src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     krf              = _mm_set1_ps(fr->ic->k_rf);
 123     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 124     crf              = _mm_set1_ps(fr->ic->c_rf);
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 132     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 133     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 137     rcutoff_scalar   = fr->rcoulomb;
 138     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 139     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 140
 141     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 142     rvdw             = _mm_set1_ps(fr->rvdw);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 176
 177         fix0             = _mm_setzero_ps();
 178         fiy0             = _mm_setzero_ps();
 179         fiz0             = _mm_setzero_ps();
 180         fix1             = _mm_setzero_ps();
 181         fiy1             = _mm_setzero_ps();
 182         fiz1             = _mm_setzero_ps();
 183         fix2             = _mm_setzero_ps();
 184         fiy2             = _mm_setzero_ps();
 185         fiz2             = _mm_setzero_ps();
 186         fix3             = _mm_setzero_ps();
 187         fiy3             = _mm_setzero_ps();
 188         fiz3             = _mm_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_ps();
 192         vvdwsum          = _mm_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                               x+j_coord_offsetC,x+j_coord_offsetD,
 211                                               &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm_sub_ps(ix0,jx0);
 215             dy00             = _mm_sub_ps(iy0,jy0);
 216             dz00             = _mm_sub_ps(iz0,jz0);
 217             dx10             = _mm_sub_ps(ix1,jx0);
 218             dy10             = _mm_sub_ps(iy1,jy0);
 219             dz10             = _mm_sub_ps(iz1,jz0);
 220             dx20             = _mm_sub_ps(ix2,jx0);
 221             dy20             = _mm_sub_ps(iy2,jy0);
 222             dz20             = _mm_sub_ps(iz2,jz0);
 223             dx30             = _mm_sub_ps(ix3,jx0);
 224             dy30             = _mm_sub_ps(iy3,jy0);
 225             dz30             = _mm_sub_ps(iz3,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 229             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 230             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 231             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 232
 233             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 234             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 235             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 236
 237             rinvsq00         = gmx_mm_inv_ps(rsq00);
 238             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 239             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 240             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                               charge+jnrC+0,charge+jnrD+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246             vdwjidx0B        = 2*vdwtype[jnrB+0];
 247             vdwjidx0C        = 2*vdwtype[jnrC+0];
 248             vdwjidx0D        = 2*vdwtype[jnrD+0];
 249
 250             fjx0             = _mm_setzero_ps();
 251             fjy0             = _mm_setzero_ps();
 252             fjz0             = _mm_setzero_ps();
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             if (gmx_mm_any_lt(rsq00,rcutoff2))
 259             {
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 263                                          vdwparam+vdwioffset0+vdwjidx0B,
 264                                          vdwparam+vdwioffset0+vdwjidx0C,
 265                                          vdwparam+vdwioffset0+vdwjidx0D,
 266                                          &c6_00,&c12_00);
 267
 268             /* LENNARD-JONES DISPERSION/REPULSION */
 269
 270             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 271             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 272             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 273             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) ,
 274                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 275             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 276
 277             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 281             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 282
 283             fscal            = fvdw;
 284
 285             fscal            = _mm_and_ps(fscal,cutoff_mask);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_ps(fscal,dx00);
 289             ty               = _mm_mul_ps(fscal,dy00);
 290             tz               = _mm_mul_ps(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_add_ps(fix0,tx);
 294             fiy0             = _mm_add_ps(fiy0,ty);
 295             fiz0             = _mm_add_ps(fiz0,tz);
 296
 297             fjx0             = _mm_add_ps(fjx0,tx);
 298             fjy0             = _mm_add_ps(fjy0,ty);
 299             fjz0             = _mm_add_ps(fjz0,tz);
 300
 301             }
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm_any_lt(rsq10,rcutoff2))
 308             {
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm_mul_ps(iq1,jq0);
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 315             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 316
 317             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm_and_ps(velec,cutoff_mask);
 321             velecsum         = _mm_add_ps(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm_and_ps(fscal,cutoff_mask);
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm_mul_ps(fscal,dx10);
 329             ty               = _mm_mul_ps(fscal,dy10);
 330             tz               = _mm_mul_ps(fscal,dz10);
 331
 332             /* Update vectorial force */
 333             fix1             = _mm_add_ps(fix1,tx);
 334             fiy1             = _mm_add_ps(fiy1,ty);
 335             fiz1             = _mm_add_ps(fiz1,tz);
 336
 337             fjx0             = _mm_add_ps(fjx0,tx);
 338             fjy0             = _mm_add_ps(fjy0,ty);
 339             fjz0             = _mm_add_ps(fjz0,tz);
 340
 341             }
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             if (gmx_mm_any_lt(rsq20,rcutoff2))
 348             {
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm_mul_ps(iq2,jq0);
 352
 353             /* REACTION-FIELD ELECTROSTATICS */
 354             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 355             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 356
 357             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velec            = _mm_and_ps(velec,cutoff_mask);
 361             velecsum         = _mm_add_ps(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             fscal            = _mm_and_ps(fscal,cutoff_mask);
 366
 367             /* Calculate temporary vectorial force */
 368             tx               = _mm_mul_ps(fscal,dx20);
 369             ty               = _mm_mul_ps(fscal,dy20);
 370             tz               = _mm_mul_ps(fscal,dz20);
 371
 372             /* Update vectorial force */
 373             fix2             = _mm_add_ps(fix2,tx);
 374             fiy2             = _mm_add_ps(fiy2,ty);
 375             fiz2             = _mm_add_ps(fiz2,tz);
 376
 377             fjx0             = _mm_add_ps(fjx0,tx);
 378             fjy0             = _mm_add_ps(fjy0,ty);
 379             fjz0             = _mm_add_ps(fjz0,tz);
 380
 381             }
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             if (gmx_mm_any_lt(rsq30,rcutoff2))
 388             {
 389
 390             /* Compute parameters for interactions between i and j atoms */
 391             qq30             = _mm_mul_ps(iq3,jq0);
 392
 393             /* REACTION-FIELD ELECTROSTATICS */
 394             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 395             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 396
 397             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             velec            = _mm_and_ps(velec,cutoff_mask);
 401             velecsum         = _mm_add_ps(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             fscal            = _mm_and_ps(fscal,cutoff_mask);
 406
 407             /* Calculate temporary vectorial force */
 408             tx               = _mm_mul_ps(fscal,dx30);
 409             ty               = _mm_mul_ps(fscal,dy30);
 410             tz               = _mm_mul_ps(fscal,dz30);
 411
 412             /* Update vectorial force */
 413             fix3             = _mm_add_ps(fix3,tx);
 414             fiy3             = _mm_add_ps(fiy3,ty);
 415             fiz3             = _mm_add_ps(fiz3,tz);
 416
 417             fjx0             = _mm_add_ps(fjx0,tx);
 418             fjy0             = _mm_add_ps(fjy0,ty);
 419             fjz0             = _mm_add_ps(fjz0,tz);
 420
 421             }
 422
 423             fjptrA             = f+j_coord_offsetA;
 424             fjptrB             = f+j_coord_offsetB;
 425             fjptrC             = f+j_coord_offsetC;
 426             fjptrD             = f+j_coord_offsetD;
 427
 428             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 429
 430             /* Inner loop uses 149 flops */
 431         }
 432
 433         if(jidx<j_index_end)
 434         {
 435
 436             /* Get j neighbor index, and coordinate index */
 437             jnrlistA         = jjnr[jidx];
 438             jnrlistB         = jjnr[jidx+1];
 439             jnrlistC         = jjnr[jidx+2];
 440             jnrlistD         = jjnr[jidx+3];
 441             /* Sign of each element will be negative for non-real atoms.
 442              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 443              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 444              */
 445             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 446             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 447             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 448             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 449             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 450             j_coord_offsetA  = DIM*jnrA;
 451             j_coord_offsetB  = DIM*jnrB;
 452             j_coord_offsetC  = DIM*jnrC;
 453             j_coord_offsetD  = DIM*jnrD;
 454
 455             /* load j atom coordinates */
 456             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 457                                               x+j_coord_offsetC,x+j_coord_offsetD,
 458                                               &jx0,&jy0,&jz0);
 459
 460             /* Calculate displacement vector */
 461             dx00             = _mm_sub_ps(ix0,jx0);
 462             dy00             = _mm_sub_ps(iy0,jy0);
 463             dz00             = _mm_sub_ps(iz0,jz0);
 464             dx10             = _mm_sub_ps(ix1,jx0);
 465             dy10             = _mm_sub_ps(iy1,jy0);
 466             dz10             = _mm_sub_ps(iz1,jz0);
 467             dx20             = _mm_sub_ps(ix2,jx0);
 468             dy20             = _mm_sub_ps(iy2,jy0);
 469             dz20             = _mm_sub_ps(iz2,jz0);
 470             dx30             = _mm_sub_ps(ix3,jx0);
 471             dy30             = _mm_sub_ps(iy3,jy0);
 472             dz30             = _mm_sub_ps(iz3,jz0);
 473
 474             /* Calculate squared distance and things based on it */
 475             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 476             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 477             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 478             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 479
 480             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 481             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 482             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 483
 484             rinvsq00         = gmx_mm_inv_ps(rsq00);
 485             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 486             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 487             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 491                                                               charge+jnrC+0,charge+jnrD+0);
 492             vdwjidx0A        = 2*vdwtype[jnrA+0];
 493             vdwjidx0B        = 2*vdwtype[jnrB+0];
 494             vdwjidx0C        = 2*vdwtype[jnrC+0];
 495             vdwjidx0D        = 2*vdwtype[jnrD+0];
 496
 497             fjx0             = _mm_setzero_ps();
 498             fjy0             = _mm_setzero_ps();
 499             fjz0             = _mm_setzero_ps();
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             if (gmx_mm_any_lt(rsq00,rcutoff2))
 506             {
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 510                                          vdwparam+vdwioffset0+vdwjidx0B,
 511                                          vdwparam+vdwioffset0+vdwjidx0C,
 512                                          vdwparam+vdwioffset0+vdwjidx0D,
 513                                          &c6_00,&c12_00);
 514
 515             /* LENNARD-JONES DISPERSION/REPULSION */
 516
 517             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 518             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 519             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 520             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) ,
 521                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 522             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 523
 524             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 525
 526             /* Update potential sum for this i atom from the interaction with this j atom. */
 527             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 528             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 529             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 530
 531             fscal            = fvdw;
 532
 533             fscal            = _mm_and_ps(fscal,cutoff_mask);
 534
 535             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm_mul_ps(fscal,dx00);
 539             ty               = _mm_mul_ps(fscal,dy00);
 540             tz               = _mm_mul_ps(fscal,dz00);
 541
 542             /* Update vectorial force */
 543             fix0             = _mm_add_ps(fix0,tx);
 544             fiy0             = _mm_add_ps(fiy0,ty);
 545             fiz0             = _mm_add_ps(fiz0,tz);
 546
 547             fjx0             = _mm_add_ps(fjx0,tx);
 548             fjy0             = _mm_add_ps(fjy0,ty);
 549             fjz0             = _mm_add_ps(fjz0,tz);
 550
 551             }
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_mm_any_lt(rsq10,rcutoff2))
 558             {
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             qq10             = _mm_mul_ps(iq1,jq0);
 562
 563             /* REACTION-FIELD ELECTROSTATICS */
 564             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 565             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 566
 567             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_and_ps(velec,cutoff_mask);
 571             velec            = _mm_andnot_ps(dummy_mask,velec);
 572             velecsum         = _mm_add_ps(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_and_ps(fscal,cutoff_mask);
 577
 578             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm_mul_ps(fscal,dx10);
 582             ty               = _mm_mul_ps(fscal,dy10);
 583             tz               = _mm_mul_ps(fscal,dz10);
 584
 585             /* Update vectorial force */
 586             fix1             = _mm_add_ps(fix1,tx);
 587             fiy1             = _mm_add_ps(fiy1,ty);
 588             fiz1             = _mm_add_ps(fiz1,tz);
 589
 590             fjx0             = _mm_add_ps(fjx0,tx);
 591             fjy0             = _mm_add_ps(fjy0,ty);
 592             fjz0             = _mm_add_ps(fjz0,tz);
 593
 594             }
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             if (gmx_mm_any_lt(rsq20,rcutoff2))
 601             {
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq20             = _mm_mul_ps(iq2,jq0);
 605
 606             /* REACTION-FIELD ELECTROSTATICS */
 607             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 608             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 609
 610             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velec            = _mm_and_ps(velec,cutoff_mask);
 614             velec            = _mm_andnot_ps(dummy_mask,velec);
 615             velecsum         = _mm_add_ps(velecsum,velec);
 616
 617             fscal            = felec;
 618
 619             fscal            = _mm_and_ps(fscal,cutoff_mask);
 620
 621             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm_mul_ps(fscal,dx20);
 625             ty               = _mm_mul_ps(fscal,dy20);
 626             tz               = _mm_mul_ps(fscal,dz20);
 627
 628             /* Update vectorial force */
 629             fix2             = _mm_add_ps(fix2,tx);
 630             fiy2             = _mm_add_ps(fiy2,ty);
 631             fiz2             = _mm_add_ps(fiz2,tz);
 632
 633             fjx0             = _mm_add_ps(fjx0,tx);
 634             fjy0             = _mm_add_ps(fjy0,ty);
 635             fjz0             = _mm_add_ps(fjz0,tz);
 636
 637             }
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             if (gmx_mm_any_lt(rsq30,rcutoff2))
 644             {
 645
 646             /* Compute parameters for interactions between i and j atoms */
 647             qq30             = _mm_mul_ps(iq3,jq0);
 648
 649             /* REACTION-FIELD ELECTROSTATICS */
 650             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 651             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 652
 653             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velec            = _mm_and_ps(velec,cutoff_mask);
 657             velec            = _mm_andnot_ps(dummy_mask,velec);
 658             velecsum         = _mm_add_ps(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm_and_ps(fscal,cutoff_mask);
 663
 664             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 665
 666             /* Calculate temporary vectorial force */
 667             tx               = _mm_mul_ps(fscal,dx30);
 668             ty               = _mm_mul_ps(fscal,dy30);
 669             tz               = _mm_mul_ps(fscal,dz30);
 670
 671             /* Update vectorial force */
 672             fix3             = _mm_add_ps(fix3,tx);
 673             fiy3             = _mm_add_ps(fiy3,ty);
 674             fiz3             = _mm_add_ps(fiz3,tz);
 675
 676             fjx0             = _mm_add_ps(fjx0,tx);
 677             fjy0             = _mm_add_ps(fjy0,ty);
 678             fjz0             = _mm_add_ps(fjz0,tz);
 679
 680             }
 681
 682             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 683             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 684             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 685             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 686
 687             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 688
 689             /* Inner loop uses 149 flops */
 690         }
 691
 692         /* End of innermost loop */
 693
 694         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 695                                               f+i_coord_offset,fshift+i_shift_offset);
 696
 697         ggid                        = gid[iidx];
 698         /* Update potential energies */
 699         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 700         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 701
 702         /* Increment number of inner iterations */
 703         inneriter                  += j_index_end - j_index_start;
 704
 705         /* Outer loop uses 26 flops */
 706     }
 707
 708     /* Increment number of outer iterations */
 709     outeriter        += nri;
 710
 711     /* Update outer/inner flops */
 712
 713     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*149);
 714 }
 715 /*
 716  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse4_1_single
 717  * Electrostatics interaction: ReactionField
 718  * VdW interaction:            LennardJones
 719  * Geometry:                   Water4-Particle
 720  * Calculate force/pot:        Force
 721  */
 722 void
 723 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse4_1_single
 724                     (t_nblist                    * gmx_restrict       nlist,
 725                      rvec                        * gmx_restrict          xx,
 726                      rvec                        * gmx_restrict          ff,
 727                      t_forcerec                  * gmx_restrict          fr,
 728                      t_mdatoms                   * gmx_restrict     mdatoms,
 729                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 730                      t_nrnb                      * gmx_restrict        nrnb)
 731 {
 732     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 733      * just 0 for non-waters.
 734      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 735      * jnr indices corresponding to data put in the four positions in the SIMD register.
 736      */
 737     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 738     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 739     int              jnrA,jnrB,jnrC,jnrD;
 740     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 741     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 742     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 743     real             rcutoff_scalar;
 744     real             *shiftvec,*fshift,*x,*f;
 745     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 746     real             scratch[4*DIM];
 747     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 748     int              vdwioffset0;
 749     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 750     int              vdwioffset1;
 751     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 752     int              vdwioffset2;
 753     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 754     int              vdwioffset3;
 755     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 756     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 757     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 758     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 759     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 760     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 761     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 762     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 763     real             *charge;
 764     int              nvdwtype;
 765     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 766     int              *vdwtype;
 767     real             *vdwparam;
 768     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 769     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 770     __m128           dummy_mask,cutoff_mask;
 771     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 772     __m128           one     = _mm_set1_ps(1.0);
 773     __m128           two     = _mm_set1_ps(2.0);
 774     x                = xx[0];
 775     f                = ff[0];
 776
 777     nri              = nlist->nri;
 778     iinr             = nlist->iinr;
 779     jindex           = nlist->jindex;
 780     jjnr             = nlist->jjnr;
 781     shiftidx         = nlist->shift;
 782     gid              = nlist->gid;
 783     shiftvec         = fr->shift_vec[0];
 784     fshift           = fr->fshift[0];
 785     facel            = _mm_set1_ps(fr->epsfac);
 786     charge           = mdatoms->chargeA;
 787     krf              = _mm_set1_ps(fr->ic->k_rf);
 788     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 789     crf              = _mm_set1_ps(fr->ic->c_rf);
 790     nvdwtype         = fr->ntype;
 791     vdwparam         = fr->nbfp;
 792     vdwtype          = mdatoms->typeA;
 793
 794     /* Setup water-specific parameters */
 795     inr              = nlist->iinr[0];
 796     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 797     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 798     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 799     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 800
 801     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 802     rcutoff_scalar   = fr->rcoulomb;
 803     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 804     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 805
 806     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 807     rvdw             = _mm_set1_ps(fr->rvdw);
 808
 809     /* Avoid stupid compiler warnings */
 810     jnrA = jnrB = jnrC = jnrD = 0;
 811     j_coord_offsetA = 0;
 812     j_coord_offsetB = 0;
 813     j_coord_offsetC = 0;
 814     j_coord_offsetD = 0;
 815
 816     outeriter        = 0;
 817     inneriter        = 0;
 818
 819     for(iidx=0;iidx<4*DIM;iidx++)
 820     {
 821         scratch[iidx] = 0.0;
 822     }
 823
 824     /* Start outer loop over neighborlists */
 825     for(iidx=0; iidx<nri; iidx++)
 826     {
 827         /* Load shift vector for this list */
 828         i_shift_offset   = DIM*shiftidx[iidx];
 829
 830         /* Load limits for loop over neighbors */
 831         j_index_start    = jindex[iidx];
 832         j_index_end      = jindex[iidx+1];
 833
 834         /* Get outer coordinate index */
 835         inr              = iinr[iidx];
 836         i_coord_offset   = DIM*inr;
 837
 838         /* Load i particle coords and add shift vector */
 839         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 840                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 841
 842         fix0             = _mm_setzero_ps();
 843         fiy0             = _mm_setzero_ps();
 844         fiz0             = _mm_setzero_ps();
 845         fix1             = _mm_setzero_ps();
 846         fiy1             = _mm_setzero_ps();
 847         fiz1             = _mm_setzero_ps();
 848         fix2             = _mm_setzero_ps();
 849         fiy2             = _mm_setzero_ps();
 850         fiz2             = _mm_setzero_ps();
 851         fix3             = _mm_setzero_ps();
 852         fiy3             = _mm_setzero_ps();
 853         fiz3             = _mm_setzero_ps();
 854
 855         /* Start inner kernel loop */
 856         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 857         {
 858
 859             /* Get j neighbor index, and coordinate index */
 860             jnrA             = jjnr[jidx];
 861             jnrB             = jjnr[jidx+1];
 862             jnrC             = jjnr[jidx+2];
 863             jnrD             = jjnr[jidx+3];
 864             j_coord_offsetA  = DIM*jnrA;
 865             j_coord_offsetB  = DIM*jnrB;
 866             j_coord_offsetC  = DIM*jnrC;
 867             j_coord_offsetD  = DIM*jnrD;
 868
 869             /* load j atom coordinates */
 870             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 871                                               x+j_coord_offsetC,x+j_coord_offsetD,
 872                                               &jx0,&jy0,&jz0);
 873
 874             /* Calculate displacement vector */
 875             dx00             = _mm_sub_ps(ix0,jx0);
 876             dy00             = _mm_sub_ps(iy0,jy0);
 877             dz00             = _mm_sub_ps(iz0,jz0);
 878             dx10             = _mm_sub_ps(ix1,jx0);
 879             dy10             = _mm_sub_ps(iy1,jy0);
 880             dz10             = _mm_sub_ps(iz1,jz0);
 881             dx20             = _mm_sub_ps(ix2,jx0);
 882             dy20             = _mm_sub_ps(iy2,jy0);
 883             dz20             = _mm_sub_ps(iz2,jz0);
 884             dx30             = _mm_sub_ps(ix3,jx0);
 885             dy30             = _mm_sub_ps(iy3,jy0);
 886             dz30             = _mm_sub_ps(iz3,jz0);
 887
 888             /* Calculate squared distance and things based on it */
 889             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 890             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 891             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 892             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 893
 894             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 895             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 896             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 897
 898             rinvsq00         = gmx_mm_inv_ps(rsq00);
 899             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 900             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 901             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 902
 903             /* Load parameters for j particles */
 904             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 905                                                               charge+jnrC+0,charge+jnrD+0);
 906             vdwjidx0A        = 2*vdwtype[jnrA+0];
 907             vdwjidx0B        = 2*vdwtype[jnrB+0];
 908             vdwjidx0C        = 2*vdwtype[jnrC+0];
 909             vdwjidx0D        = 2*vdwtype[jnrD+0];
 910
 911             fjx0             = _mm_setzero_ps();
 912             fjy0             = _mm_setzero_ps();
 913             fjz0             = _mm_setzero_ps();
 914
 915             /**************************
 916              * CALCULATE INTERACTIONS *
 917              **************************/
 918
 919             if (gmx_mm_any_lt(rsq00,rcutoff2))
 920             {
 921
 922             /* Compute parameters for interactions between i and j atoms */
 923             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 924                                          vdwparam+vdwioffset0+vdwjidx0B,
 925                                          vdwparam+vdwioffset0+vdwjidx0C,
 926                                          vdwparam+vdwioffset0+vdwjidx0D,
 927                                          &c6_00,&c12_00);
 928
 929             /* LENNARD-JONES DISPERSION/REPULSION */
 930
 931             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 932             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 933
 934             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 935
 936             fscal            = fvdw;
 937
 938             fscal            = _mm_and_ps(fscal,cutoff_mask);
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm_mul_ps(fscal,dx00);
 942             ty               = _mm_mul_ps(fscal,dy00);
 943             tz               = _mm_mul_ps(fscal,dz00);
 944
 945             /* Update vectorial force */
 946             fix0             = _mm_add_ps(fix0,tx);
 947             fiy0             = _mm_add_ps(fiy0,ty);
 948             fiz0             = _mm_add_ps(fiz0,tz);
 949
 950             fjx0             = _mm_add_ps(fjx0,tx);
 951             fjy0             = _mm_add_ps(fjy0,ty);
 952             fjz0             = _mm_add_ps(fjz0,tz);
 953
 954             }
 955
 956             /**************************
 957              * CALCULATE INTERACTIONS *
 958              **************************/
 959
 960             if (gmx_mm_any_lt(rsq10,rcutoff2))
 961             {
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq10             = _mm_mul_ps(iq1,jq0);
 965
 966             /* REACTION-FIELD ELECTROSTATICS */
 967             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 968
 969             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 970
 971             fscal            = felec;
 972
 973             fscal            = _mm_and_ps(fscal,cutoff_mask);
 974
 975             /* Calculate temporary vectorial force */
 976             tx               = _mm_mul_ps(fscal,dx10);
 977             ty               = _mm_mul_ps(fscal,dy10);
 978             tz               = _mm_mul_ps(fscal,dz10);
 979
 980             /* Update vectorial force */
 981             fix1             = _mm_add_ps(fix1,tx);
 982             fiy1             = _mm_add_ps(fiy1,ty);
 983             fiz1             = _mm_add_ps(fiz1,tz);
 984
 985             fjx0             = _mm_add_ps(fjx0,tx);
 986             fjy0             = _mm_add_ps(fjy0,ty);
 987             fjz0             = _mm_add_ps(fjz0,tz);
 988
 989             }
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             if (gmx_mm_any_lt(rsq20,rcutoff2))
 996             {
 997
 998             /* Compute parameters for interactions between i and j atoms */
 999             qq20             = _mm_mul_ps(iq2,jq0);
1000
1001             /* REACTION-FIELD ELECTROSTATICS */
1002             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1003
1004             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1005
1006             fscal            = felec;
1007
1008             fscal            = _mm_and_ps(fscal,cutoff_mask);
1009
1010             /* Calculate temporary vectorial force */
1011             tx               = _mm_mul_ps(fscal,dx20);
1012             ty               = _mm_mul_ps(fscal,dy20);
1013             tz               = _mm_mul_ps(fscal,dz20);
1014
1015             /* Update vectorial force */
1016             fix2             = _mm_add_ps(fix2,tx);
1017             fiy2             = _mm_add_ps(fiy2,ty);
1018             fiz2             = _mm_add_ps(fiz2,tz);
1019
1020             fjx0             = _mm_add_ps(fjx0,tx);
1021             fjy0             = _mm_add_ps(fjy0,ty);
1022             fjz0             = _mm_add_ps(fjz0,tz);
1023
1024             }
1025
1026             /**************************
1027              * CALCULATE INTERACTIONS *
1028              **************************/
1029
1030             if (gmx_mm_any_lt(rsq30,rcutoff2))
1031             {
1032
1033             /* Compute parameters for interactions between i and j atoms */
1034             qq30             = _mm_mul_ps(iq3,jq0);
1035
1036             /* REACTION-FIELD ELECTROSTATICS */
1037             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1038
1039             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1040
1041             fscal            = felec;
1042
1043             fscal            = _mm_and_ps(fscal,cutoff_mask);
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm_mul_ps(fscal,dx30);
1047             ty               = _mm_mul_ps(fscal,dy30);
1048             tz               = _mm_mul_ps(fscal,dz30);
1049
1050             /* Update vectorial force */
1051             fix3             = _mm_add_ps(fix3,tx);
1052             fiy3             = _mm_add_ps(fiy3,ty);
1053             fiz3             = _mm_add_ps(fiz3,tz);
1054
1055             fjx0             = _mm_add_ps(fjx0,tx);
1056             fjy0             = _mm_add_ps(fjy0,ty);
1057             fjz0             = _mm_add_ps(fjz0,tz);
1058
1059             }
1060
1061             fjptrA             = f+j_coord_offsetA;
1062             fjptrB             = f+j_coord_offsetB;
1063             fjptrC             = f+j_coord_offsetC;
1064             fjptrD             = f+j_coord_offsetD;
1065
1066             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1067
1068             /* Inner loop uses 120 flops */
1069         }
1070
1071         if(jidx<j_index_end)
1072         {
1073
1074             /* Get j neighbor index, and coordinate index */
1075             jnrlistA         = jjnr[jidx];
1076             jnrlistB         = jjnr[jidx+1];
1077             jnrlistC         = jjnr[jidx+2];
1078             jnrlistD         = jjnr[jidx+3];
1079             /* Sign of each element will be negative for non-real atoms.
1080              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1081              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1082              */
1083             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1084             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1085             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1086             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1087             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1088             j_coord_offsetA  = DIM*jnrA;
1089             j_coord_offsetB  = DIM*jnrB;
1090             j_coord_offsetC  = DIM*jnrC;
1091             j_coord_offsetD  = DIM*jnrD;
1092
1093             /* load j atom coordinates */
1094             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1095                                               x+j_coord_offsetC,x+j_coord_offsetD,
1096                                               &jx0,&jy0,&jz0);
1097
1098             /* Calculate displacement vector */
1099             dx00             = _mm_sub_ps(ix0,jx0);
1100             dy00             = _mm_sub_ps(iy0,jy0);
1101             dz00             = _mm_sub_ps(iz0,jz0);
1102             dx10             = _mm_sub_ps(ix1,jx0);
1103             dy10             = _mm_sub_ps(iy1,jy0);
1104             dz10             = _mm_sub_ps(iz1,jz0);
1105             dx20             = _mm_sub_ps(ix2,jx0);
1106             dy20             = _mm_sub_ps(iy2,jy0);
1107             dz20             = _mm_sub_ps(iz2,jz0);
1108             dx30             = _mm_sub_ps(ix3,jx0);
1109             dy30             = _mm_sub_ps(iy3,jy0);
1110             dz30             = _mm_sub_ps(iz3,jz0);
1111
1112             /* Calculate squared distance and things based on it */
1113             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1114             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1115             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1116             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1117
1118             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1119             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1120             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1121
1122             rinvsq00         = gmx_mm_inv_ps(rsq00);
1123             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1124             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1125             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1126
1127             /* Load parameters for j particles */
1128             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1129                                                               charge+jnrC+0,charge+jnrD+0);
1130             vdwjidx0A        = 2*vdwtype[jnrA+0];
1131             vdwjidx0B        = 2*vdwtype[jnrB+0];
1132             vdwjidx0C        = 2*vdwtype[jnrC+0];
1133             vdwjidx0D        = 2*vdwtype[jnrD+0];
1134
1135             fjx0             = _mm_setzero_ps();
1136             fjy0             = _mm_setzero_ps();
1137             fjz0             = _mm_setzero_ps();
1138
1139             /**************************
1140              * CALCULATE INTERACTIONS *
1141              **************************/
1142
1143             if (gmx_mm_any_lt(rsq00,rcutoff2))
1144             {
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1148                                          vdwparam+vdwioffset0+vdwjidx0B,
1149                                          vdwparam+vdwioffset0+vdwjidx0C,
1150                                          vdwparam+vdwioffset0+vdwjidx0D,
1151                                          &c6_00,&c12_00);
1152
1153             /* LENNARD-JONES DISPERSION/REPULSION */
1154
1155             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1156             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1157
1158             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1159
1160             fscal            = fvdw;
1161
1162             fscal            = _mm_and_ps(fscal,cutoff_mask);
1163
1164             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1165
1166             /* Calculate temporary vectorial force */
1167             tx               = _mm_mul_ps(fscal,dx00);
1168             ty               = _mm_mul_ps(fscal,dy00);
1169             tz               = _mm_mul_ps(fscal,dz00);
1170
1171             /* Update vectorial force */
1172             fix0             = _mm_add_ps(fix0,tx);
1173             fiy0             = _mm_add_ps(fiy0,ty);
1174             fiz0             = _mm_add_ps(fiz0,tz);
1175
1176             fjx0             = _mm_add_ps(fjx0,tx);
1177             fjy0             = _mm_add_ps(fjy0,ty);
1178             fjz0             = _mm_add_ps(fjz0,tz);
1179
1180             }
1181
1182             /**************************
1183              * CALCULATE INTERACTIONS *
1184              **************************/
1185
1186             if (gmx_mm_any_lt(rsq10,rcutoff2))
1187             {
1188
1189             /* Compute parameters for interactions between i and j atoms */
1190             qq10             = _mm_mul_ps(iq1,jq0);
1191
1192             /* REACTION-FIELD ELECTROSTATICS */
1193             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1194
1195             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm_and_ps(fscal,cutoff_mask);
1200
1201             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1202
1203             /* Calculate temporary vectorial force */
1204             tx               = _mm_mul_ps(fscal,dx10);
1205             ty               = _mm_mul_ps(fscal,dy10);
1206             tz               = _mm_mul_ps(fscal,dz10);
1207
1208             /* Update vectorial force */
1209             fix1             = _mm_add_ps(fix1,tx);
1210             fiy1             = _mm_add_ps(fiy1,ty);
1211             fiz1             = _mm_add_ps(fiz1,tz);
1212
1213             fjx0             = _mm_add_ps(fjx0,tx);
1214             fjy0             = _mm_add_ps(fjy0,ty);
1215             fjz0             = _mm_add_ps(fjz0,tz);
1216
1217             }
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             if (gmx_mm_any_lt(rsq20,rcutoff2))
1224             {
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq20             = _mm_mul_ps(iq2,jq0);
1228
1229             /* REACTION-FIELD ELECTROSTATICS */
1230             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1231
1232             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1233
1234             fscal            = felec;
1235
1236             fscal            = _mm_and_ps(fscal,cutoff_mask);
1237
1238             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1239
1240             /* Calculate temporary vectorial force */
1241             tx               = _mm_mul_ps(fscal,dx20);
1242             ty               = _mm_mul_ps(fscal,dy20);
1243             tz               = _mm_mul_ps(fscal,dz20);
1244
1245             /* Update vectorial force */
1246             fix2             = _mm_add_ps(fix2,tx);
1247             fiy2             = _mm_add_ps(fiy2,ty);
1248             fiz2             = _mm_add_ps(fiz2,tz);
1249
1250             fjx0             = _mm_add_ps(fjx0,tx);
1251             fjy0             = _mm_add_ps(fjy0,ty);
1252             fjz0             = _mm_add_ps(fjz0,tz);
1253
1254             }
1255
1256             /**************************
1257              * CALCULATE INTERACTIONS *
1258              **************************/
1259
1260             if (gmx_mm_any_lt(rsq30,rcutoff2))
1261             {
1262
1263             /* Compute parameters for interactions between i and j atoms */
1264             qq30             = _mm_mul_ps(iq3,jq0);
1265
1266             /* REACTION-FIELD ELECTROSTATICS */
1267             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1268
1269             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1270
1271             fscal            = felec;
1272
1273             fscal            = _mm_and_ps(fscal,cutoff_mask);
1274
1275             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1276
1277             /* Calculate temporary vectorial force */
1278             tx               = _mm_mul_ps(fscal,dx30);
1279             ty               = _mm_mul_ps(fscal,dy30);
1280             tz               = _mm_mul_ps(fscal,dz30);
1281
1282             /* Update vectorial force */
1283             fix3             = _mm_add_ps(fix3,tx);
1284             fiy3             = _mm_add_ps(fiy3,ty);
1285             fiz3             = _mm_add_ps(fiz3,tz);
1286
1287             fjx0             = _mm_add_ps(fjx0,tx);
1288             fjy0             = _mm_add_ps(fjy0,ty);
1289             fjz0             = _mm_add_ps(fjz0,tz);
1290
1291             }
1292
1293             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1294             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1295             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1296             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1297
1298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1299
1300             /* Inner loop uses 120 flops */
1301         }
1302
1303         /* End of innermost loop */
1304
1305         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1306                                               f+i_coord_offset,fshift+i_shift_offset);
1307
1308         /* Increment number of inner iterations */
1309         inneriter                  += j_index_end - j_index_start;
1310
1311         /* Outer loop uses 24 flops */
1312     }
1313
1314     /* Increment number of outer iterations */
1315     outeriter        += nri;
1316
1317     /* Update outer/inner flops */
1318
1319     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*120);
1320 }