src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJ_GeomW3P1_sse2_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 sse2_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_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse2_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          ewitab;
 103     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 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     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 134     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 135     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 170
 171         fix0             = _mm_setzero_ps();
 172         fiy0             = _mm_setzero_ps();
 173         fiz0             = _mm_setzero_ps();
 174         fix1             = _mm_setzero_ps();
 175         fiy1             = _mm_setzero_ps();
 176         fiz1             = _mm_setzero_ps();
 177         fix2             = _mm_setzero_ps();
 178         fiy2             = _mm_setzero_ps();
 179         fiz2             = _mm_setzero_ps();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm_setzero_ps();
 183         vvdwsum          = _mm_setzero_ps();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               x+j_coord_offsetC,x+j_coord_offsetD,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_ps(ix0,jx0);
 206             dy00             = _mm_sub_ps(iy0,jy0);
 207             dz00             = _mm_sub_ps(iz0,jz0);
 208             dx10             = _mm_sub_ps(ix1,jx0);
 209             dy10             = _mm_sub_ps(iy1,jy0);
 210             dz10             = _mm_sub_ps(iz1,jz0);
 211             dx20             = _mm_sub_ps(ix2,jx0);
 212             dy20             = _mm_sub_ps(iy2,jy0);
 213             dz20             = _mm_sub_ps(iz2,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 217             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 218             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 219
 220             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 221             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 222             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 223
 224             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 225             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 226             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 230                                                               charge+jnrC+0,charge+jnrD+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233             vdwjidx0C        = 2*vdwtype[jnrC+0];
 234             vdwjidx0D        = 2*vdwtype[jnrD+0];
 235
 236             fjx0             = _mm_setzero_ps();
 237             fjy0             = _mm_setzero_ps();
 238             fjz0             = _mm_setzero_ps();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             r00              = _mm_mul_ps(rsq00,rinv00);
 245
 246             /* Compute parameters for interactions between i and j atoms */
 247             qq00             = _mm_mul_ps(iq0,jq0);
 248             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 249                                          vdwparam+vdwioffset0+vdwjidx0B,
 250                                          vdwparam+vdwioffset0+vdwjidx0C,
 251                                          vdwparam+vdwioffset0+vdwjidx0D,
 252                                          &c6_00,&c12_00);
 253
 254             /* EWALD ELECTROSTATICS */
 255
 256             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 257             ewrt             = _mm_mul_ps(r00,ewtabscale);
 258             ewitab           = _mm_cvttps_epi32(ewrt);
 259             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 260             ewitab           = _mm_slli_epi32(ewitab,2);
 261             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 262             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 263             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 264             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 265             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 266             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 267             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 268             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 269             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 270
 271             /* LENNARD-JONES DISPERSION/REPULSION */
 272
 273             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 274             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 275             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 276             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 277             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             velecsum         = _mm_add_ps(velecsum,velec);
 281             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 282
 283             fscal            = _mm_add_ps(felec,fvdw);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_ps(fscal,dx00);
 287             ty               = _mm_mul_ps(fscal,dy00);
 288             tz               = _mm_mul_ps(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_add_ps(fix0,tx);
 292             fiy0             = _mm_add_ps(fiy0,ty);
 293             fiz0             = _mm_add_ps(fiz0,tz);
 294
 295             fjx0             = _mm_add_ps(fjx0,tx);
 296             fjy0             = _mm_add_ps(fjy0,ty);
 297             fjz0             = _mm_add_ps(fjz0,tz);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r10              = _mm_mul_ps(rsq10,rinv10);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_ps(iq1,jq0);
 307
 308             /* EWALD ELECTROSTATICS */
 309
 310             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 311             ewrt             = _mm_mul_ps(r10,ewtabscale);
 312             ewitab           = _mm_cvttps_epi32(ewrt);
 313             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 314             ewitab           = _mm_slli_epi32(ewitab,2);
 315             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 316             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 317             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 318             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 319             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 320             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 321             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 322             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 323             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velecsum         = _mm_add_ps(velecsum,velec);
 327
 328             fscal            = felec;
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_ps(fscal,dx10);
 332             ty               = _mm_mul_ps(fscal,dy10);
 333             tz               = _mm_mul_ps(fscal,dz10);
 334
 335             /* Update vectorial force */
 336             fix1             = _mm_add_ps(fix1,tx);
 337             fiy1             = _mm_add_ps(fiy1,ty);
 338             fiz1             = _mm_add_ps(fiz1,tz);
 339
 340             fjx0             = _mm_add_ps(fjx0,tx);
 341             fjy0             = _mm_add_ps(fjy0,ty);
 342             fjz0             = _mm_add_ps(fjz0,tz);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             r20              = _mm_mul_ps(rsq20,rinv20);
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm_mul_ps(iq2,jq0);
 352
 353             /* EWALD ELECTROSTATICS */
 354
 355             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 356             ewrt             = _mm_mul_ps(r20,ewtabscale);
 357             ewitab           = _mm_cvttps_epi32(ewrt);
 358             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 359             ewitab           = _mm_slli_epi32(ewitab,2);
 360             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 361             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 362             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 363             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 364             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 365             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 366             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 367             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 368             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velecsum         = _mm_add_ps(velecsum,velec);
 372
 373             fscal            = felec;
 374
 375             /* Calculate temporary vectorial force */
 376             tx               = _mm_mul_ps(fscal,dx20);
 377             ty               = _mm_mul_ps(fscal,dy20);
 378             tz               = _mm_mul_ps(fscal,dz20);
 379
 380             /* Update vectorial force */
 381             fix2             = _mm_add_ps(fix2,tx);
 382             fiy2             = _mm_add_ps(fiy2,ty);
 383             fiz2             = _mm_add_ps(fiz2,tz);
 384
 385             fjx0             = _mm_add_ps(fjx0,tx);
 386             fjy0             = _mm_add_ps(fjy0,ty);
 387             fjz0             = _mm_add_ps(fjz0,tz);
 388
 389             fjptrA             = f+j_coord_offsetA;
 390             fjptrB             = f+j_coord_offsetB;
 391             fjptrC             = f+j_coord_offsetC;
 392             fjptrD             = f+j_coord_offsetD;
 393
 394             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 395
 396             /* Inner loop uses 135 flops */
 397         }
 398
 399         if(jidx<j_index_end)
 400         {
 401
 402             /* Get j neighbor index, and coordinate index */
 403             jnrlistA         = jjnr[jidx];
 404             jnrlistB         = jjnr[jidx+1];
 405             jnrlistC         = jjnr[jidx+2];
 406             jnrlistD         = jjnr[jidx+3];
 407             /* Sign of each element will be negative for non-real atoms.
 408              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 409              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 410              */
 411             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 412             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 413             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 414             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 415             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 416             j_coord_offsetA  = DIM*jnrA;
 417             j_coord_offsetB  = DIM*jnrB;
 418             j_coord_offsetC  = DIM*jnrC;
 419             j_coord_offsetD  = DIM*jnrD;
 420
 421             /* load j atom coordinates */
 422             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 423                                               x+j_coord_offsetC,x+j_coord_offsetD,
 424                                               &jx0,&jy0,&jz0);
 425
 426             /* Calculate displacement vector */
 427             dx00             = _mm_sub_ps(ix0,jx0);
 428             dy00             = _mm_sub_ps(iy0,jy0);
 429             dz00             = _mm_sub_ps(iz0,jz0);
 430             dx10             = _mm_sub_ps(ix1,jx0);
 431             dy10             = _mm_sub_ps(iy1,jy0);
 432             dz10             = _mm_sub_ps(iz1,jz0);
 433             dx20             = _mm_sub_ps(ix2,jx0);
 434             dy20             = _mm_sub_ps(iy2,jy0);
 435             dz20             = _mm_sub_ps(iz2,jz0);
 436
 437             /* Calculate squared distance and things based on it */
 438             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 439             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 440             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 441
 442             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 443             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 444             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 445
 446             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 447             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 448             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 449
 450             /* Load parameters for j particles */
 451             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 452                                                               charge+jnrC+0,charge+jnrD+0);
 453             vdwjidx0A        = 2*vdwtype[jnrA+0];
 454             vdwjidx0B        = 2*vdwtype[jnrB+0];
 455             vdwjidx0C        = 2*vdwtype[jnrC+0];
 456             vdwjidx0D        = 2*vdwtype[jnrD+0];
 457
 458             fjx0             = _mm_setzero_ps();
 459             fjy0             = _mm_setzero_ps();
 460             fjz0             = _mm_setzero_ps();
 461
 462             /**************************
 463              * CALCULATE INTERACTIONS *
 464              **************************/
 465
 466             r00              = _mm_mul_ps(rsq00,rinv00);
 467             r00              = _mm_andnot_ps(dummy_mask,r00);
 468
 469             /* Compute parameters for interactions between i and j atoms */
 470             qq00             = _mm_mul_ps(iq0,jq0);
 471             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 472                                          vdwparam+vdwioffset0+vdwjidx0B,
 473                                          vdwparam+vdwioffset0+vdwjidx0C,
 474                                          vdwparam+vdwioffset0+vdwjidx0D,
 475                                          &c6_00,&c12_00);
 476
 477             /* EWALD ELECTROSTATICS */
 478
 479             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 480             ewrt             = _mm_mul_ps(r00,ewtabscale);
 481             ewitab           = _mm_cvttps_epi32(ewrt);
 482             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 483             ewitab           = _mm_slli_epi32(ewitab,2);
 484             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 485             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 486             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 487             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 488             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 489             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 490             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 491             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 492             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 493
 494             /* LENNARD-JONES DISPERSION/REPULSION */
 495
 496             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 497             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 498             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 499             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 500             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 501
 502             /* Update potential sum for this i atom from the interaction with this j atom. */
 503             velec            = _mm_andnot_ps(dummy_mask,velec);
 504             velecsum         = _mm_add_ps(velecsum,velec);
 505             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 506             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 507
 508             fscal            = _mm_add_ps(felec,fvdw);
 509
 510             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 511
 512             /* Calculate temporary vectorial force */
 513             tx               = _mm_mul_ps(fscal,dx00);
 514             ty               = _mm_mul_ps(fscal,dy00);
 515             tz               = _mm_mul_ps(fscal,dz00);
 516
 517             /* Update vectorial force */
 518             fix0             = _mm_add_ps(fix0,tx);
 519             fiy0             = _mm_add_ps(fiy0,ty);
 520             fiz0             = _mm_add_ps(fiz0,tz);
 521
 522             fjx0             = _mm_add_ps(fjx0,tx);
 523             fjy0             = _mm_add_ps(fjy0,ty);
 524             fjz0             = _mm_add_ps(fjz0,tz);
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             r10              = _mm_mul_ps(rsq10,rinv10);
 531             r10              = _mm_andnot_ps(dummy_mask,r10);
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             qq10             = _mm_mul_ps(iq1,jq0);
 535
 536             /* EWALD ELECTROSTATICS */
 537
 538             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 539             ewrt             = _mm_mul_ps(r10,ewtabscale);
 540             ewitab           = _mm_cvttps_epi32(ewrt);
 541             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 542             ewitab           = _mm_slli_epi32(ewitab,2);
 543             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 544             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 545             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 546             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 547             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 548             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 549             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 550             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 551             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm_andnot_ps(dummy_mask,velec);
 555             velecsum         = _mm_add_ps(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm_mul_ps(fscal,dx10);
 563             ty               = _mm_mul_ps(fscal,dy10);
 564             tz               = _mm_mul_ps(fscal,dz10);
 565
 566             /* Update vectorial force */
 567             fix1             = _mm_add_ps(fix1,tx);
 568             fiy1             = _mm_add_ps(fiy1,ty);
 569             fiz1             = _mm_add_ps(fiz1,tz);
 570
 571             fjx0             = _mm_add_ps(fjx0,tx);
 572             fjy0             = _mm_add_ps(fjy0,ty);
 573             fjz0             = _mm_add_ps(fjz0,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r20              = _mm_mul_ps(rsq20,rinv20);
 580             r20              = _mm_andnot_ps(dummy_mask,r20);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq20             = _mm_mul_ps(iq2,jq0);
 584
 585             /* EWALD ELECTROSTATICS */
 586
 587             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 588             ewrt             = _mm_mul_ps(r20,ewtabscale);
 589             ewitab           = _mm_cvttps_epi32(ewrt);
 590             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 591             ewitab           = _mm_slli_epi32(ewitab,2);
 592             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 593             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 594             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 595             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 596             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 597             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 598             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 599             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 600             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             velec            = _mm_andnot_ps(dummy_mask,velec);
 604             velecsum         = _mm_add_ps(velecsum,velec);
 605
 606             fscal            = felec;
 607
 608             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 609
 610             /* Calculate temporary vectorial force */
 611             tx               = _mm_mul_ps(fscal,dx20);
 612             ty               = _mm_mul_ps(fscal,dy20);
 613             tz               = _mm_mul_ps(fscal,dz20);
 614
 615             /* Update vectorial force */
 616             fix2             = _mm_add_ps(fix2,tx);
 617             fiy2             = _mm_add_ps(fiy2,ty);
 618             fiz2             = _mm_add_ps(fiz2,tz);
 619
 620             fjx0             = _mm_add_ps(fjx0,tx);
 621             fjy0             = _mm_add_ps(fjy0,ty);
 622             fjz0             = _mm_add_ps(fjz0,tz);
 623
 624             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 625             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 626             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 627             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 628
 629             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 630
 631             /* Inner loop uses 138 flops */
 632         }
 633
 634         /* End of innermost loop */
 635
 636         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 637                                               f+i_coord_offset,fshift+i_shift_offset);
 638
 639         ggid                        = gid[iidx];
 640         /* Update potential energies */
 641         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 642         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 643
 644         /* Increment number of inner iterations */
 645         inneriter                  += j_index_end - j_index_start;
 646
 647         /* Outer loop uses 20 flops */
 648     }
 649
 650     /* Increment number of outer iterations */
 651     outeriter        += nri;
 652
 653     /* Update outer/inner flops */
 654
 655     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*138);
 656 }
 657 /*
 658  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse2_single
 659  * Electrostatics interaction: Ewald
 660  * VdW interaction:            LennardJones
 661  * Geometry:                   Water3-Particle
 662  * Calculate force/pot:        Force
 663  */
 664 void
 665 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse2_single
 666                     (t_nblist                    * gmx_restrict       nlist,
 667                      rvec                        * gmx_restrict          xx,
 668                      rvec                        * gmx_restrict          ff,
 669                      t_forcerec                  * gmx_restrict          fr,
 670                      t_mdatoms                   * gmx_restrict     mdatoms,
 671                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 672                      t_nrnb                      * gmx_restrict        nrnb)
 673 {
 674     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 675      * just 0 for non-waters.
 676      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 677      * jnr indices corresponding to data put in the four positions in the SIMD register.
 678      */
 679     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 680     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 681     int              jnrA,jnrB,jnrC,jnrD;
 682     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 683     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 684     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 685     real             rcutoff_scalar;
 686     real             *shiftvec,*fshift,*x,*f;
 687     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 688     real             scratch[4*DIM];
 689     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 690     int              vdwioffset0;
 691     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 692     int              vdwioffset1;
 693     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 694     int              vdwioffset2;
 695     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 696     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 697     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 698     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 699     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 700     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 701     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 702     real             *charge;
 703     int              nvdwtype;
 704     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 705     int              *vdwtype;
 706     real             *vdwparam;
 707     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 708     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 709     __m128i          ewitab;
 710     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 711     real             *ewtab;
 712     __m128           dummy_mask,cutoff_mask;
 713     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 714     __m128           one     = _mm_set1_ps(1.0);
 715     __m128           two     = _mm_set1_ps(2.0);
 716     x                = xx[0];
 717     f                = ff[0];
 718
 719     nri              = nlist->nri;
 720     iinr             = nlist->iinr;
 721     jindex           = nlist->jindex;
 722     jjnr             = nlist->jjnr;
 723     shiftidx         = nlist->shift;
 724     gid              = nlist->gid;
 725     shiftvec         = fr->shift_vec[0];
 726     fshift           = fr->fshift[0];
 727     facel            = _mm_set1_ps(fr->epsfac);
 728     charge           = mdatoms->chargeA;
 729     nvdwtype         = fr->ntype;
 730     vdwparam         = fr->nbfp;
 731     vdwtype          = mdatoms->typeA;
 732
 733     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 734     ewtab            = fr->ic->tabq_coul_F;
 735     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 736     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 737
 738     /* Setup water-specific parameters */
 739     inr              = nlist->iinr[0];
 740     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 741     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 742     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 743     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 744
 745     /* Avoid stupid compiler warnings */
 746     jnrA = jnrB = jnrC = jnrD = 0;
 747     j_coord_offsetA = 0;
 748     j_coord_offsetB = 0;
 749     j_coord_offsetC = 0;
 750     j_coord_offsetD = 0;
 751
 752     outeriter        = 0;
 753     inneriter        = 0;
 754
 755     for(iidx=0;iidx<4*DIM;iidx++)
 756     {
 757         scratch[iidx] = 0.0;
 758     }
 759
 760     /* Start outer loop over neighborlists */
 761     for(iidx=0; iidx<nri; iidx++)
 762     {
 763         /* Load shift vector for this list */
 764         i_shift_offset   = DIM*shiftidx[iidx];
 765
 766         /* Load limits for loop over neighbors */
 767         j_index_start    = jindex[iidx];
 768         j_index_end      = jindex[iidx+1];
 769
 770         /* Get outer coordinate index */
 771         inr              = iinr[iidx];
 772         i_coord_offset   = DIM*inr;
 773
 774         /* Load i particle coords and add shift vector */
 775         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 776                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 777
 778         fix0             = _mm_setzero_ps();
 779         fiy0             = _mm_setzero_ps();
 780         fiz0             = _mm_setzero_ps();
 781         fix1             = _mm_setzero_ps();
 782         fiy1             = _mm_setzero_ps();
 783         fiz1             = _mm_setzero_ps();
 784         fix2             = _mm_setzero_ps();
 785         fiy2             = _mm_setzero_ps();
 786         fiz2             = _mm_setzero_ps();
 787
 788         /* Start inner kernel loop */
 789         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 790         {
 791
 792             /* Get j neighbor index, and coordinate index */
 793             jnrA             = jjnr[jidx];
 794             jnrB             = jjnr[jidx+1];
 795             jnrC             = jjnr[jidx+2];
 796             jnrD             = jjnr[jidx+3];
 797             j_coord_offsetA  = DIM*jnrA;
 798             j_coord_offsetB  = DIM*jnrB;
 799             j_coord_offsetC  = DIM*jnrC;
 800             j_coord_offsetD  = DIM*jnrD;
 801
 802             /* load j atom coordinates */
 803             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 804                                               x+j_coord_offsetC,x+j_coord_offsetD,
 805                                               &jx0,&jy0,&jz0);
 806
 807             /* Calculate displacement vector */
 808             dx00             = _mm_sub_ps(ix0,jx0);
 809             dy00             = _mm_sub_ps(iy0,jy0);
 810             dz00             = _mm_sub_ps(iz0,jz0);
 811             dx10             = _mm_sub_ps(ix1,jx0);
 812             dy10             = _mm_sub_ps(iy1,jy0);
 813             dz10             = _mm_sub_ps(iz1,jz0);
 814             dx20             = _mm_sub_ps(ix2,jx0);
 815             dy20             = _mm_sub_ps(iy2,jy0);
 816             dz20             = _mm_sub_ps(iz2,jz0);
 817
 818             /* Calculate squared distance and things based on it */
 819             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 820             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 821             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 822
 823             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 824             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 825             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 826
 827             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 828             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 829             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 830
 831             /* Load parameters for j particles */
 832             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 833                                                               charge+jnrC+0,charge+jnrD+0);
 834             vdwjidx0A        = 2*vdwtype[jnrA+0];
 835             vdwjidx0B        = 2*vdwtype[jnrB+0];
 836             vdwjidx0C        = 2*vdwtype[jnrC+0];
 837             vdwjidx0D        = 2*vdwtype[jnrD+0];
 838
 839             fjx0             = _mm_setzero_ps();
 840             fjy0             = _mm_setzero_ps();
 841             fjz0             = _mm_setzero_ps();
 842
 843             /**************************
 844              * CALCULATE INTERACTIONS *
 845              **************************/
 846
 847             r00              = _mm_mul_ps(rsq00,rinv00);
 848
 849             /* Compute parameters for interactions between i and j atoms */
 850             qq00             = _mm_mul_ps(iq0,jq0);
 851             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 852                                          vdwparam+vdwioffset0+vdwjidx0B,
 853                                          vdwparam+vdwioffset0+vdwjidx0C,
 854                                          vdwparam+vdwioffset0+vdwjidx0D,
 855                                          &c6_00,&c12_00);
 856
 857             /* EWALD ELECTROSTATICS */
 858
 859             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 860             ewrt             = _mm_mul_ps(r00,ewtabscale);
 861             ewitab           = _mm_cvttps_epi32(ewrt);
 862             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 863             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 864                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 865                                          &ewtabF,&ewtabFn);
 866             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 867             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 868
 869             /* LENNARD-JONES DISPERSION/REPULSION */
 870
 871             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 872             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 873
 874             fscal            = _mm_add_ps(felec,fvdw);
 875
 876             /* Calculate temporary vectorial force */
 877             tx               = _mm_mul_ps(fscal,dx00);
 878             ty               = _mm_mul_ps(fscal,dy00);
 879             tz               = _mm_mul_ps(fscal,dz00);
 880
 881             /* Update vectorial force */
 882             fix0             = _mm_add_ps(fix0,tx);
 883             fiy0             = _mm_add_ps(fiy0,ty);
 884             fiz0             = _mm_add_ps(fiz0,tz);
 885
 886             fjx0             = _mm_add_ps(fjx0,tx);
 887             fjy0             = _mm_add_ps(fjy0,ty);
 888             fjz0             = _mm_add_ps(fjz0,tz);
 889
 890             /**************************
 891              * CALCULATE INTERACTIONS *
 892              **************************/
 893
 894             r10              = _mm_mul_ps(rsq10,rinv10);
 895
 896             /* Compute parameters for interactions between i and j atoms */
 897             qq10             = _mm_mul_ps(iq1,jq0);
 898
 899             /* EWALD ELECTROSTATICS */
 900
 901             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 902             ewrt             = _mm_mul_ps(r10,ewtabscale);
 903             ewitab           = _mm_cvttps_epi32(ewrt);
 904             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 905             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 906                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 907                                          &ewtabF,&ewtabFn);
 908             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 909             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 910
 911             fscal            = felec;
 912
 913             /* Calculate temporary vectorial force */
 914             tx               = _mm_mul_ps(fscal,dx10);
 915             ty               = _mm_mul_ps(fscal,dy10);
 916             tz               = _mm_mul_ps(fscal,dz10);
 917
 918             /* Update vectorial force */
 919             fix1             = _mm_add_ps(fix1,tx);
 920             fiy1             = _mm_add_ps(fiy1,ty);
 921             fiz1             = _mm_add_ps(fiz1,tz);
 922
 923             fjx0             = _mm_add_ps(fjx0,tx);
 924             fjy0             = _mm_add_ps(fjy0,ty);
 925             fjz0             = _mm_add_ps(fjz0,tz);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             r20              = _mm_mul_ps(rsq20,rinv20);
 932
 933             /* Compute parameters for interactions between i and j atoms */
 934             qq20             = _mm_mul_ps(iq2,jq0);
 935
 936             /* EWALD ELECTROSTATICS */
 937
 938             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 939             ewrt             = _mm_mul_ps(r20,ewtabscale);
 940             ewitab           = _mm_cvttps_epi32(ewrt);
 941             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 942             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 943                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 944                                          &ewtabF,&ewtabFn);
 945             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 946             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 947
 948             fscal            = felec;
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm_mul_ps(fscal,dx20);
 952             ty               = _mm_mul_ps(fscal,dy20);
 953             tz               = _mm_mul_ps(fscal,dz20);
 954
 955             /* Update vectorial force */
 956             fix2             = _mm_add_ps(fix2,tx);
 957             fiy2             = _mm_add_ps(fiy2,ty);
 958             fiz2             = _mm_add_ps(fiz2,tz);
 959
 960             fjx0             = _mm_add_ps(fjx0,tx);
 961             fjy0             = _mm_add_ps(fjy0,ty);
 962             fjz0             = _mm_add_ps(fjz0,tz);
 963
 964             fjptrA             = f+j_coord_offsetA;
 965             fjptrB             = f+j_coord_offsetB;
 966             fjptrC             = f+j_coord_offsetC;
 967             fjptrD             = f+j_coord_offsetD;
 968
 969             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 970
 971             /* Inner loop uses 115 flops */
 972         }
 973
 974         if(jidx<j_index_end)
 975         {
 976
 977             /* Get j neighbor index, and coordinate index */
 978             jnrlistA         = jjnr[jidx];
 979             jnrlistB         = jjnr[jidx+1];
 980             jnrlistC         = jjnr[jidx+2];
 981             jnrlistD         = jjnr[jidx+3];
 982             /* Sign of each element will be negative for non-real atoms.
 983              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 984              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 985              */
 986             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 987             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 988             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 989             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 990             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 991             j_coord_offsetA  = DIM*jnrA;
 992             j_coord_offsetB  = DIM*jnrB;
 993             j_coord_offsetC  = DIM*jnrC;
 994             j_coord_offsetD  = DIM*jnrD;
 995
 996             /* load j atom coordinates */
 997             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 998                                               x+j_coord_offsetC,x+j_coord_offsetD,
 999                                               &jx0,&jy0,&jz0);
1000
1001             /* Calculate displacement vector */
1002             dx00             = _mm_sub_ps(ix0,jx0);
1003             dy00             = _mm_sub_ps(iy0,jy0);
1004             dz00             = _mm_sub_ps(iz0,jz0);
1005             dx10             = _mm_sub_ps(ix1,jx0);
1006             dy10             = _mm_sub_ps(iy1,jy0);
1007             dz10             = _mm_sub_ps(iz1,jz0);
1008             dx20             = _mm_sub_ps(ix2,jx0);
1009             dy20             = _mm_sub_ps(iy2,jy0);
1010             dz20             = _mm_sub_ps(iz2,jz0);
1011
1012             /* Calculate squared distance and things based on it */
1013             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1014             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1015             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1016
1017             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1018             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1019             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1020
1021             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1022             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1023             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1024
1025             /* Load parameters for j particles */
1026             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1027                                                               charge+jnrC+0,charge+jnrD+0);
1028             vdwjidx0A        = 2*vdwtype[jnrA+0];
1029             vdwjidx0B        = 2*vdwtype[jnrB+0];
1030             vdwjidx0C        = 2*vdwtype[jnrC+0];
1031             vdwjidx0D        = 2*vdwtype[jnrD+0];
1032
1033             fjx0             = _mm_setzero_ps();
1034             fjy0             = _mm_setzero_ps();
1035             fjz0             = _mm_setzero_ps();
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             r00              = _mm_mul_ps(rsq00,rinv00);
1042             r00              = _mm_andnot_ps(dummy_mask,r00);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq00             = _mm_mul_ps(iq0,jq0);
1046             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1047                                          vdwparam+vdwioffset0+vdwjidx0B,
1048                                          vdwparam+vdwioffset0+vdwjidx0C,
1049                                          vdwparam+vdwioffset0+vdwjidx0D,
1050                                          &c6_00,&c12_00);
1051
1052             /* EWALD ELECTROSTATICS */
1053
1054             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1055             ewrt             = _mm_mul_ps(r00,ewtabscale);
1056             ewitab           = _mm_cvttps_epi32(ewrt);
1057             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1058             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1059                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1060                                          &ewtabF,&ewtabFn);
1061             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1062             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1063
1064             /* LENNARD-JONES DISPERSION/REPULSION */
1065
1066             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1067             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1068
1069             fscal            = _mm_add_ps(felec,fvdw);
1070
1071             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1072
1073             /* Calculate temporary vectorial force */
1074             tx               = _mm_mul_ps(fscal,dx00);
1075             ty               = _mm_mul_ps(fscal,dy00);
1076             tz               = _mm_mul_ps(fscal,dz00);
1077
1078             /* Update vectorial force */
1079             fix0             = _mm_add_ps(fix0,tx);
1080             fiy0             = _mm_add_ps(fiy0,ty);
1081             fiz0             = _mm_add_ps(fiz0,tz);
1082
1083             fjx0             = _mm_add_ps(fjx0,tx);
1084             fjy0             = _mm_add_ps(fjy0,ty);
1085             fjz0             = _mm_add_ps(fjz0,tz);
1086
1087             /**************************
1088              * CALCULATE INTERACTIONS *
1089              **************************/
1090
1091             r10              = _mm_mul_ps(rsq10,rinv10);
1092             r10              = _mm_andnot_ps(dummy_mask,r10);
1093
1094             /* Compute parameters for interactions between i and j atoms */
1095             qq10             = _mm_mul_ps(iq1,jq0);
1096
1097             /* EWALD ELECTROSTATICS */
1098
1099             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1100             ewrt             = _mm_mul_ps(r10,ewtabscale);
1101             ewitab           = _mm_cvttps_epi32(ewrt);
1102             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1103             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1104                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1105                                          &ewtabF,&ewtabFn);
1106             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1107             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1108
1109             fscal            = felec;
1110
1111             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1112
1113             /* Calculate temporary vectorial force */
1114             tx               = _mm_mul_ps(fscal,dx10);
1115             ty               = _mm_mul_ps(fscal,dy10);
1116             tz               = _mm_mul_ps(fscal,dz10);
1117
1118             /* Update vectorial force */
1119             fix1             = _mm_add_ps(fix1,tx);
1120             fiy1             = _mm_add_ps(fiy1,ty);
1121             fiz1             = _mm_add_ps(fiz1,tz);
1122
1123             fjx0             = _mm_add_ps(fjx0,tx);
1124             fjy0             = _mm_add_ps(fjy0,ty);
1125             fjz0             = _mm_add_ps(fjz0,tz);
1126
1127             /**************************
1128              * CALCULATE INTERACTIONS *
1129              **************************/
1130
1131             r20              = _mm_mul_ps(rsq20,rinv20);
1132             r20              = _mm_andnot_ps(dummy_mask,r20);
1133
1134             /* Compute parameters for interactions between i and j atoms */
1135             qq20             = _mm_mul_ps(iq2,jq0);
1136
1137             /* EWALD ELECTROSTATICS */
1138
1139             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1140             ewrt             = _mm_mul_ps(r20,ewtabscale);
1141             ewitab           = _mm_cvttps_epi32(ewrt);
1142             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1143             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1144                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1145                                          &ewtabF,&ewtabFn);
1146             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1147             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1148
1149             fscal            = felec;
1150
1151             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1152
1153             /* Calculate temporary vectorial force */
1154             tx               = _mm_mul_ps(fscal,dx20);
1155             ty               = _mm_mul_ps(fscal,dy20);
1156             tz               = _mm_mul_ps(fscal,dz20);
1157
1158             /* Update vectorial force */
1159             fix2             = _mm_add_ps(fix2,tx);
1160             fiy2             = _mm_add_ps(fiy2,ty);
1161             fiz2             = _mm_add_ps(fiz2,tz);
1162
1163             fjx0             = _mm_add_ps(fjx0,tx);
1164             fjy0             = _mm_add_ps(fjy0,ty);
1165             fjz0             = _mm_add_ps(fjz0,tz);
1166
1167             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1168             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1169             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1170             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1171
1172             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1173
1174             /* Inner loop uses 118 flops */
1175         }
1176
1177         /* End of innermost loop */
1178
1179         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1180                                               f+i_coord_offset,fshift+i_shift_offset);
1181
1182         /* Increment number of inner iterations */
1183         inneriter                  += j_index_end - j_index_start;
1184
1185         /* Outer loop uses 18 flops */
1186     }
1187
1188     /* Increment number of outer iterations */
1189     outeriter        += nri;
1190
1191     /* Update outer/inner flops */
1192
1193     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1194 }