src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single 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_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_single
  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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     krf              = _mm_set1_ps(fr->ic->k_rf);
 109     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 110     crf              = _mm_set1_ps(fr->ic->c_rf);
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     /* Setup water-specific parameters */
 116     inr              = nlist->iinr[0];
 117     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 118     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 119     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 120     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 126
 127     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 128     rvdw             = _mm_set1_ps(fr->rvdw);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 162
 163         fix0             = _mm_setzero_ps();
 164         fiy0             = _mm_setzero_ps();
 165         fiz0             = _mm_setzero_ps();
 166         fix1             = _mm_setzero_ps();
 167         fiy1             = _mm_setzero_ps();
 168         fiz1             = _mm_setzero_ps();
 169         fix2             = _mm_setzero_ps();
 170         fiy2             = _mm_setzero_ps();
 171         fiz2             = _mm_setzero_ps();
 172         fix3             = _mm_setzero_ps();
 173         fiy3             = _mm_setzero_ps();
 174         fiz3             = _mm_setzero_ps();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_ps();
 178         vvdwsum          = _mm_setzero_ps();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               x+j_coord_offsetC,x+j_coord_offsetD,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_ps(ix0,jx0);
 201             dy00             = _mm_sub_ps(iy0,jy0);
 202             dz00             = _mm_sub_ps(iz0,jz0);
 203             dx10             = _mm_sub_ps(ix1,jx0);
 204             dy10             = _mm_sub_ps(iy1,jy0);
 205             dz10             = _mm_sub_ps(iz1,jz0);
 206             dx20             = _mm_sub_ps(ix2,jx0);
 207             dy20             = _mm_sub_ps(iy2,jy0);
 208             dz20             = _mm_sub_ps(iz2,jz0);
 209             dx30             = _mm_sub_ps(ix3,jx0);
 210             dy30             = _mm_sub_ps(iy3,jy0);
 211             dz30             = _mm_sub_ps(iz3,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 217             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 218
 219             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 220             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 221             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 222
 223             rinvsq00         = gmx_mm_inv_ps(rsq00);
 224             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 225             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 226             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 230                                                               charge+jnrC+0,charge+jnrD+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233             vdwjidx0C        = 2*vdwtype[jnrC+0];
 234             vdwjidx0D        = 2*vdwtype[jnrD+0];
 235
 236             fjx0             = _mm_setzero_ps();
 237             fjy0             = _mm_setzero_ps();
 238             fjz0             = _mm_setzero_ps();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             if (gmx_mm_any_lt(rsq00,rcutoff2))
 245             {
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 249                                          vdwparam+vdwioffset0+vdwjidx0B,
 250                                          vdwparam+vdwioffset0+vdwjidx0C,
 251                                          vdwparam+vdwioffset0+vdwjidx0D,
 252                                          &c6_00,&c12_00);
 253
 254             /* LENNARD-JONES DISPERSION/REPULSION */
 255
 256             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 257             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 258             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 259             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 260                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 261             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 262
 263             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 267             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = fvdw;
 270
 271             fscal            = _mm_and_ps(fscal,cutoff_mask);
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm_mul_ps(fscal,dx00);
 275             ty               = _mm_mul_ps(fscal,dy00);
 276             tz               = _mm_mul_ps(fscal,dz00);
 277
 278             /* Update vectorial force */
 279             fix0             = _mm_add_ps(fix0,tx);
 280             fiy0             = _mm_add_ps(fiy0,ty);
 281             fiz0             = _mm_add_ps(fiz0,tz);
 282
 283             fjx0             = _mm_add_ps(fjx0,tx);
 284             fjy0             = _mm_add_ps(fjy0,ty);
 285             fjz0             = _mm_add_ps(fjz0,tz);
 286
 287             }
 288
 289             /**************************
 290              * CALCULATE INTERACTIONS *
 291              **************************/
 292
 293             if (gmx_mm_any_lt(rsq10,rcutoff2))
 294             {
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq10             = _mm_mul_ps(iq1,jq0);
 298
 299             /* REACTION-FIELD ELECTROSTATICS */
 300             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 301             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 302
 303             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velec            = _mm_and_ps(velec,cutoff_mask);
 307             velecsum         = _mm_add_ps(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             fscal            = _mm_and_ps(fscal,cutoff_mask);
 312
 313             /* Calculate temporary vectorial force */
 314             tx               = _mm_mul_ps(fscal,dx10);
 315             ty               = _mm_mul_ps(fscal,dy10);
 316             tz               = _mm_mul_ps(fscal,dz10);
 317
 318             /* Update vectorial force */
 319             fix1             = _mm_add_ps(fix1,tx);
 320             fiy1             = _mm_add_ps(fiy1,ty);
 321             fiz1             = _mm_add_ps(fiz1,tz);
 322
 323             fjx0             = _mm_add_ps(fjx0,tx);
 324             fjy0             = _mm_add_ps(fjy0,ty);
 325             fjz0             = _mm_add_ps(fjz0,tz);
 326
 327             }
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             if (gmx_mm_any_lt(rsq20,rcutoff2))
 334             {
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq20             = _mm_mul_ps(iq2,jq0);
 338
 339             /* REACTION-FIELD ELECTROSTATICS */
 340             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 341             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 342
 343             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm_and_ps(velec,cutoff_mask);
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             fscal            = _mm_and_ps(fscal,cutoff_mask);
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm_mul_ps(fscal,dx20);
 355             ty               = _mm_mul_ps(fscal,dy20);
 356             tz               = _mm_mul_ps(fscal,dz20);
 357
 358             /* Update vectorial force */
 359             fix2             = _mm_add_ps(fix2,tx);
 360             fiy2             = _mm_add_ps(fiy2,ty);
 361             fiz2             = _mm_add_ps(fiz2,tz);
 362
 363             fjx0             = _mm_add_ps(fjx0,tx);
 364             fjy0             = _mm_add_ps(fjy0,ty);
 365             fjz0             = _mm_add_ps(fjz0,tz);
 366
 367             }
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             if (gmx_mm_any_lt(rsq30,rcutoff2))
 374             {
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq30             = _mm_mul_ps(iq3,jq0);
 378
 379             /* REACTION-FIELD ELECTROSTATICS */
 380             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 381             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 382
 383             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velec            = _mm_and_ps(velec,cutoff_mask);
 387             velecsum         = _mm_add_ps(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             fscal            = _mm_and_ps(fscal,cutoff_mask);
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm_mul_ps(fscal,dx30);
 395             ty               = _mm_mul_ps(fscal,dy30);
 396             tz               = _mm_mul_ps(fscal,dz30);
 397
 398             /* Update vectorial force */
 399             fix3             = _mm_add_ps(fix3,tx);
 400             fiy3             = _mm_add_ps(fiy3,ty);
 401             fiz3             = _mm_add_ps(fiz3,tz);
 402
 403             fjx0             = _mm_add_ps(fjx0,tx);
 404             fjy0             = _mm_add_ps(fjy0,ty);
 405             fjz0             = _mm_add_ps(fjz0,tz);
 406
 407             }
 408
 409             fjptrA             = f+j_coord_offsetA;
 410             fjptrB             = f+j_coord_offsetB;
 411             fjptrC             = f+j_coord_offsetC;
 412             fjptrD             = f+j_coord_offsetD;
 413
 414             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 415
 416             /* Inner loop uses 149 flops */
 417         }
 418
 419         if(jidx<j_index_end)
 420         {
 421
 422             /* Get j neighbor index, and coordinate index */
 423             jnrlistA         = jjnr[jidx];
 424             jnrlistB         = jjnr[jidx+1];
 425             jnrlistC         = jjnr[jidx+2];
 426             jnrlistD         = jjnr[jidx+3];
 427             /* Sign of each element will be negative for non-real atoms.
 428              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 429              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 430              */
 431             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 432             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 433             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 434             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 435             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 443                                               x+j_coord_offsetC,x+j_coord_offsetD,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_ps(ix0,jx0);
 448             dy00             = _mm_sub_ps(iy0,jy0);
 449             dz00             = _mm_sub_ps(iz0,jz0);
 450             dx10             = _mm_sub_ps(ix1,jx0);
 451             dy10             = _mm_sub_ps(iy1,jy0);
 452             dz10             = _mm_sub_ps(iz1,jz0);
 453             dx20             = _mm_sub_ps(ix2,jx0);
 454             dy20             = _mm_sub_ps(iy2,jy0);
 455             dz20             = _mm_sub_ps(iz2,jz0);
 456             dx30             = _mm_sub_ps(ix3,jx0);
 457             dy30             = _mm_sub_ps(iy3,jy0);
 458             dz30             = _mm_sub_ps(iz3,jz0);
 459
 460             /* Calculate squared distance and things based on it */
 461             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 462             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 463             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 464             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 465
 466             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 467             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 468             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 469
 470             rinvsq00         = gmx_mm_inv_ps(rsq00);
 471             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 472             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 473             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 474
 475             /* Load parameters for j particles */
 476             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 477                                                               charge+jnrC+0,charge+jnrD+0);
 478             vdwjidx0A        = 2*vdwtype[jnrA+0];
 479             vdwjidx0B        = 2*vdwtype[jnrB+0];
 480             vdwjidx0C        = 2*vdwtype[jnrC+0];
 481             vdwjidx0D        = 2*vdwtype[jnrD+0];
 482
 483             fjx0             = _mm_setzero_ps();
 484             fjy0             = _mm_setzero_ps();
 485             fjz0             = _mm_setzero_ps();
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             if (gmx_mm_any_lt(rsq00,rcutoff2))
 492             {
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 496                                          vdwparam+vdwioffset0+vdwjidx0B,
 497                                          vdwparam+vdwioffset0+vdwjidx0C,
 498                                          vdwparam+vdwioffset0+vdwjidx0D,
 499                                          &c6_00,&c12_00);
 500
 501             /* LENNARD-JONES DISPERSION/REPULSION */
 502
 503             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 504             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 505             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 506             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 507                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 508             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 509
 510             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 511
 512             /* Update potential sum for this i atom from the interaction with this j atom. */
 513             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 514             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 515             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 516
 517             fscal            = fvdw;
 518
 519             fscal            = _mm_and_ps(fscal,cutoff_mask);
 520
 521             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 522
 523             /* Calculate temporary vectorial force */
 524             tx               = _mm_mul_ps(fscal,dx00);
 525             ty               = _mm_mul_ps(fscal,dy00);
 526             tz               = _mm_mul_ps(fscal,dz00);
 527
 528             /* Update vectorial force */
 529             fix0             = _mm_add_ps(fix0,tx);
 530             fiy0             = _mm_add_ps(fiy0,ty);
 531             fiz0             = _mm_add_ps(fiz0,tz);
 532
 533             fjx0             = _mm_add_ps(fjx0,tx);
 534             fjy0             = _mm_add_ps(fjy0,ty);
 535             fjz0             = _mm_add_ps(fjz0,tz);
 536
 537             }
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             if (gmx_mm_any_lt(rsq10,rcutoff2))
 544             {
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq10             = _mm_mul_ps(iq1,jq0);
 548
 549             /* REACTION-FIELD ELECTROSTATICS */
 550             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 551             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 552
 553             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 554
 555             /* Update potential sum for this i atom from the interaction with this j atom. */
 556             velec            = _mm_and_ps(velec,cutoff_mask);
 557             velec            = _mm_andnot_ps(dummy_mask,velec);
 558             velecsum         = _mm_add_ps(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_and_ps(fscal,cutoff_mask);
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_ps(fscal,dx10);
 568             ty               = _mm_mul_ps(fscal,dy10);
 569             tz               = _mm_mul_ps(fscal,dz10);
 570
 571             /* Update vectorial force */
 572             fix1             = _mm_add_ps(fix1,tx);
 573             fiy1             = _mm_add_ps(fiy1,ty);
 574             fiz1             = _mm_add_ps(fiz1,tz);
 575
 576             fjx0             = _mm_add_ps(fjx0,tx);
 577             fjy0             = _mm_add_ps(fjy0,ty);
 578             fjz0             = _mm_add_ps(fjz0,tz);
 579
 580             }
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             if (gmx_mm_any_lt(rsq20,rcutoff2))
 587             {
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq20             = _mm_mul_ps(iq2,jq0);
 591
 592             /* REACTION-FIELD ELECTROSTATICS */
 593             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 594             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 595
 596             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             velec            = _mm_and_ps(velec,cutoff_mask);
 600             velec            = _mm_andnot_ps(dummy_mask,velec);
 601             velecsum         = _mm_add_ps(velecsum,velec);
 602
 603             fscal            = felec;
 604
 605             fscal            = _mm_and_ps(fscal,cutoff_mask);
 606
 607             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm_mul_ps(fscal,dx20);
 611             ty               = _mm_mul_ps(fscal,dy20);
 612             tz               = _mm_mul_ps(fscal,dz20);
 613
 614             /* Update vectorial force */
 615             fix2             = _mm_add_ps(fix2,tx);
 616             fiy2             = _mm_add_ps(fiy2,ty);
 617             fiz2             = _mm_add_ps(fiz2,tz);
 618
 619             fjx0             = _mm_add_ps(fjx0,tx);
 620             fjy0             = _mm_add_ps(fjy0,ty);
 621             fjz0             = _mm_add_ps(fjz0,tz);
 622
 623             }
 624
 625             /**************************
 626              * CALCULATE INTERACTIONS *
 627              **************************/
 628
 629             if (gmx_mm_any_lt(rsq30,rcutoff2))
 630             {
 631
 632             /* Compute parameters for interactions between i and j atoms */
 633             qq30             = _mm_mul_ps(iq3,jq0);
 634
 635             /* REACTION-FIELD ELECTROSTATICS */
 636             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 637             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 638
 639             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velec            = _mm_and_ps(velec,cutoff_mask);
 643             velec            = _mm_andnot_ps(dummy_mask,velec);
 644             velecsum         = _mm_add_ps(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm_and_ps(fscal,cutoff_mask);
 649
 650             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 651
 652             /* Calculate temporary vectorial force */
 653             tx               = _mm_mul_ps(fscal,dx30);
 654             ty               = _mm_mul_ps(fscal,dy30);
 655             tz               = _mm_mul_ps(fscal,dz30);
 656
 657             /* Update vectorial force */
 658             fix3             = _mm_add_ps(fix3,tx);
 659             fiy3             = _mm_add_ps(fiy3,ty);
 660             fiz3             = _mm_add_ps(fiz3,tz);
 661
 662             fjx0             = _mm_add_ps(fjx0,tx);
 663             fjy0             = _mm_add_ps(fjy0,ty);
 664             fjz0             = _mm_add_ps(fjz0,tz);
 665
 666             }
 667
 668             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 669             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 670             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 671             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 672
 673             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 674
 675             /* Inner loop uses 149 flops */
 676         }
 677
 678         /* End of innermost loop */
 679
 680         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 681                                               f+i_coord_offset,fshift+i_shift_offset);
 682
 683         ggid                        = gid[iidx];
 684         /* Update potential energies */
 685         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 686         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 687
 688         /* Increment number of inner iterations */
 689         inneriter                  += j_index_end - j_index_start;
 690
 691         /* Outer loop uses 26 flops */
 692     }
 693
 694     /* Increment number of outer iterations */
 695     outeriter        += nri;
 696
 697     /* Update outer/inner flops */
 698
 699     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*149);
 700 }
 701 /*
 702  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse4_1_single
 703  * Electrostatics interaction: ReactionField
 704  * VdW interaction:            LennardJones
 705  * Geometry:                   Water4-Particle
 706  * Calculate force/pot:        Force
 707  */
 708 void
 709 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse4_1_single
 710                     (t_nblist * gmx_restrict                nlist,
 711                      rvec * gmx_restrict                    xx,
 712                      rvec * gmx_restrict                    ff,
 713                      t_forcerec * gmx_restrict              fr,
 714                      t_mdatoms * gmx_restrict               mdatoms,
 715                      nb_kernel_data_t * gmx_restrict        kernel_data,
 716                      t_nrnb * gmx_restrict                  nrnb)
 717 {
 718     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 719      * just 0 for non-waters.
 720      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 721      * jnr indices corresponding to data put in the four positions in the SIMD register.
 722      */
 723     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 724     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 725     int              jnrA,jnrB,jnrC,jnrD;
 726     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 727     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 728     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 729     real             rcutoff_scalar;
 730     real             *shiftvec,*fshift,*x,*f;
 731     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 732     real             scratch[4*DIM];
 733     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 734     int              vdwioffset0;
 735     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 736     int              vdwioffset1;
 737     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 738     int              vdwioffset2;
 739     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 740     int              vdwioffset3;
 741     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 742     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 743     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 744     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 745     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 746     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 747     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 748     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 749     real             *charge;
 750     int              nvdwtype;
 751     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 752     int              *vdwtype;
 753     real             *vdwparam;
 754     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 755     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 756     __m128           dummy_mask,cutoff_mask;
 757     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 758     __m128           one     = _mm_set1_ps(1.0);
 759     __m128           two     = _mm_set1_ps(2.0);
 760     x                = xx[0];
 761     f                = ff[0];
 762
 763     nri              = nlist->nri;
 764     iinr             = nlist->iinr;
 765     jindex           = nlist->jindex;
 766     jjnr             = nlist->jjnr;
 767     shiftidx         = nlist->shift;
 768     gid              = nlist->gid;
 769     shiftvec         = fr->shift_vec[0];
 770     fshift           = fr->fshift[0];
 771     facel            = _mm_set1_ps(fr->epsfac);
 772     charge           = mdatoms->chargeA;
 773     krf              = _mm_set1_ps(fr->ic->k_rf);
 774     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 775     crf              = _mm_set1_ps(fr->ic->c_rf);
 776     nvdwtype         = fr->ntype;
 777     vdwparam         = fr->nbfp;
 778     vdwtype          = mdatoms->typeA;
 779
 780     /* Setup water-specific parameters */
 781     inr              = nlist->iinr[0];
 782     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 783     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 784     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 785     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 786
 787     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 788     rcutoff_scalar   = fr->rcoulomb;
 789     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 790     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 791
 792     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 793     rvdw             = _mm_set1_ps(fr->rvdw);
 794
 795     /* Avoid stupid compiler warnings */
 796     jnrA = jnrB = jnrC = jnrD = 0;
 797     j_coord_offsetA = 0;
 798     j_coord_offsetB = 0;
 799     j_coord_offsetC = 0;
 800     j_coord_offsetD = 0;
 801
 802     outeriter        = 0;
 803     inneriter        = 0;
 804
 805     for(iidx=0;iidx<4*DIM;iidx++)
 806     {
 807         scratch[iidx] = 0.0;
 808     }
 809
 810     /* Start outer loop over neighborlists */
 811     for(iidx=0; iidx<nri; iidx++)
 812     {
 813         /* Load shift vector for this list */
 814         i_shift_offset   = DIM*shiftidx[iidx];
 815
 816         /* Load limits for loop over neighbors */
 817         j_index_start    = jindex[iidx];
 818         j_index_end      = jindex[iidx+1];
 819
 820         /* Get outer coordinate index */
 821         inr              = iinr[iidx];
 822         i_coord_offset   = DIM*inr;
 823
 824         /* Load i particle coords and add shift vector */
 825         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 826                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 827
 828         fix0             = _mm_setzero_ps();
 829         fiy0             = _mm_setzero_ps();
 830         fiz0             = _mm_setzero_ps();
 831         fix1             = _mm_setzero_ps();
 832         fiy1             = _mm_setzero_ps();
 833         fiz1             = _mm_setzero_ps();
 834         fix2             = _mm_setzero_ps();
 835         fiy2             = _mm_setzero_ps();
 836         fiz2             = _mm_setzero_ps();
 837         fix3             = _mm_setzero_ps();
 838         fiy3             = _mm_setzero_ps();
 839         fiz3             = _mm_setzero_ps();
 840
 841         /* Start inner kernel loop */
 842         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 843         {
 844
 845             /* Get j neighbor index, and coordinate index */
 846             jnrA             = jjnr[jidx];
 847             jnrB             = jjnr[jidx+1];
 848             jnrC             = jjnr[jidx+2];
 849             jnrD             = jjnr[jidx+3];
 850             j_coord_offsetA  = DIM*jnrA;
 851             j_coord_offsetB  = DIM*jnrB;
 852             j_coord_offsetC  = DIM*jnrC;
 853             j_coord_offsetD  = DIM*jnrD;
 854
 855             /* load j atom coordinates */
 856             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 857                                               x+j_coord_offsetC,x+j_coord_offsetD,
 858                                               &jx0,&jy0,&jz0);
 859
 860             /* Calculate displacement vector */
 861             dx00             = _mm_sub_ps(ix0,jx0);
 862             dy00             = _mm_sub_ps(iy0,jy0);
 863             dz00             = _mm_sub_ps(iz0,jz0);
 864             dx10             = _mm_sub_ps(ix1,jx0);
 865             dy10             = _mm_sub_ps(iy1,jy0);
 866             dz10             = _mm_sub_ps(iz1,jz0);
 867             dx20             = _mm_sub_ps(ix2,jx0);
 868             dy20             = _mm_sub_ps(iy2,jy0);
 869             dz20             = _mm_sub_ps(iz2,jz0);
 870             dx30             = _mm_sub_ps(ix3,jx0);
 871             dy30             = _mm_sub_ps(iy3,jy0);
 872             dz30             = _mm_sub_ps(iz3,jz0);
 873
 874             /* Calculate squared distance and things based on it */
 875             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 876             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 877             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 878             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 879
 880             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 881             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 882             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 883
 884             rinvsq00         = gmx_mm_inv_ps(rsq00);
 885             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 886             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 887             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 888
 889             /* Load parameters for j particles */
 890             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 891                                                               charge+jnrC+0,charge+jnrD+0);
 892             vdwjidx0A        = 2*vdwtype[jnrA+0];
 893             vdwjidx0B        = 2*vdwtype[jnrB+0];
 894             vdwjidx0C        = 2*vdwtype[jnrC+0];
 895             vdwjidx0D        = 2*vdwtype[jnrD+0];
 896
 897             fjx0             = _mm_setzero_ps();
 898             fjy0             = _mm_setzero_ps();
 899             fjz0             = _mm_setzero_ps();
 900
 901             /**************************
 902              * CALCULATE INTERACTIONS *
 903              **************************/
 904
 905             if (gmx_mm_any_lt(rsq00,rcutoff2))
 906             {
 907
 908             /* Compute parameters for interactions between i and j atoms */
 909             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 910                                          vdwparam+vdwioffset0+vdwjidx0B,
 911                                          vdwparam+vdwioffset0+vdwjidx0C,
 912                                          vdwparam+vdwioffset0+vdwjidx0D,
 913                                          &c6_00,&c12_00);
 914
 915             /* LENNARD-JONES DISPERSION/REPULSION */
 916
 917             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 918             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 919
 920             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 921
 922             fscal            = fvdw;
 923
 924             fscal            = _mm_and_ps(fscal,cutoff_mask);
 925
 926             /* Calculate temporary vectorial force */
 927             tx               = _mm_mul_ps(fscal,dx00);
 928             ty               = _mm_mul_ps(fscal,dy00);
 929             tz               = _mm_mul_ps(fscal,dz00);
 930
 931             /* Update vectorial force */
 932             fix0             = _mm_add_ps(fix0,tx);
 933             fiy0             = _mm_add_ps(fiy0,ty);
 934             fiz0             = _mm_add_ps(fiz0,tz);
 935
 936             fjx0             = _mm_add_ps(fjx0,tx);
 937             fjy0             = _mm_add_ps(fjy0,ty);
 938             fjz0             = _mm_add_ps(fjz0,tz);
 939
 940             }
 941
 942             /**************************
 943              * CALCULATE INTERACTIONS *
 944              **************************/
 945
 946             if (gmx_mm_any_lt(rsq10,rcutoff2))
 947             {
 948
 949             /* Compute parameters for interactions between i and j atoms */
 950             qq10             = _mm_mul_ps(iq1,jq0);
 951
 952             /* REACTION-FIELD ELECTROSTATICS */
 953             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 954
 955             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 956
 957             fscal            = felec;
 958
 959             fscal            = _mm_and_ps(fscal,cutoff_mask);
 960
 961             /* Calculate temporary vectorial force */
 962             tx               = _mm_mul_ps(fscal,dx10);
 963             ty               = _mm_mul_ps(fscal,dy10);
 964             tz               = _mm_mul_ps(fscal,dz10);
 965
 966             /* Update vectorial force */
 967             fix1             = _mm_add_ps(fix1,tx);
 968             fiy1             = _mm_add_ps(fiy1,ty);
 969             fiz1             = _mm_add_ps(fiz1,tz);
 970
 971             fjx0             = _mm_add_ps(fjx0,tx);
 972             fjy0             = _mm_add_ps(fjy0,ty);
 973             fjz0             = _mm_add_ps(fjz0,tz);
 974
 975             }
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             if (gmx_mm_any_lt(rsq20,rcutoff2))
 982             {
 983
 984             /* Compute parameters for interactions between i and j atoms */
 985             qq20             = _mm_mul_ps(iq2,jq0);
 986
 987             /* REACTION-FIELD ELECTROSTATICS */
 988             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 989
 990             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 991
 992             fscal            = felec;
 993
 994             fscal            = _mm_and_ps(fscal,cutoff_mask);
 995
 996             /* Calculate temporary vectorial force */
 997             tx               = _mm_mul_ps(fscal,dx20);
 998             ty               = _mm_mul_ps(fscal,dy20);
 999             tz               = _mm_mul_ps(fscal,dz20);
1000
1001             /* Update vectorial force */
1002             fix2             = _mm_add_ps(fix2,tx);
1003             fiy2             = _mm_add_ps(fiy2,ty);
1004             fiz2             = _mm_add_ps(fiz2,tz);
1005
1006             fjx0             = _mm_add_ps(fjx0,tx);
1007             fjy0             = _mm_add_ps(fjy0,ty);
1008             fjz0             = _mm_add_ps(fjz0,tz);
1009
1010             }
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             if (gmx_mm_any_lt(rsq30,rcutoff2))
1017             {
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq30             = _mm_mul_ps(iq3,jq0);
1021
1022             /* REACTION-FIELD ELECTROSTATICS */
1023             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1024
1025             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1026
1027             fscal            = felec;
1028
1029             fscal            = _mm_and_ps(fscal,cutoff_mask);
1030
1031             /* Calculate temporary vectorial force */
1032             tx               = _mm_mul_ps(fscal,dx30);
1033             ty               = _mm_mul_ps(fscal,dy30);
1034             tz               = _mm_mul_ps(fscal,dz30);
1035
1036             /* Update vectorial force */
1037             fix3             = _mm_add_ps(fix3,tx);
1038             fiy3             = _mm_add_ps(fiy3,ty);
1039             fiz3             = _mm_add_ps(fiz3,tz);
1040
1041             fjx0             = _mm_add_ps(fjx0,tx);
1042             fjy0             = _mm_add_ps(fjy0,ty);
1043             fjz0             = _mm_add_ps(fjz0,tz);
1044
1045             }
1046
1047             fjptrA             = f+j_coord_offsetA;
1048             fjptrB             = f+j_coord_offsetB;
1049             fjptrC             = f+j_coord_offsetC;
1050             fjptrD             = f+j_coord_offsetD;
1051
1052             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1053
1054             /* Inner loop uses 120 flops */
1055         }
1056
1057         if(jidx<j_index_end)
1058         {
1059
1060             /* Get j neighbor index, and coordinate index */
1061             jnrlistA         = jjnr[jidx];
1062             jnrlistB         = jjnr[jidx+1];
1063             jnrlistC         = jjnr[jidx+2];
1064             jnrlistD         = jjnr[jidx+3];
1065             /* Sign of each element will be negative for non-real atoms.
1066              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1067              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1068              */
1069             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1070             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1071             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1072             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1073             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1074             j_coord_offsetA  = DIM*jnrA;
1075             j_coord_offsetB  = DIM*jnrB;
1076             j_coord_offsetC  = DIM*jnrC;
1077             j_coord_offsetD  = DIM*jnrD;
1078
1079             /* load j atom coordinates */
1080             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1081                                               x+j_coord_offsetC,x+j_coord_offsetD,
1082                                               &jx0,&jy0,&jz0);
1083
1084             /* Calculate displacement vector */
1085             dx00             = _mm_sub_ps(ix0,jx0);
1086             dy00             = _mm_sub_ps(iy0,jy0);
1087             dz00             = _mm_sub_ps(iz0,jz0);
1088             dx10             = _mm_sub_ps(ix1,jx0);
1089             dy10             = _mm_sub_ps(iy1,jy0);
1090             dz10             = _mm_sub_ps(iz1,jz0);
1091             dx20             = _mm_sub_ps(ix2,jx0);
1092             dy20             = _mm_sub_ps(iy2,jy0);
1093             dz20             = _mm_sub_ps(iz2,jz0);
1094             dx30             = _mm_sub_ps(ix3,jx0);
1095             dy30             = _mm_sub_ps(iy3,jy0);
1096             dz30             = _mm_sub_ps(iz3,jz0);
1097
1098             /* Calculate squared distance and things based on it */
1099             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1100             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1101             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1102             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1103
1104             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1105             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1106             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1107
1108             rinvsq00         = gmx_mm_inv_ps(rsq00);
1109             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1110             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1111             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1112
1113             /* Load parameters for j particles */
1114             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1115                                                               charge+jnrC+0,charge+jnrD+0);
1116             vdwjidx0A        = 2*vdwtype[jnrA+0];
1117             vdwjidx0B        = 2*vdwtype[jnrB+0];
1118             vdwjidx0C        = 2*vdwtype[jnrC+0];
1119             vdwjidx0D        = 2*vdwtype[jnrD+0];
1120
1121             fjx0             = _mm_setzero_ps();
1122             fjy0             = _mm_setzero_ps();
1123             fjz0             = _mm_setzero_ps();
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             if (gmx_mm_any_lt(rsq00,rcutoff2))
1130             {
1131
1132             /* Compute parameters for interactions between i and j atoms */
1133             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1134                                          vdwparam+vdwioffset0+vdwjidx0B,
1135                                          vdwparam+vdwioffset0+vdwjidx0C,
1136                                          vdwparam+vdwioffset0+vdwjidx0D,
1137                                          &c6_00,&c12_00);
1138
1139             /* LENNARD-JONES DISPERSION/REPULSION */
1140
1141             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1142             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1143
1144             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1145
1146             fscal            = fvdw;
1147
1148             fscal            = _mm_and_ps(fscal,cutoff_mask);
1149
1150             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1151
1152             /* Calculate temporary vectorial force */
1153             tx               = _mm_mul_ps(fscal,dx00);
1154             ty               = _mm_mul_ps(fscal,dy00);
1155             tz               = _mm_mul_ps(fscal,dz00);
1156
1157             /* Update vectorial force */
1158             fix0             = _mm_add_ps(fix0,tx);
1159             fiy0             = _mm_add_ps(fiy0,ty);
1160             fiz0             = _mm_add_ps(fiz0,tz);
1161
1162             fjx0             = _mm_add_ps(fjx0,tx);
1163             fjy0             = _mm_add_ps(fjy0,ty);
1164             fjz0             = _mm_add_ps(fjz0,tz);
1165
1166             }
1167
1168             /**************************
1169              * CALCULATE INTERACTIONS *
1170              **************************/
1171
1172             if (gmx_mm_any_lt(rsq10,rcutoff2))
1173             {
1174
1175             /* Compute parameters for interactions between i and j atoms */
1176             qq10             = _mm_mul_ps(iq1,jq0);
1177
1178             /* REACTION-FIELD ELECTROSTATICS */
1179             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1180
1181             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1182
1183             fscal            = felec;
1184
1185             fscal            = _mm_and_ps(fscal,cutoff_mask);
1186
1187             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1188
1189             /* Calculate temporary vectorial force */
1190             tx               = _mm_mul_ps(fscal,dx10);
1191             ty               = _mm_mul_ps(fscal,dy10);
1192             tz               = _mm_mul_ps(fscal,dz10);
1193
1194             /* Update vectorial force */
1195             fix1             = _mm_add_ps(fix1,tx);
1196             fiy1             = _mm_add_ps(fiy1,ty);
1197             fiz1             = _mm_add_ps(fiz1,tz);
1198
1199             fjx0             = _mm_add_ps(fjx0,tx);
1200             fjy0             = _mm_add_ps(fjy0,ty);
1201             fjz0             = _mm_add_ps(fjz0,tz);
1202
1203             }
1204
1205             /**************************
1206              * CALCULATE INTERACTIONS *
1207              **************************/
1208
1209             if (gmx_mm_any_lt(rsq20,rcutoff2))
1210             {
1211
1212             /* Compute parameters for interactions between i and j atoms */
1213             qq20             = _mm_mul_ps(iq2,jq0);
1214
1215             /* REACTION-FIELD ELECTROSTATICS */
1216             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1217
1218             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1219
1220             fscal            = felec;
1221
1222             fscal            = _mm_and_ps(fscal,cutoff_mask);
1223
1224             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1225
1226             /* Calculate temporary vectorial force */
1227             tx               = _mm_mul_ps(fscal,dx20);
1228             ty               = _mm_mul_ps(fscal,dy20);
1229             tz               = _mm_mul_ps(fscal,dz20);
1230
1231             /* Update vectorial force */
1232             fix2             = _mm_add_ps(fix2,tx);
1233             fiy2             = _mm_add_ps(fiy2,ty);
1234             fiz2             = _mm_add_ps(fiz2,tz);
1235
1236             fjx0             = _mm_add_ps(fjx0,tx);
1237             fjy0             = _mm_add_ps(fjy0,ty);
1238             fjz0             = _mm_add_ps(fjz0,tz);
1239
1240             }
1241
1242             /**************************
1243              * CALCULATE INTERACTIONS *
1244              **************************/
1245
1246             if (gmx_mm_any_lt(rsq30,rcutoff2))
1247             {
1248
1249             /* Compute parameters for interactions between i and j atoms */
1250             qq30             = _mm_mul_ps(iq3,jq0);
1251
1252             /* REACTION-FIELD ELECTROSTATICS */
1253             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1254
1255             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1256
1257             fscal            = felec;
1258
1259             fscal            = _mm_and_ps(fscal,cutoff_mask);
1260
1261             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1262
1263             /* Calculate temporary vectorial force */
1264             tx               = _mm_mul_ps(fscal,dx30);
1265             ty               = _mm_mul_ps(fscal,dy30);
1266             tz               = _mm_mul_ps(fscal,dz30);
1267
1268             /* Update vectorial force */
1269             fix3             = _mm_add_ps(fix3,tx);
1270             fiy3             = _mm_add_ps(fiy3,ty);
1271             fiz3             = _mm_add_ps(fiz3,tz);
1272
1273             fjx0             = _mm_add_ps(fjx0,tx);
1274             fjy0             = _mm_add_ps(fjy0,ty);
1275             fjz0             = _mm_add_ps(fjz0,tz);
1276
1277             }
1278
1279             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1280             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1281             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1282             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1283
1284             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1285
1286             /* Inner loop uses 120 flops */
1287         }
1288
1289         /* End of innermost loop */
1290
1291         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1292                                               f+i_coord_offset,fshift+i_shift_offset);
1293
1294         /* Increment number of inner iterations */
1295         inneriter                  += j_index_end - j_index_start;
1296
1297         /* Outer loop uses 24 flops */
1298     }
1299
1300     /* Increment number of outer iterations */
1301     outeriter        += nri;
1302
1303     /* Update outer/inner flops */
1304
1305     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*120);
1306 }