal,dx00);
 520             ty               = _mm_mul_pd(fscal,dy00);
 521             tz               = _mm_mul_pd(fscal,dz00);
 522
 523             /* Update vectorial force */
 524             fix0             = _mm_add_pd(fix0,tx);
 525             fiy0             = _mm_add_pd(fiy0,ty);
 526             fiz0             = _mm_add_pd(fiz0,tz);
 527
 528             fjx0             = _mm_add_pd(fjx0,tx);
 529             fjy0             = _mm_add_pd(fjy0,ty);
 530             fjz0             = _mm_add_pd(fjz0,tz);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm_any_lt(rsq10,rcutoff2))
 537             {
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq10             = _mm_mul_pd(iq1,jq0);
 541
 542             /* REACTION-FIELD ELECTROSTATICS */
 543             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 544             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 545
 546             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm_and_pd(velec,cutoff_mask);
 550             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 551             velecsum         = _mm_add_pd(velecsum,velec);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm_and_pd(fscal,cutoff_mask);
 556
 557             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_pd(fscal,dx10);
 561             ty               = _mm_mul_pd(fscal,dy10);
 562             tz               = _mm_mul_pd(fscal,dz10);
 563
 564             /* Update vectorial force */
 565             fix1             = _mm_add_pd(fix1,tx);
 566             fiy1             = _mm_add_pd(fiy1,ty);
 567             fiz1             = _mm_add_pd(fiz1,tz);
 568
 569             fjx0             = _mm_add_pd(fjx0,tx);
 570             fjy0             = _mm_add_pd(fjy0,ty);
 571             fjz0             = _mm_add_pd(fjz0,tz);
 572
 573             }
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             if (gmx_mm_any_lt(rsq20,rcutoff2))
 580             {
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq20             = _mm_mul_pd(iq2,jq0);
 584
 585             /* REACTION-FIELD ELECTROSTATICS */
 586             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 587             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 588
 589             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 590
 591             /* Update potential sum for this i atom from the interaction with this j atom. */
 592             velec            = _mm_and_pd(velec,cutoff_mask);
 593             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 594             velecsum         = _mm_add_pd(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             fscal            = _mm_and_pd(fscal,cutoff_mask);
 599
 600             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 601
 602             /* Calculate temporary vectorial force */
 603             tx               = _mm_mul_pd(fscal,dx20);
 604             ty               = _mm_mul_pd(fscal,dy20);
 605             tz               = _mm_mul_pd(fscal,dz20);
 606
 607             /* Update vectorial force */
 608             fix2             = _mm_add_pd(fix2,tx);
 609             fiy2             = _mm_add_pd(fiy2,ty);
 610             fiz2             = _mm_add_pd(fiz2,tz);
 611
 612             fjx0             = _mm_add_pd(fjx0,tx);
 613             fjy0             = _mm_add_pd(fjy0,ty);
 614             fjz0             = _mm_add_pd(fjz0,tz);
 615
 616             }
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             if (gmx_mm_any_lt(rsq30,rcutoff2))
 623             {
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq30             = _mm_mul_pd(iq3,jq0);
 627
 628             /* REACTION-FIELD ELECTROSTATICS */
 629             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_add_pd(rinv30,_mm_mul_pd(krf,rsq30)),crf));
 630             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
 631
 632             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velec            = _mm_and_pd(velec,cutoff_mask);
 636             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 637             velecsum         = _mm_add_pd(velecsum,velec);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_and_pd(fscal,cutoff_mask);
 642
 643             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm_mul_pd(fscal,dx30);
 647             ty               = _mm_mul_pd(fscal,dy30);
 648             tz               = _mm_mul_pd(fscal,dz30);
 649
 650             /* Update vectorial force */
 651             fix3             = _mm_add_pd(fix3,tx);
 652             fiy3             = _mm_add_pd(fiy3,ty);
 653             fiz3             = _mm_add_pd(fiz3,tz);
 654
 655             fjx0             = _mm_add_pd(fjx0,tx);
 656             fjy0             = _mm_add_pd(fjy0,ty);
 657             fjz0             = _mm_add_pd(fjz0,tz);
 658
 659             }
 660
 661             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 662
 663             /* Inner loop uses 167 flops */
 664         }
 665
 666         /* End of innermost loop */
 667
 668         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 669                                               f+i_coord_offset,fshift+i_shift_offset);
 670
 671         ggid                        = gid[iidx];
 672         /* Update potential energies */
 673         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 674         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 675
 676         /* Increment number of inner iterations */
 677         inneriter                  += j_index_end - j_index_start;
 678
 679         /* Outer loop uses 26 flops */
 680     }
 681
 682     /* Increment number of outer iterations */
 683     outeriter        += nri;
 684
 685     /* Update outer/inner flops */
 686
 687     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 688 }
 689 /*
 690  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_double
 691  * Electrostatics interaction: ReactionField
 692  * VdW interaction:            CubicSplineTable
 693  * Geometry:                   Water4-Particle
 694  * Calculate force/pot:        Force
 695  */
 696 void
 697 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_double
 698                     (t_nblist * gmx_restrict                nlist,
 699                      rvec * gmx_restrict                    xx,
 700                      rvec * gmx_restrict                    ff,
 701                      t_forcerec * gmx_restrict              fr,
 702                      t_mdatoms * gmx_restrict               mdatoms,
 703                      nb_kernel_data_t * gmx_restrict        kernel_data,
 704                      t_nrnb * gmx_restrict                  nrnb)
 705 {
 706     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 707      * just 0 for non-waters.
 708      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 709      * jnr indices corresponding to data put in the four positions in the SIMD register.
 710      */
 711     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 712     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 713     int              jnrA,jnrB;
 714     int              j_coord_offsetA,j_coord_offsetB;
 715     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 716     real             rcutoff_scalar;
 717     real             *shiftvec,*fshift,*x,*f;
 718     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 719     int              vdwioffset0;
 720     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 721     int              vdwioffset1;
 722     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 723     int              vdwioffset2;
 724     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 725     int              vdwioffset3;
 726     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 727     int              vdwjidx0A,vdwjidx0B;
 728     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 729     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 730     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 731     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 732     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 733     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 734     real             *charge;
 735     int              nvdwtype;
 736     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 737     int              *vdwtype;
 738     real             *vdwparam;
 739     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 740     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 741     __m128i          vfitab;
 742     __m128i          ifour       = _mm_set1_epi32(4);
 743     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 744     real             *vftab;
 745     __m128d          dummy_mask,cutoff_mask;
 746     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 747     __m128d          one     = _mm_set1_pd(1.0);
 748     __m128d          two     = _mm_set1_pd(2.0);
 749     x                = xx[0];
 750     f                = ff[0];
 751
 752     nri              = nlist->nri;
 753     iinr             = nlist->iinr;
 754     jindex           = nlist->jindex;
 755     jjnr             = nlist->jjnr;
 756     shiftidx         = nlist->shift;
 757     gid              = nlist->gid;
 758     shiftvec         = fr->shift_vec[0];
 759     fshift           = fr->fshift[0];
 760     facel            = _mm_set1_pd(fr->epsfac);
 761     charge           = mdatoms->chargeA;
 762     krf              = _mm_set1_pd(fr->ic->k_rf);
 763     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 764     crf              = _mm_set1_pd(fr->ic->c_rf);
 765     nvdwtype         = fr->ntype;
 766     vdwparam         = fr->nbfp;
 767     vdwtype          = mdatoms->typeA;
 768
 769     vftab            = kernel_data->table_vdw->data;
 770     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 771
 772     /* Setup water-specific parameters */
 773     inr              = nlist->iinr[0];
 774     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 775     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 776     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 777     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 778
 779     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 780     rcutoff_scalar   = fr->rcoulomb;
 781     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 782     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 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             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 863             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 864             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 865
 866             /* Load parameters for j particles */
 867             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 868             vdwjidx0A        = 2*vdwtype[jnrA+0];
 869             vdwjidx0B        = 2*vdwtype[jnrB+0];
 870
 871             fjx0             = _mm_setzero_pd();
 872             fjy0             = _mm_setzero_pd();
 873             fjz0             = _mm_setzero_pd();
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             r00              = _mm_mul_pd(rsq00,rinv00);
 880
 881             /* Compute parameters for interactions between i and j atoms */
 882             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 883                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 884
 885             /* Calculate table index by multiplying r with table scale and truncate to integer */
 886             rt               = _mm_mul_pd(r00,vftabscale);
 887             vfitab           = _mm_cvttpd_epi32(rt);
 888             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 889             vfitab           = _mm_slli_epi32(vfitab,3);
 890
 891             /* CUBIC SPLINE TABLE DISPERSION */
 892             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 893             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 894             GMX_MM_TRANSPOSE2_PD(Y,F);
 895             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 896             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 897             GMX_MM_TRANSPOSE2_PD(G,H);
 898             Heps             = _mm_mul_pd(vfeps,H);
 899             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 900             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 901             fvdw6            = _mm_mul_pd(c6_00,FF);
 902
 903             /* CUBIC SPLINE TABLE REPULSION */
 904             vfitab           = _mm_add_epi32(vfitab,ifour);
 905             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 906             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 907             GMX_MM_TRANSPOSE2_PD(Y,F);
 908             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 909             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 910             GMX_MM_TRANSPOSE2_PD(G,H);
 911             Heps             = _mm_mul_pd(vfeps,H);
 912             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 913             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 914             fvdw12           = _mm_mul_pd(c12_00,FF);
 915             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 916
 917             fscal            = fvdw;
 918
 919             /* Calculate temporary vectorial force */
 920             tx               = _mm_mul_pd(fscal,dx00);
 921             ty               = _mm_mul_pd(fscal,dy00);
 922             tz               = _mm_mul_pd(fscal,dz00);
 923
 924             /* Update vectorial force */
 925             fix0             = _mm_add_pd(fix0,tx);
 926             fiy0             = _mm_add_pd(fiy0,ty);
 927             fiz0             = _mm_add_pd(fiz0,tz);
 928
 929             fjx0             = _mm_add_pd(fjx0,tx);
 930             fjy0             = _mm_add_pd(fjy0,ty);
 931             fjz0             = _mm_add_pd(fjz0,tz);
 932
 933             /**************************
 934              * CALCULATE INTERACTIONS *
 935              **************************/
 936
 937             if (gmx_mm_any_lt(rsq10,rcutoff2))
 938             {
 939
 940             /* Compute parameters for interactions between i and j atoms */
 941             qq10             = _mm_mul_pd(iq1,jq0);
 942
 943             /* REACTION-FIELD ELECTROSTATICS */
 944             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 945
 946             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 947
 948             fscal            = felec;
 949
 950             fscal            = _mm_and_pd(fscal,cutoff_mask);
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm_mul_pd(fscal,dx10);
 954             ty               = _mm_mul_pd(fscal,dy10);
 955             tz               = _mm_mul_pd(fscal,dz10);
 956
 957             /* Update vectorial force */
 958             fix1             = _mm_add_pd(fix1,tx);
 959             fiy1             = _mm_add_pd(fiy1,ty);
 960             fiz1             = _mm_add_pd(fiz1,tz);
 961
 962             fjx0             = _mm_add_pd(fjx0,tx);
 963             fjy0             = _mm_add_pd(fjy0,ty);
 964             fjz0             = _mm_add_pd(fjz0,tz);
 965
 966             }
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             if (gmx_mm_any_lt(rsq20,rcutoff2))
 973             {
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq20             = _mm_mul_pd(iq2,jq0);
 977
 978             /* REACTION-FIELD ELECTROSTATICS */
 979             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 980
 981             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 982
 983             fscal            = felec;
 984
 985             fscal            = _mm_and_pd(fscal,cutoff_mask);
 986
 987             /* Calculate temporary vectorial force */
 988             tx               = _mm_mul_pd(fscal,dx20);
 989             ty               = _mm_mul_pd(fscal,dy20);
 990             tz               = _mm_mul_pd(fscal,dz20);
 991
 992             /* Update vectorial force */
 993             fix2             = _mm_add_pd(fix2,tx);
 994             fiy2             = _mm_add_pd(fiy2,ty);
 995             fiz2             = _mm_add_pd(fiz2,tz);
 996
 997             fjx0             = _mm_add_pd(fjx0,tx);
 998             fjy0             = _mm_add_pd(fjy0,ty);
 999             fjz0             = _mm_add_pd(fjz0,tz);
1000
1001             }
1002
1003             /**************************
1004              * CALCULATE INTERACTIONS *
1005              **************************/
1006
1007             if (gmx_mm_any_lt(rsq30,rcutoff2))
1008             {
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq30             = _mm_mul_pd(iq3,jq0);
1012
1013             /* REACTION-FIELD ELECTROSTATICS */
1014             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1015
1016             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1017
1018             fscal            = felec;
1019
1020             fscal            = _mm_and_pd(fscal,cutoff_mask);
1021
1022             /* Calculate temporary vectorial force */
1023             tx               = _mm_mul_pd(fscal,dx30);
1024             ty               = _mm_mul_pd(fscal,dy30);
1025             tz               = _mm_mul_pd(fscal,dz30);
1026
1027             /* Update vectorial force */
1028             fix3             = _mm_add_pd(fix3,tx);
1029             fiy3             = _mm_add_pd(fiy3,ty);
1030             fiz3             = _mm_add_pd(fiz3,tz);
1031
1032             fjx0             = _mm_add_pd(fjx0,tx);
1033             fjy0             = _mm_add_pd(fjy0,ty);
1034             fjz0             = _mm_add_pd(fjz0,tz);
1035
1036             }
1037
1038             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1039
1040             /* Inner loop uses 141 flops */
1041         }
1042
1043         if(jidx<j_index_end)
1044         {
1045
1046             jnrA             = jjnr[jidx];
1047             j_coord_offsetA  = DIM*jnrA;
1048
1049             /* load j atom coordinates */
1050             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1051                                               &jx0,&jy0,&jz0);
1052
1053             /* Calculate displacement vector */
1054             dx00             = _mm_sub_pd(ix0,jx0);
1055             dy00             = _mm_sub_pd(iy0,jy0);
1056             dz00             = _mm_sub_pd(iz0,jz0);
1057             dx10             = _mm_sub_pd(ix1,jx0);
1058             dy10             = _mm_sub_pd(iy1,jy0);
1059             dz10             = _mm_sub_pd(iz1,jz0);
1060             dx20             = _mm_sub_pd(ix2,jx0);
1061             dy20             = _mm_sub_pd(iy2,jy0);
1062             dz20             = _mm_sub_pd(iz2,jz0);
1063             dx30             = _mm_sub_pd(ix3,jx0);
1064             dy30             = _mm_sub_pd(iy3,jy0);
1065             dz30             = _mm_sub_pd(iz3,jz0);
1066
1067             /* Calculate squared distance and things based on it */
1068             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1069             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1070             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1071             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1072
1073             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1074             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1075             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1076             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1077
1078             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1079             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1080             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1081
1082             /* Load parameters for j particles */
1083             jq0              = _mm_load_sd(charge+jnrA+0);
1084             vdwjidx0A        = 2*vdwtype[jnrA+0];
1085
1086             fjx0             = _mm_setzero_pd();
1087             fjy0             = _mm_setzero_pd();
1088             fjz0             = _mm_setzero_pd();
1089
1090             /**************************
1091              * CALCULATE INTERACTIONS *
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             /* Calculate table index by multiplying r with table scale and truncate to integer */
1100             rt               = _mm_mul_pd(r00,vftabscale);
1101             vfitab           = _mm_cvttpd_epi32(rt);
1102             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1103             vfitab           = _mm_slli_epi32(vfitab,3);
1104
1105             /* CUBIC SPLINE TABLE DISPERSION */
1106             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1107             F                = _mm_setzero_pd();
1108             GMX_MM_TRANSPOSE2_PD(Y,F);
1109             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1110             H                = _mm_setzero_pd();
1111             GMX_MM_TRANSPOSE2_PD(G,H);
1112             Heps             = _mm_mul_pd(vfeps,H);
1113             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1114             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1115             fvdw6            = _mm_mul_pd(c6_00,FF);
1116
1117             /* CUBIC SPLINE TABLE REPULSION */
1118             vfitab           = _mm_add_epi32(vfitab,ifour);
1119             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1120             F                = _mm_setzero_pd();
1121             GMX_MM_TRANSPOSE2_PD(Y,F);
1122             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1123             H                = _mm_setzero_pd();
1124             GMX_MM_TRANSPOSE2_PD(G,H);
1125             Heps             = _mm_mul_pd(vfeps,H);
1126             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1127             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1128             fvdw12           = _mm_mul_pd(c12_00,FF);
1129             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1130
1131             fscal            = fvdw;
1132
1133             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1134
1135             /* Calculate temporary vectorial force */
1136             tx               = _mm_mul_pd(fscal,dx00);
1137             ty               = _mm_mul_pd(fscal,dy00);
1138             tz               = _mm_mul_pd(fscal,dz00);
1139
1140             /* Update vectorial force */
1141             fix0             = _mm_add_pd(fix0,tx);
1142             fiy0             = _mm_add_pd(fiy0,ty);
1143             fiz0             = _mm_add_pd(fiz0,tz);
1144
1145             fjx0             = _mm_add_pd(fjx0,tx);
1146             fjy0             = _mm_add_pd(fjy0,ty);
1147             fjz0             = _mm_add_pd(fjz0,tz);
1148
1149             /**************************
1150              * CALCULATE INTERACTIONS *
1151              **************************/
1152
1153             if (gmx_mm_any_lt(rsq10,rcutoff2))
1154             {
1155
1156             /* Compute parameters for interactions between i and j atoms */
1157             qq10             = _mm_mul_pd(iq1,jq0);
1158
1159             /* REACTION-FIELD ELECTROSTATICS */
1160             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1161
1162             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1163
1164             fscal            = felec;
1165
1166             fscal            = _mm_and_pd(fscal,cutoff_mask);
1167
1168             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1169
1170             /* Calculate temporary vectorial force */
1171             tx               = _mm_mul_pd(fscal,dx10);
1172             ty               = _mm_mul_pd(fscal,dy10);
1173             tz               = _mm_mul_pd(fscal,dz10);
1174
1175             /* Update vectorial force */
1176             fix1             = _mm_add_pd(fix1,tx);
1177             fiy1             = _mm_add_pd(fiy1,ty);
1178             fiz1             = _mm_add_pd(fiz1,tz);
1179
1180             fjx0             = _mm_add_pd(fjx0,tx);
1181             fjy0             = _mm_add_pd(fjy0,ty);
1182             fjz0             = _mm_add_pd(fjz0,tz);
1183
1184             }
1185
1186             /**************************
1187              * CALCULATE INTERACTIONS *
1188              **************************/
1189
1190             if (gmx_mm_any_lt(rsq20,rcutoff2))
1191             {
1192
1193             /* Compute parameters for interactions between i and j atoms */
1194             qq20             = _mm_mul_pd(iq2,jq0);
1195
1196             /* REACTION-FIELD ELECTROSTATICS */
1197             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1198
1199             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1200
1201             fscal            = felec;
1202
1203             fscal            = _mm_and_pd(fscal,cutoff_mask);
1204
1205             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1206
1207             /* Calculate temporary vectorial force */
1208             tx               = _mm_mul_pd(fscal,dx20);
1209             ty               = _mm_mul_pd(fscal,dy20);
1210             tz               = _mm_mul_pd(fscal,dz20);
1211
1212             /* Update vectorial force */
1213             fix2             = _mm_add_pd(fix2,tx);
1214             fiy2             = _mm_add_pd(fiy2,ty);
1215             fiz2             = _mm_add_pd(fiz2,tz);
1216
1217             fjx0             = _mm_add_pd(fjx0,tx);
1218             fjy0             = _mm_add_pd(fjy0,ty);
1219             fjz0             = _mm_add_pd(fjz0,tz);
1220
1221             }
1222
1223             /**************************
1224              * CALCULATE INTERACTIONS *
1225              **************************/
1226
1227             if (gmx_mm_any_lt(rsq30,rcutoff2))
1228             {
1229
1230             /* Compute parameters for interactions between i and j atoms */
1231             qq30             = _mm_mul_pd(iq3,jq0);
1232
1233             /* REACTION-FIELD ELECTROSTATICS */
1234             felec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_mul_pd(rinv30,rinvsq30),krf2));
1235
1236             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1237
1238             fscal            = felec;
1239
1240             fscal            = _mm_and_pd(fscal,cutoff_mask);
1241
1242             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1243
1244             /* Calculate temporary vectorial force */
1245             tx               = _mm_mul_pd(fscal,dx30);
1246             ty               = _mm_mul_pd(fscal,dy30);
1247             tz               = _mm_mul_pd(fscal,dz30);
1248
1249             /* Update vectorial force */
1250             fix3             = _mm_add_pd(fix3,tx);
1251             fiy3             = _mm_add_pd(fiy3,ty);
1252             fiz3             = _mm_add_pd(fiz3,tz);
1253
1254             fjx0             = _mm_add_pd(fjx0,tx);
1255             fjy0             = _mm_add_pd(fjy0,ty);
1256             fjz0             = _mm_add_pd(fjz0,tz);
1257
1258             }
1259
1260             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1261
1262             /* Inner loop uses 141 flops */
1263         }
1264
1265         /* End of innermost loop */
1266
1267         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1268                                               f+i_coord_offset,fshift+i_shift_offset);
1269
1270         /* Increment number of inner iterations */
1271         inneriter                  += j_index_end - j_index_start;
1272
1273         /* Outer loop uses 24 flops */
1274     }
1275
1276     /* Increment number of outer iterations */
1277     outeriter        += nri;
1278
1279     /* Update outer/inner flops */
1280
1281     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*141);
1282 }