src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJ_GeomW3P1_sse2_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse2_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128i          ewitab;
 102     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     real             *ewtab;
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->ic->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 128     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_ps();
 171         fiy0             = _mm_setzero_ps();
 172         fiz0             = _mm_setzero_ps();
 173         fix1             = _mm_setzero_ps();
 174         fiy1             = _mm_setzero_ps();
 175         fiz1             = _mm_setzero_ps();
 176         fix2             = _mm_setzero_ps();
 177         fiy2             = _mm_setzero_ps();
 178         fiz2             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 216             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 217             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 218
 219             rinv00           = sse2_invsqrt_f(rsq00);
 220             rinv10           = sse2_invsqrt_f(rsq10);
 221             rinv20           = sse2_invsqrt_f(rsq20);
 222
 223             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 224             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 225             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                               charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             fjx0             = _mm_setzero_ps();
 236             fjy0             = _mm_setzero_ps();
 237             fjz0             = _mm_setzero_ps();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm_mul_ps(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm_mul_ps(iq0,jq0);
 247             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,
 249                                          vdwparam+vdwioffset0+vdwjidx0C,
 250                                          vdwparam+vdwioffset0+vdwjidx0D,
 251                                          &c6_00,&c12_00);
 252
 253             /* EWALD ELECTROSTATICS */
 254
 255             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 256             ewrt             = _mm_mul_ps(r00,ewtabscale);
 257             ewitab           = _mm_cvttps_epi32(ewrt);
 258             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 259             ewitab           = _mm_slli_epi32(ewitab,2);
 260             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 261             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 262             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 263             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 264             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 265             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 266             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 267             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 268             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 269
 270             /* LENNARD-JONES DISPERSION/REPULSION */
 271
 272             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 273             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 274             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 275             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 276             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velecsum         = _mm_add_ps(velecsum,velec);
 280             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 281
 282             fscal            = _mm_add_ps(felec,fvdw);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_ps(fscal,dx00);
 286             ty               = _mm_mul_ps(fscal,dy00);
 287             tz               = _mm_mul_ps(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_add_ps(fix0,tx);
 291             fiy0             = _mm_add_ps(fiy0,ty);
 292             fiz0             = _mm_add_ps(fiz0,tz);
 293
 294             fjx0             = _mm_add_ps(fjx0,tx);
 295             fjy0             = _mm_add_ps(fjy0,ty);
 296             fjz0             = _mm_add_ps(fjz0,tz);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r10              = _mm_mul_ps(rsq10,rinv10);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_ps(iq1,jq0);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 310             ewrt             = _mm_mul_ps(r10,ewtabscale);
 311             ewitab           = _mm_cvttps_epi32(ewrt);
 312             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 313             ewitab           = _mm_slli_epi32(ewitab,2);
 314             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 315             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 316             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 317             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 318             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 319             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 320             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 321             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 322             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velecsum         = _mm_add_ps(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm_mul_ps(fscal,dx10);
 331             ty               = _mm_mul_ps(fscal,dy10);
 332             tz               = _mm_mul_ps(fscal,dz10);
 333
 334             /* Update vectorial force */
 335             fix1             = _mm_add_ps(fix1,tx);
 336             fiy1             = _mm_add_ps(fiy1,ty);
 337             fiz1             = _mm_add_ps(fiz1,tz);
 338
 339             fjx0             = _mm_add_ps(fjx0,tx);
 340             fjy0             = _mm_add_ps(fjy0,ty);
 341             fjz0             = _mm_add_ps(fjz0,tz);
 342
 343             /**************************
 344              * CALCULATE INTERACTIONS *
 345              **************************/
 346
 347             r20              = _mm_mul_ps(rsq20,rinv20);
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq20             = _mm_mul_ps(iq2,jq0);
 351
 352             /* EWALD ELECTROSTATICS */
 353
 354             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 355             ewrt             = _mm_mul_ps(r20,ewtabscale);
 356             ewitab           = _mm_cvttps_epi32(ewrt);
 357             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 358             ewitab           = _mm_slli_epi32(ewitab,2);
 359             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 360             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 361             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 362             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 363             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 364             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 365             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 366             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 367             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velecsum         = _mm_add_ps(velecsum,velec);
 371
 372             fscal            = felec;
 373
 374             /* Calculate temporary vectorial force */
 375             tx               = _mm_mul_ps(fscal,dx20);
 376             ty               = _mm_mul_ps(fscal,dy20);
 377             tz               = _mm_mul_ps(fscal,dz20);
 378
 379             /* Update vectorial force */
 380             fix2             = _mm_add_ps(fix2,tx);
 381             fiy2             = _mm_add_ps(fiy2,ty);
 382             fiz2             = _mm_add_ps(fiz2,tz);
 383
 384             fjx0             = _mm_add_ps(fjx0,tx);
 385             fjy0             = _mm_add_ps(fjy0,ty);
 386             fjz0             = _mm_add_ps(fjz0,tz);
 387
 388             fjptrA             = f+j_coord_offsetA;
 389             fjptrB             = f+j_coord_offsetB;
 390             fjptrC             = f+j_coord_offsetC;
 391             fjptrD             = f+j_coord_offsetD;
 392
 393             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 394
 395             /* Inner loop uses 135 flops */
 396         }
 397
 398         if(jidx<j_index_end)
 399         {
 400
 401             /* Get j neighbor index, and coordinate index */
 402             jnrlistA         = jjnr[jidx];
 403             jnrlistB         = jjnr[jidx+1];
 404             jnrlistC         = jjnr[jidx+2];
 405             jnrlistD         = jjnr[jidx+3];
 406             /* Sign of each element will be negative for non-real atoms.
 407              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 408              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 409              */
 410             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 411             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 412             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 413             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 414             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 415             j_coord_offsetA  = DIM*jnrA;
 416             j_coord_offsetB  = DIM*jnrB;
 417             j_coord_offsetC  = DIM*jnrC;
 418             j_coord_offsetD  = DIM*jnrD;
 419
 420             /* load j atom coordinates */
 421             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 422                                               x+j_coord_offsetC,x+j_coord_offsetD,
 423                                               &jx0,&jy0,&jz0);
 424
 425             /* Calculate displacement vector */
 426             dx00             = _mm_sub_ps(ix0,jx0);
 427             dy00             = _mm_sub_ps(iy0,jy0);
 428             dz00             = _mm_sub_ps(iz0,jz0);
 429             dx10             = _mm_sub_ps(ix1,jx0);
 430             dy10             = _mm_sub_ps(iy1,jy0);
 431             dz10             = _mm_sub_ps(iz1,jz0);
 432             dx20             = _mm_sub_ps(ix2,jx0);
 433             dy20             = _mm_sub_ps(iy2,jy0);
 434             dz20             = _mm_sub_ps(iz2,jz0);
 435
 436             /* Calculate squared distance and things based on it */
 437             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 438             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 439             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 440
 441             rinv00           = sse2_invsqrt_f(rsq00);
 442             rinv10           = sse2_invsqrt_f(rsq10);
 443             rinv20           = sse2_invsqrt_f(rsq20);
 444
 445             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 446             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 447             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 448
 449             /* Load parameters for j particles */
 450             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 451                                                               charge+jnrC+0,charge+jnrD+0);
 452             vdwjidx0A        = 2*vdwtype[jnrA+0];
 453             vdwjidx0B        = 2*vdwtype[jnrB+0];
 454             vdwjidx0C        = 2*vdwtype[jnrC+0];
 455             vdwjidx0D        = 2*vdwtype[jnrD+0];
 456
 457             fjx0             = _mm_setzero_ps();
 458             fjy0             = _mm_setzero_ps();
 459             fjz0             = _mm_setzero_ps();
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             r00              = _mm_mul_ps(rsq00,rinv00);
 466             r00              = _mm_andnot_ps(dummy_mask,r00);
 467
 468             /* Compute parameters for interactions between i and j atoms */
 469             qq00             = _mm_mul_ps(iq0,jq0);
 470             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 471                                          vdwparam+vdwioffset0+vdwjidx0B,
 472                                          vdwparam+vdwioffset0+vdwjidx0C,
 473                                          vdwparam+vdwioffset0+vdwjidx0D,
 474                                          &c6_00,&c12_00);
 475
 476             /* EWALD ELECTROSTATICS */
 477
 478             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 479             ewrt             = _mm_mul_ps(r00,ewtabscale);
 480             ewitab           = _mm_cvttps_epi32(ewrt);
 481             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 482             ewitab           = _mm_slli_epi32(ewitab,2);
 483             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 484             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 485             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 486             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 487             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 488             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 489             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 490             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 491             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 492
 493             /* LENNARD-JONES DISPERSION/REPULSION */
 494
 495             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 496             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 497             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 498             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 499             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             velec            = _mm_andnot_ps(dummy_mask,velec);
 503             velecsum         = _mm_add_ps(velecsum,velec);
 504             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 505             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 506
 507             fscal            = _mm_add_ps(felec,fvdw);
 508
 509             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm_mul_ps(fscal,dx00);
 513             ty               = _mm_mul_ps(fscal,dy00);
 514             tz               = _mm_mul_ps(fscal,dz00);
 515
 516             /* Update vectorial force */
 517             fix0             = _mm_add_ps(fix0,tx);
 518             fiy0             = _mm_add_ps(fiy0,ty);
 519             fiz0             = _mm_add_ps(fiz0,tz);
 520
 521             fjx0             = _mm_add_ps(fjx0,tx);
 522             fjy0             = _mm_add_ps(fjy0,ty);
 523             fjz0             = _mm_add_ps(fjz0,tz);
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             r10              = _mm_mul_ps(rsq10,rinv10);
 530             r10              = _mm_andnot_ps(dummy_mask,r10);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq10             = _mm_mul_ps(iq1,jq0);
 534
 535             /* EWALD ELECTROSTATICS */
 536
 537             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 538             ewrt             = _mm_mul_ps(r10,ewtabscale);
 539             ewitab           = _mm_cvttps_epi32(ewrt);
 540             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 541             ewitab           = _mm_slli_epi32(ewitab,2);
 542             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 543             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 544             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 545             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 546             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 547             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 548             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 549             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 550             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm_andnot_ps(dummy_mask,velec);
 554             velecsum         = _mm_add_ps(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 559
 560             /* Calculate temporary vectorial force */
 561             tx               = _mm_mul_ps(fscal,dx10);
 562             ty               = _mm_mul_ps(fscal,dy10);
 563             tz               = _mm_mul_ps(fscal,dz10);
 564
 565             /* Update vectorial force */
 566             fix1             = _mm_add_ps(fix1,tx);
 567             fiy1             = _mm_add_ps(fiy1,ty);
 568             fiz1             = _mm_add_ps(fiz1,tz);
 569
 570             fjx0             = _mm_add_ps(fjx0,tx);
 571             fjy0             = _mm_add_ps(fjy0,ty);
 572             fjz0             = _mm_add_ps(fjz0,tz);
 573
 574             /**************************
 575              * CALCULATE INTERACTIONS *
 576              **************************/
 577
 578             r20              = _mm_mul_ps(rsq20,rinv20);
 579             r20              = _mm_andnot_ps(dummy_mask,r20);
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             qq20             = _mm_mul_ps(iq2,jq0);
 583
 584             /* EWALD ELECTROSTATICS */
 585
 586             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 587             ewrt             = _mm_mul_ps(r20,ewtabscale);
 588             ewitab           = _mm_cvttps_epi32(ewrt);
 589             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 590             ewitab           = _mm_slli_epi32(ewitab,2);
 591             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 592             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 593             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 594             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 595             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 596             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 597             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 598             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 599             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 600
 601             /* Update potential sum for this i atom from the interaction with this j atom. */
 602             velec            = _mm_andnot_ps(dummy_mask,velec);
 603             velecsum         = _mm_add_ps(velecsum,velec);
 604
 605             fscal            = felec;
 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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 624             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 625             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 626             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 627
 628             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 629
 630             /* Inner loop uses 138 flops */
 631         }
 632
 633         /* End of innermost loop */
 634
 635         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 636                                               f+i_coord_offset,fshift+i_shift_offset);
 637
 638         ggid                        = gid[iidx];
 639         /* Update potential energies */
 640         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 641         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 642
 643         /* Increment number of inner iterations */
 644         inneriter                  += j_index_end - j_index_start;
 645
 646         /* Outer loop uses 20 flops */
 647     }
 648
 649     /* Increment number of outer iterations */
 650     outeriter        += nri;
 651
 652     /* Update outer/inner flops */
 653
 654     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*138);
 655 }
 656 /*
 657  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse2_single
 658  * Electrostatics interaction: Ewald
 659  * VdW interaction:            LennardJones
 660  * Geometry:                   Water3-Particle
 661  * Calculate force/pot:        Force
 662  */
 663 void
 664 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse2_single
 665                     (t_nblist                    * gmx_restrict       nlist,
 666                      rvec                        * gmx_restrict          xx,
 667                      rvec                        * gmx_restrict          ff,
 668                      struct t_forcerec           * gmx_restrict          fr,
 669                      t_mdatoms                   * gmx_restrict     mdatoms,
 670                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 671                      t_nrnb                      * gmx_restrict        nrnb)
 672 {
 673     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 674      * just 0 for non-waters.
 675      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 676      * jnr indices corresponding to data put in the four positions in the SIMD register.
 677      */
 678     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 679     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 680     int              jnrA,jnrB,jnrC,jnrD;
 681     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 682     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 683     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 684     real             rcutoff_scalar;
 685     real             *shiftvec,*fshift,*x,*f;
 686     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 687     real             scratch[4*DIM];
 688     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 689     int              vdwioffset0;
 690     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 691     int              vdwioffset1;
 692     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 693     int              vdwioffset2;
 694     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 695     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 696     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 697     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 698     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 699     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 700     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 701     real             *charge;
 702     int              nvdwtype;
 703     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 704     int              *vdwtype;
 705     real             *vdwparam;
 706     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 707     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 708     __m128i          ewitab;
 709     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 710     real             *ewtab;
 711     __m128           dummy_mask,cutoff_mask;
 712     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 713     __m128           one     = _mm_set1_ps(1.0);
 714     __m128           two     = _mm_set1_ps(2.0);
 715     x                = xx[0];
 716     f                = ff[0];
 717
 718     nri              = nlist->nri;
 719     iinr             = nlist->iinr;
 720     jindex           = nlist->jindex;
 721     jjnr             = nlist->jjnr;
 722     shiftidx         = nlist->shift;
 723     gid              = nlist->gid;
 724     shiftvec         = fr->shift_vec[0];
 725     fshift           = fr->fshift[0];
 726     facel            = _mm_set1_ps(fr->ic->epsfac);
 727     charge           = mdatoms->chargeA;
 728     nvdwtype         = fr->ntype;
 729     vdwparam         = fr->nbfp;
 730     vdwtype          = mdatoms->typeA;
 731
 732     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 733     ewtab            = fr->ic->tabq_coul_F;
 734     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 735     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 736
 737     /* Setup water-specific parameters */
 738     inr              = nlist->iinr[0];
 739     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 740     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 741     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 742     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 743
 744     /* Avoid stupid compiler warnings */
 745     jnrA = jnrB = jnrC = jnrD = 0;
 746     j_coord_offsetA = 0;
 747     j_coord_offsetB = 0;
 748     j_coord_offsetC = 0;
 749     j_coord_offsetD = 0;
 750
 751     outeriter        = 0;
 752     inneriter        = 0;
 753
 754     for(iidx=0;iidx<4*DIM;iidx++)
 755     {
 756         scratch[iidx] = 0.0;
 757     }
 758
 759     /* Start outer loop over neighborlists */
 760     for(iidx=0; iidx<nri; iidx++)
 761     {
 762         /* Load shift vector for this list */
 763         i_shift_offset   = DIM*shiftidx[iidx];
 764
 765         /* Load limits for loop over neighbors */
 766         j_index_start    = jindex[iidx];
 767         j_index_end      = jindex[iidx+1];
 768
 769         /* Get outer coordinate index */
 770         inr              = iinr[iidx];
 771         i_coord_offset   = DIM*inr;
 772
 773         /* Load i particle coords and add shift vector */
 774         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 775                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 776
 777         fix0             = _mm_setzero_ps();
 778         fiy0             = _mm_setzero_ps();
 779         fiz0             = _mm_setzero_ps();
 780         fix1             = _mm_setzero_ps();
 781         fiy1             = _mm_setzero_ps();
 782         fiz1             = _mm_setzero_ps();
 783         fix2             = _mm_setzero_ps();
 784         fiy2             = _mm_setzero_ps();
 785         fiz2             = _mm_setzero_ps();
 786
 787         /* Start inner kernel loop */
 788         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 789         {
 790
 791             /* Get j neighbor index, and coordinate index */
 792             jnrA             = jjnr[jidx];
 793             jnrB             = jjnr[jidx+1];
 794             jnrC             = jjnr[jidx+2];
 795             jnrD             = jjnr[jidx+3];
 796             j_coord_offsetA  = DIM*jnrA;
 797             j_coord_offsetB  = DIM*jnrB;
 798             j_coord_offsetC  = DIM*jnrC;
 799             j_coord_offsetD  = DIM*jnrD;
 800
 801             /* load j atom coordinates */
 802             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 803                                               x+j_coord_offsetC,x+j_coord_offsetD,
 804                                               &jx0,&jy0,&jz0);
 805
 806             /* Calculate displacement vector */
 807             dx00             = _mm_sub_ps(ix0,jx0);
 808             dy00             = _mm_sub_ps(iy0,jy0);
 809             dz00             = _mm_sub_ps(iz0,jz0);
 810             dx10             = _mm_sub_ps(ix1,jx0);
 811             dy10             = _mm_sub_ps(iy1,jy0);
 812             dz10             = _mm_sub_ps(iz1,jz0);
 813             dx20             = _mm_sub_ps(ix2,jx0);
 814             dy20             = _mm_sub_ps(iy2,jy0);
 815             dz20             = _mm_sub_ps(iz2,jz0);
 816
 817             /* Calculate squared distance and things based on it */
 818             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 819             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 820             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 821
 822             rinv00           = sse2_invsqrt_f(rsq00);
 823             rinv10           = sse2_invsqrt_f(rsq10);
 824             rinv20           = sse2_invsqrt_f(rsq20);
 825
 826             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 827             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 828             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 829
 830             /* Load parameters for j particles */
 831             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 832                                                               charge+jnrC+0,charge+jnrD+0);
 833             vdwjidx0A        = 2*vdwtype[jnrA+0];
 834             vdwjidx0B        = 2*vdwtype[jnrB+0];
 835             vdwjidx0C        = 2*vdwtype[jnrC+0];
 836             vdwjidx0D        = 2*vdwtype[jnrD+0];
 837
 838             fjx0             = _mm_setzero_ps();
 839             fjy0             = _mm_setzero_ps();
 840             fjz0             = _mm_setzero_ps();
 841
 842             /**************************
 843              * CALCULATE INTERACTIONS *
 844              **************************/
 845
 846             r00              = _mm_mul_ps(rsq00,rinv00);
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             qq00             = _mm_mul_ps(iq0,jq0);
 850             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 851                                          vdwparam+vdwioffset0+vdwjidx0B,
 852                                          vdwparam+vdwioffset0+vdwjidx0C,
 853                                          vdwparam+vdwioffset0+vdwjidx0D,
 854                                          &c6_00,&c12_00);
 855
 856             /* EWALD ELECTROSTATICS */
 857
 858             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 859             ewrt             = _mm_mul_ps(r00,ewtabscale);
 860             ewitab           = _mm_cvttps_epi32(ewrt);
 861             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 862             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 863                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 864                                          &ewtabF,&ewtabFn);
 865             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 866             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 867
 868             /* LENNARD-JONES DISPERSION/REPULSION */
 869
 870             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 871             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 872
 873             fscal            = _mm_add_ps(felec,fvdw);
 874
 875             /* Calculate temporary vectorial force */
 876             tx               = _mm_mul_ps(fscal,dx00);
 877             ty               = _mm_mul_ps(fscal,dy00);
 878             tz               = _mm_mul_ps(fscal,dz00);
 879
 880             /* Update vectorial force */
 881             fix0             = _mm_add_ps(fix0,tx);
 882             fiy0             = _mm_add_ps(fiy0,ty);
 883             fiz0             = _mm_add_ps(fiz0,tz);
 884
 885             fjx0             = _mm_add_ps(fjx0,tx);
 886             fjy0             = _mm_add_ps(fjy0,ty);
 887             fjz0             = _mm_add_ps(fjz0,tz);
 888
 889             /**************************
 890              * CALCULATE INTERACTIONS *
 891              **************************/
 892
 893             r10              = _mm_mul_ps(rsq10,rinv10);
 894
 895             /* Compute parameters for interactions between i and j atoms */
 896             qq10             = _mm_mul_ps(iq1,jq0);
 897
 898             /* EWALD ELECTROSTATICS */
 899
 900             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 901             ewrt             = _mm_mul_ps(r10,ewtabscale);
 902             ewitab           = _mm_cvttps_epi32(ewrt);
 903             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 904             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 905                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 906                                          &ewtabF,&ewtabFn);
 907             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 908             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 909
 910             fscal            = felec;
 911
 912             /* Calculate temporary vectorial force */
 913             tx               = _mm_mul_ps(fscal,dx10);
 914             ty               = _mm_mul_ps(fscal,dy10);
 915             tz               = _mm_mul_ps(fscal,dz10);
 916
 917             /* Update vectorial force */
 918             fix1             = _mm_add_ps(fix1,tx);
 919             fiy1             = _mm_add_ps(fiy1,ty);
 920             fiz1             = _mm_add_ps(fiz1,tz);
 921
 922             fjx0             = _mm_add_ps(fjx0,tx);
 923             fjy0             = _mm_add_ps(fjy0,ty);
 924             fjz0             = _mm_add_ps(fjz0,tz);
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             r20              = _mm_mul_ps(rsq20,rinv20);
 931
 932             /* Compute parameters for interactions between i and j atoms */
 933             qq20             = _mm_mul_ps(iq2,jq0);
 934
 935             /* EWALD ELECTROSTATICS */
 936
 937             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 938             ewrt             = _mm_mul_ps(r20,ewtabscale);
 939             ewitab           = _mm_cvttps_epi32(ewrt);
 940             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 941             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 942                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 943                                          &ewtabF,&ewtabFn);
 944             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 945             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 946
 947             fscal            = felec;
 948
 949             /* Calculate temporary vectorial force */
 950             tx               = _mm_mul_ps(fscal,dx20);
 951             ty               = _mm_mul_ps(fscal,dy20);
 952             tz               = _mm_mul_ps(fscal,dz20);
 953
 954             /* Update vectorial force */
 955             fix2             = _mm_add_ps(fix2,tx);
 956             fiy2             = _mm_add_ps(fiy2,ty);
 957             fiz2             = _mm_add_ps(fiz2,tz);
 958
 959             fjx0             = _mm_add_ps(fjx0,tx);
 960             fjy0             = _mm_add_ps(fjy0,ty);
 961             fjz0             = _mm_add_ps(fjz0,tz);
 962
 963             fjptrA             = f+j_coord_offsetA;
 964             fjptrB             = f+j_coord_offsetB;
 965             fjptrC             = f+j_coord_offsetC;
 966             fjptrD             = f+j_coord_offsetD;
 967
 968             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 969
 970             /* Inner loop uses 115 flops */
 971         }
 972
 973         if(jidx<j_index_end)
 974         {
 975
 976             /* Get j neighbor index, and coordinate index */
 977             jnrlistA         = jjnr[jidx];
 978             jnrlistB         = jjnr[jidx+1];
 979             jnrlistC         = jjnr[jidx+2];
 980             jnrlistD         = jjnr[jidx+3];
 981             /* Sign of each element will be negative for non-real atoms.
 982              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 983              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 984              */
 985             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 986             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 987             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 988             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 989             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 990             j_coord_offsetA  = DIM*jnrA;
 991             j_coord_offsetB  = DIM*jnrB;
 992             j_coord_offsetC  = DIM*jnrC;
 993             j_coord_offsetD  = DIM*jnrD;
 994
 995             /* load j atom coordinates */
 996             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 997                                               x+j_coord_offsetC,x+j_coord_offsetD,
 998                                               &jx0,&jy0,&jz0);
 999
1000             /* Calculate displacement vector */
1001             dx00             = _mm_sub_ps(ix0,jx0);
1002             dy00             = _mm_sub_ps(iy0,jy0);
1003             dz00             = _mm_sub_ps(iz0,jz0);
1004             dx10             = _mm_sub_ps(ix1,jx0);
1005             dy10             = _mm_sub_ps(iy1,jy0);
1006             dz10             = _mm_sub_ps(iz1,jz0);
1007             dx20             = _mm_sub_ps(ix2,jx0);
1008             dy20             = _mm_sub_ps(iy2,jy0);
1009             dz20             = _mm_sub_ps(iz2,jz0);
1010
1011             /* Calculate squared distance and things based on it */
1012             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1013             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1014             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1015
1016             rinv00           = sse2_invsqrt_f(rsq00);
1017             rinv10           = sse2_invsqrt_f(rsq10);
1018             rinv20           = sse2_invsqrt_f(rsq20);
1019
1020             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1021             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1022             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1023
1024             /* Load parameters for j particles */
1025             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1026                                                               charge+jnrC+0,charge+jnrD+0);
1027             vdwjidx0A        = 2*vdwtype[jnrA+0];
1028             vdwjidx0B        = 2*vdwtype[jnrB+0];
1029             vdwjidx0C        = 2*vdwtype[jnrC+0];
1030             vdwjidx0D        = 2*vdwtype[jnrD+0];
1031
1032             fjx0             = _mm_setzero_ps();
1033             fjy0             = _mm_setzero_ps();
1034             fjz0             = _mm_setzero_ps();
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             r00              = _mm_mul_ps(rsq00,rinv00);
1041             r00              = _mm_andnot_ps(dummy_mask,r00);
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq00             = _mm_mul_ps(iq0,jq0);
1045             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1046                                          vdwparam+vdwioffset0+vdwjidx0B,
1047                                          vdwparam+vdwioffset0+vdwjidx0C,
1048                                          vdwparam+vdwioffset0+vdwjidx0D,
1049                                          &c6_00,&c12_00);
1050
1051             /* EWALD ELECTROSTATICS */
1052
1053             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1054             ewrt             = _mm_mul_ps(r00,ewtabscale);
1055             ewitab           = _mm_cvttps_epi32(ewrt);
1056             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1057             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1058                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1059                                          &ewtabF,&ewtabFn);
1060             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1061             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1062
1063             /* LENNARD-JONES DISPERSION/REPULSION */
1064
1065             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1066             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1067
1068             fscal            = _mm_add_ps(felec,fvdw);
1069
1070             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1071
1072             /* Calculate temporary vectorial force */
1073             tx               = _mm_mul_ps(fscal,dx00);
1074             ty               = _mm_mul_ps(fscal,dy00);
1075             tz               = _mm_mul_ps(fscal,dz00);
1076
1077             /* Update vectorial force */
1078             fix0             = _mm_add_ps(fix0,tx);
1079             fiy0             = _mm_add_ps(fiy0,ty);
1080             fiz0             = _mm_add_ps(fiz0,tz);
1081
1082             fjx0             = _mm_add_ps(fjx0,tx);
1083             fjy0             = _mm_add_ps(fjy0,ty);
1084             fjz0             = _mm_add_ps(fjz0,tz);
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             r10              = _mm_mul_ps(rsq10,rinv10);
1091             r10              = _mm_andnot_ps(dummy_mask,r10);
1092
1093             /* Compute parameters for interactions between i and j atoms */
1094             qq10             = _mm_mul_ps(iq1,jq0);
1095
1096             /* EWALD ELECTROSTATICS */
1097
1098             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1099             ewrt             = _mm_mul_ps(r10,ewtabscale);
1100             ewitab           = _mm_cvttps_epi32(ewrt);
1101             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1102             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1103                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1104                                          &ewtabF,&ewtabFn);
1105             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1106             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1107
1108             fscal            = felec;
1109
1110             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm_mul_ps(fscal,dx10);
1114             ty               = _mm_mul_ps(fscal,dy10);
1115             tz               = _mm_mul_ps(fscal,dz10);
1116
1117             /* Update vectorial force */
1118             fix1             = _mm_add_ps(fix1,tx);
1119             fiy1             = _mm_add_ps(fiy1,ty);
1120             fiz1             = _mm_add_ps(fiz1,tz);
1121
1122             fjx0             = _mm_add_ps(fjx0,tx);
1123             fjy0             = _mm_add_ps(fjy0,ty);
1124             fjz0             = _mm_add_ps(fjz0,tz);
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             r20              = _mm_mul_ps(rsq20,rinv20);
1131             r20              = _mm_andnot_ps(dummy_mask,r20);
1132
1133             /* Compute parameters for interactions between i and j atoms */
1134             qq20             = _mm_mul_ps(iq2,jq0);
1135
1136             /* EWALD ELECTROSTATICS */
1137
1138             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1139             ewrt             = _mm_mul_ps(r20,ewtabscale);
1140             ewitab           = _mm_cvttps_epi32(ewrt);
1141             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1142             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1143                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1144                                          &ewtabF,&ewtabFn);
1145             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1146             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1147
1148             fscal            = felec;
1149
1150             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1151
1152             /* Calculate temporary vectorial force */
1153             tx               = _mm_mul_ps(fscal,dx20);
1154             ty               = _mm_mul_ps(fscal,dy20);
1155             tz               = _mm_mul_ps(fscal,dz20);
1156
1157             /* Update vectorial force */
1158             fix2             = _mm_add_ps(fix2,tx);
1159             fiy2             = _mm_add_ps(fiy2,ty);
1160             fiz2             = _mm_add_ps(fiz2,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1167             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1168             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1169             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1170
1171             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1172
1173             /* Inner loop uses 118 flops */
1174         }
1175
1176         /* End of innermost loop */
1177
1178         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1179                                               f+i_coord_offset,fshift+i_shift_offset);
1180
1181         /* Increment number of inner iterations */
1182         inneriter                  += j_index_end - j_index_start;
1183
1184         /* Outer loop uses 18 flops */
1185     }
1186
1187     /* Increment number of outer iterations */
1188     outeriter        += nri;
1189
1190     /* Update outer/inner flops */
1191
1192     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1193 }