src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJ_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 136     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 137     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176         fix1             = _mm_setzero_ps();
 177         fiy1             = _mm_setzero_ps();
 178         fiz1             = _mm_setzero_ps();
 179         fix2             = _mm_setzero_ps();
 180         fiy2             = _mm_setzero_ps();
 181         fiz2             = _mm_setzero_ps();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm_setzero_ps();
 185         vvdwsum          = _mm_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198             j_coord_offsetC  = DIM*jnrC;
 199             j_coord_offsetD  = DIM*jnrD;
 200
 201             /* load j atom coordinates */
 202             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 203                                               x+j_coord_offsetC,x+j_coord_offsetD,
 204                                               &jx0,&jy0,&jz0);
 205
 206             /* Calculate displacement vector */
 207             dx00             = _mm_sub_ps(ix0,jx0);
 208             dy00             = _mm_sub_ps(iy0,jy0);
 209             dz00             = _mm_sub_ps(iz0,jz0);
 210             dx10             = _mm_sub_ps(ix1,jx0);
 211             dy10             = _mm_sub_ps(iy1,jy0);
 212             dz10             = _mm_sub_ps(iz1,jz0);
 213             dx20             = _mm_sub_ps(ix2,jx0);
 214             dy20             = _mm_sub_ps(iy2,jy0);
 215             dz20             = _mm_sub_ps(iz2,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 221
 222             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 223             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 224             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 225
 226             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 227             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 232                                                               charge+jnrC+0,charge+jnrD+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235             vdwjidx0C        = 2*vdwtype[jnrC+0];
 236             vdwjidx0D        = 2*vdwtype[jnrD+0];
 237
 238             fjx0             = _mm_setzero_ps();
 239             fjy0             = _mm_setzero_ps();
 240             fjz0             = _mm_setzero_ps();
 241
 242             /**************************
 243              * CALCULATE INTERACTIONS *
 244              **************************/
 245
 246             r00              = _mm_mul_ps(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             qq00             = _mm_mul_ps(iq0,jq0);
 250             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 251                                          vdwparam+vdwioffset0+vdwjidx0B,
 252                                          vdwparam+vdwioffset0+vdwjidx0C,
 253                                          vdwparam+vdwioffset0+vdwjidx0D,
 254                                          &c6_00,&c12_00);
 255
 256             /* EWALD ELECTROSTATICS */
 257
 258             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 259             ewrt             = _mm_mul_ps(r00,ewtabscale);
 260             ewitab           = _mm_cvttps_epi32(ewrt);
 261             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 262             ewitab           = _mm_slli_epi32(ewitab,2);
 263             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 264             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 265             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 266             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 267             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 268             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 269             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 270             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 271             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 272
 273             /* LENNARD-JONES DISPERSION/REPULSION */
 274
 275             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 276             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 277             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 278             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 279             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_ps(velecsum,velec);
 283             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 284
 285             fscal            = _mm_add_ps(felec,fvdw);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_ps(fscal,dx00);
 289             ty               = _mm_mul_ps(fscal,dy00);
 290             tz               = _mm_mul_ps(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_add_ps(fix0,tx);
 294             fiy0             = _mm_add_ps(fiy0,ty);
 295             fiz0             = _mm_add_ps(fiz0,tz);
 296
 297             fjx0             = _mm_add_ps(fjx0,tx);
 298             fjy0             = _mm_add_ps(fjy0,ty);
 299             fjz0             = _mm_add_ps(fjz0,tz);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             r10              = _mm_mul_ps(rsq10,rinv10);
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm_mul_ps(iq1,jq0);
 309
 310             /* EWALD ELECTROSTATICS */
 311
 312             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 313             ewrt             = _mm_mul_ps(r10,ewtabscale);
 314             ewitab           = _mm_cvttps_epi32(ewrt);
 315             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 316             ewitab           = _mm_slli_epi32(ewitab,2);
 317             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 318             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 319             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 320             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 321             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 322             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 323             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 324             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 325             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm_add_ps(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_ps(fscal,dx10);
 334             ty               = _mm_mul_ps(fscal,dy10);
 335             tz               = _mm_mul_ps(fscal,dz10);
 336
 337             /* Update vectorial force */
 338             fix1             = _mm_add_ps(fix1,tx);
 339             fiy1             = _mm_add_ps(fiy1,ty);
 340             fiz1             = _mm_add_ps(fiz1,tz);
 341
 342             fjx0             = _mm_add_ps(fjx0,tx);
 343             fjy0             = _mm_add_ps(fjy0,ty);
 344             fjz0             = _mm_add_ps(fjz0,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             r20              = _mm_mul_ps(rsq20,rinv20);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm_mul_ps(iq2,jq0);
 354
 355             /* EWALD ELECTROSTATICS */
 356
 357             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 358             ewrt             = _mm_mul_ps(r20,ewtabscale);
 359             ewitab           = _mm_cvttps_epi32(ewrt);
 360             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 361             ewitab           = _mm_slli_epi32(ewitab,2);
 362             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 363             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 364             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 365             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 366             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 367             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 368             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 369             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 370             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm_add_ps(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm_mul_ps(fscal,dx20);
 379             ty               = _mm_mul_ps(fscal,dy20);
 380             tz               = _mm_mul_ps(fscal,dz20);
 381
 382             /* Update vectorial force */
 383             fix2             = _mm_add_ps(fix2,tx);
 384             fiy2             = _mm_add_ps(fiy2,ty);
 385             fiz2             = _mm_add_ps(fiz2,tz);
 386
 387             fjx0             = _mm_add_ps(fjx0,tx);
 388             fjy0             = _mm_add_ps(fjy0,ty);
 389             fjz0             = _mm_add_ps(fjz0,tz);
 390
 391             fjptrA             = f+j_coord_offsetA;
 392             fjptrB             = f+j_coord_offsetB;
 393             fjptrC             = f+j_coord_offsetC;
 394             fjptrD             = f+j_coord_offsetD;
 395
 396             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 397
 398             /* Inner loop uses 135 flops */
 399         }
 400
 401         if(jidx<j_index_end)
 402         {
 403
 404             /* Get j neighbor index, and coordinate index */
 405             jnrlistA         = jjnr[jidx];
 406             jnrlistB         = jjnr[jidx+1];
 407             jnrlistC         = jjnr[jidx+2];
 408             jnrlistD         = jjnr[jidx+3];
 409             /* Sign of each element will be negative for non-real atoms.
 410              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 411              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 412              */
 413             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 414             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 415             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 416             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 417             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 418             j_coord_offsetA  = DIM*jnrA;
 419             j_coord_offsetB  = DIM*jnrB;
 420             j_coord_offsetC  = DIM*jnrC;
 421             j_coord_offsetD  = DIM*jnrD;
 422
 423             /* load j atom coordinates */
 424             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 425                                               x+j_coord_offsetC,x+j_coord_offsetD,
 426                                               &jx0,&jy0,&jz0);
 427
 428             /* Calculate displacement vector */
 429             dx00             = _mm_sub_ps(ix0,jx0);
 430             dy00             = _mm_sub_ps(iy0,jy0);
 431             dz00             = _mm_sub_ps(iz0,jz0);
 432             dx10             = _mm_sub_ps(ix1,jx0);
 433             dy10             = _mm_sub_ps(iy1,jy0);
 434             dz10             = _mm_sub_ps(iz1,jz0);
 435             dx20             = _mm_sub_ps(ix2,jx0);
 436             dy20             = _mm_sub_ps(iy2,jy0);
 437             dz20             = _mm_sub_ps(iz2,jz0);
 438
 439             /* Calculate squared distance and things based on it */
 440             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 441             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 442             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 443
 444             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 445             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 446             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 447
 448             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 449             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 450             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 451
 452             /* Load parameters for j particles */
 453             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 454                                                               charge+jnrC+0,charge+jnrD+0);
 455             vdwjidx0A        = 2*vdwtype[jnrA+0];
 456             vdwjidx0B        = 2*vdwtype[jnrB+0];
 457             vdwjidx0C        = 2*vdwtype[jnrC+0];
 458             vdwjidx0D        = 2*vdwtype[jnrD+0];
 459
 460             fjx0             = _mm_setzero_ps();
 461             fjy0             = _mm_setzero_ps();
 462             fjz0             = _mm_setzero_ps();
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r00              = _mm_mul_ps(rsq00,rinv00);
 469             r00              = _mm_andnot_ps(dummy_mask,r00);
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq00             = _mm_mul_ps(iq0,jq0);
 473             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 474                                          vdwparam+vdwioffset0+vdwjidx0B,
 475                                          vdwparam+vdwioffset0+vdwjidx0C,
 476                                          vdwparam+vdwioffset0+vdwjidx0D,
 477                                          &c6_00,&c12_00);
 478
 479             /* EWALD ELECTROSTATICS */
 480
 481             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 482             ewrt             = _mm_mul_ps(r00,ewtabscale);
 483             ewitab           = _mm_cvttps_epi32(ewrt);
 484             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 485             ewitab           = _mm_slli_epi32(ewitab,2);
 486             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 487             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 488             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 489             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 490             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 491             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 492             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 493             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 494             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 495
 496             /* LENNARD-JONES DISPERSION/REPULSION */
 497
 498             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 499             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 500             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 501             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 502             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velec            = _mm_andnot_ps(dummy_mask,velec);
 506             velecsum         = _mm_add_ps(velecsum,velec);
 507             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 508             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 509
 510             fscal            = _mm_add_ps(felec,fvdw);
 511
 512             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 513
 514             /* Calculate temporary vectorial force */
 515             tx               = _mm_mul_ps(fscal,dx00);
 516             ty               = _mm_mul_ps(fscal,dy00);
 517             tz               = _mm_mul_ps(fscal,dz00);
 518
 519             /* Update vectorial force */
 520             fix0             = _mm_add_ps(fix0,tx);
 521             fiy0             = _mm_add_ps(fiy0,ty);
 522             fiz0             = _mm_add_ps(fiz0,tz);
 523
 524             fjx0             = _mm_add_ps(fjx0,tx);
 525             fjy0             = _mm_add_ps(fjy0,ty);
 526             fjz0             = _mm_add_ps(fjz0,tz);
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             r10              = _mm_mul_ps(rsq10,rinv10);
 533             r10              = _mm_andnot_ps(dummy_mask,r10);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq10             = _mm_mul_ps(iq1,jq0);
 537
 538             /* EWALD ELECTROSTATICS */
 539
 540             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 541             ewrt             = _mm_mul_ps(r10,ewtabscale);
 542             ewitab           = _mm_cvttps_epi32(ewrt);
 543             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 544             ewitab           = _mm_slli_epi32(ewitab,2);
 545             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 546             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 547             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 548             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 549             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 550             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 551             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 552             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 553             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 554
 555             /* Update potential sum for this i atom from the interaction with this j atom. */
 556             velec            = _mm_andnot_ps(dummy_mask,velec);
 557             velecsum         = _mm_add_ps(velecsum,velec);
 558
 559             fscal            = felec;
 560
 561             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 562
 563             /* Calculate temporary vectorial force */
 564             tx               = _mm_mul_ps(fscal,dx10);
 565             ty               = _mm_mul_ps(fscal,dy10);
 566             tz               = _mm_mul_ps(fscal,dz10);
 567
 568             /* Update vectorial force */
 569             fix1             = _mm_add_ps(fix1,tx);
 570             fiy1             = _mm_add_ps(fiy1,ty);
 571             fiz1             = _mm_add_ps(fiz1,tz);
 572
 573             fjx0             = _mm_add_ps(fjx0,tx);
 574             fjy0             = _mm_add_ps(fjy0,ty);
 575             fjz0             = _mm_add_ps(fjz0,tz);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             r20              = _mm_mul_ps(rsq20,rinv20);
 582             r20              = _mm_andnot_ps(dummy_mask,r20);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq20             = _mm_mul_ps(iq2,jq0);
 586
 587             /* EWALD ELECTROSTATICS */
 588
 589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 590             ewrt             = _mm_mul_ps(r20,ewtabscale);
 591             ewitab           = _mm_cvttps_epi32(ewrt);
 592             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 593             ewitab           = _mm_slli_epi32(ewitab,2);
 594             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 595             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 596             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 597             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 598             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 599             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 600             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 601             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 602             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             velec            = _mm_andnot_ps(dummy_mask,velec);
 606             velecsum         = _mm_add_ps(velecsum,velec);
 607
 608             fscal            = felec;
 609
 610             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm_mul_ps(fscal,dx20);
 614             ty               = _mm_mul_ps(fscal,dy20);
 615             tz               = _mm_mul_ps(fscal,dz20);
 616
 617             /* Update vectorial force */
 618             fix2             = _mm_add_ps(fix2,tx);
 619             fiy2             = _mm_add_ps(fiy2,ty);
 620             fiz2             = _mm_add_ps(fiz2,tz);
 621
 622             fjx0             = _mm_add_ps(fjx0,tx);
 623             fjy0             = _mm_add_ps(fjy0,ty);
 624             fjz0             = _mm_add_ps(fjz0,tz);
 625
 626             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 627             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 628             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 629             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 630
 631             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 632
 633             /* Inner loop uses 138 flops */
 634         }
 635
 636         /* End of innermost loop */
 637
 638         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 639                                               f+i_coord_offset,fshift+i_shift_offset);
 640
 641         ggid                        = gid[iidx];
 642         /* Update potential energies */
 643         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 644         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 645
 646         /* Increment number of inner iterations */
 647         inneriter                  += j_index_end - j_index_start;
 648
 649         /* Outer loop uses 20 flops */
 650     }
 651
 652     /* Increment number of outer iterations */
 653     outeriter        += nri;
 654
 655     /* Update outer/inner flops */
 656
 657     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*138);
 658 }
 659 /*
 660  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse4_1_single
 661  * Electrostatics interaction: Ewald
 662  * VdW interaction:            LennardJones
 663  * Geometry:                   Water3-Particle
 664  * Calculate force/pot:        Force
 665  */
 666 void
 667 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse4_1_single
 668                     (t_nblist                    * gmx_restrict       nlist,
 669                      rvec                        * gmx_restrict          xx,
 670                      rvec                        * gmx_restrict          ff,
 671                      t_forcerec                  * gmx_restrict          fr,
 672                      t_mdatoms                   * gmx_restrict     mdatoms,
 673                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 674                      t_nrnb                      * gmx_restrict        nrnb)
 675 {
 676     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 677      * just 0 for non-waters.
 678      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 679      * jnr indices corresponding to data put in the four positions in the SIMD register.
 680      */
 681     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 682     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 683     int              jnrA,jnrB,jnrC,jnrD;
 684     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 685     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 686     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 687     real             rcutoff_scalar;
 688     real             *shiftvec,*fshift,*x,*f;
 689     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 690     real             scratch[4*DIM];
 691     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 692     int              vdwioffset0;
 693     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 694     int              vdwioffset1;
 695     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 696     int              vdwioffset2;
 697     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 698     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 699     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 700     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 701     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 702     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 703     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 704     real             *charge;
 705     int              nvdwtype;
 706     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 707     int              *vdwtype;
 708     real             *vdwparam;
 709     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 710     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 711     __m128i          ewitab;
 712     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 713     real             *ewtab;
 714     __m128           dummy_mask,cutoff_mask;
 715     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 716     __m128           one     = _mm_set1_ps(1.0);
 717     __m128           two     = _mm_set1_ps(2.0);
 718     x                = xx[0];
 719     f                = ff[0];
 720
 721     nri              = nlist->nri;
 722     iinr             = nlist->iinr;
 723     jindex           = nlist->jindex;
 724     jjnr             = nlist->jjnr;
 725     shiftidx         = nlist->shift;
 726     gid              = nlist->gid;
 727     shiftvec         = fr->shift_vec[0];
 728     fshift           = fr->fshift[0];
 729     facel            = _mm_set1_ps(fr->epsfac);
 730     charge           = mdatoms->chargeA;
 731     nvdwtype         = fr->ntype;
 732     vdwparam         = fr->nbfp;
 733     vdwtype          = mdatoms->typeA;
 734
 735     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 736     ewtab            = fr->ic->tabq_coul_F;
 737     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 738     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 739
 740     /* Setup water-specific parameters */
 741     inr              = nlist->iinr[0];
 742     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 743     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 744     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 745     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 746
 747     /* Avoid stupid compiler warnings */
 748     jnrA = jnrB = jnrC = jnrD = 0;
 749     j_coord_offsetA = 0;
 750     j_coord_offsetB = 0;
 751     j_coord_offsetC = 0;
 752     j_coord_offsetD = 0;
 753
 754     outeriter        = 0;
 755     inneriter        = 0;
 756
 757     for(iidx=0;iidx<4*DIM;iidx++)
 758     {
 759         scratch[iidx] = 0.0;
 760     }
 761
 762     /* Start outer loop over neighborlists */
 763     for(iidx=0; iidx<nri; iidx++)
 764     {
 765         /* Load shift vector for this list */
 766         i_shift_offset   = DIM*shiftidx[iidx];
 767
 768         /* Load limits for loop over neighbors */
 769         j_index_start    = jindex[iidx];
 770         j_index_end      = jindex[iidx+1];
 771
 772         /* Get outer coordinate index */
 773         inr              = iinr[iidx];
 774         i_coord_offset   = DIM*inr;
 775
 776         /* Load i particle coords and add shift vector */
 777         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 778                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 779
 780         fix0             = _mm_setzero_ps();
 781         fiy0             = _mm_setzero_ps();
 782         fiz0             = _mm_setzero_ps();
 783         fix1             = _mm_setzero_ps();
 784         fiy1             = _mm_setzero_ps();
 785         fiz1             = _mm_setzero_ps();
 786         fix2             = _mm_setzero_ps();
 787         fiy2             = _mm_setzero_ps();
 788         fiz2             = _mm_setzero_ps();
 789
 790         /* Start inner kernel loop */
 791         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 792         {
 793
 794             /* Get j neighbor index, and coordinate index */
 795             jnrA             = jjnr[jidx];
 796             jnrB             = jjnr[jidx+1];
 797             jnrC             = jjnr[jidx+2];
 798             jnrD             = jjnr[jidx+3];
 799             j_coord_offsetA  = DIM*jnrA;
 800             j_coord_offsetB  = DIM*jnrB;
 801             j_coord_offsetC  = DIM*jnrC;
 802             j_coord_offsetD  = DIM*jnrD;
 803
 804             /* load j atom coordinates */
 805             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 806                                               x+j_coord_offsetC,x+j_coord_offsetD,
 807                                               &jx0,&jy0,&jz0);
 808
 809             /* Calculate displacement vector */
 810             dx00             = _mm_sub_ps(ix0,jx0);
 811             dy00             = _mm_sub_ps(iy0,jy0);
 812             dz00             = _mm_sub_ps(iz0,jz0);
 813             dx10             = _mm_sub_ps(ix1,jx0);
 814             dy10             = _mm_sub_ps(iy1,jy0);
 815             dz10             = _mm_sub_ps(iz1,jz0);
 816             dx20             = _mm_sub_ps(ix2,jx0);
 817             dy20             = _mm_sub_ps(iy2,jy0);
 818             dz20             = _mm_sub_ps(iz2,jz0);
 819
 820             /* Calculate squared distance and things based on it */
 821             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 822             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 823             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 824
 825             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 826             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 827             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 828
 829             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 830             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 831             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 832
 833             /* Load parameters for j particles */
 834             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 835                                                               charge+jnrC+0,charge+jnrD+0);
 836             vdwjidx0A        = 2*vdwtype[jnrA+0];
 837             vdwjidx0B        = 2*vdwtype[jnrB+0];
 838             vdwjidx0C        = 2*vdwtype[jnrC+0];
 839             vdwjidx0D        = 2*vdwtype[jnrD+0];
 840
 841             fjx0             = _mm_setzero_ps();
 842             fjy0             = _mm_setzero_ps();
 843             fjz0             = _mm_setzero_ps();
 844
 845             /**************************
 846              * CALCULATE INTERACTIONS *
 847              **************************/
 848
 849             r00              = _mm_mul_ps(rsq00,rinv00);
 850
 851             /* Compute parameters for interactions between i and j atoms */
 852             qq00             = _mm_mul_ps(iq0,jq0);
 853             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 854                                          vdwparam+vdwioffset0+vdwjidx0B,
 855                                          vdwparam+vdwioffset0+vdwjidx0C,
 856                                          vdwparam+vdwioffset0+vdwjidx0D,
 857                                          &c6_00,&c12_00);
 858
 859             /* EWALD ELECTROSTATICS */
 860
 861             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 862             ewrt             = _mm_mul_ps(r00,ewtabscale);
 863             ewitab           = _mm_cvttps_epi32(ewrt);
 864             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 865             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 866                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 867                                          &ewtabF,&ewtabFn);
 868             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 869             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 870
 871             /* LENNARD-JONES DISPERSION/REPULSION */
 872
 873             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 874             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 875
 876             fscal            = _mm_add_ps(felec,fvdw);
 877
 878             /* Calculate temporary vectorial force */
 879             tx               = _mm_mul_ps(fscal,dx00);
 880             ty               = _mm_mul_ps(fscal,dy00);
 881             tz               = _mm_mul_ps(fscal,dz00);
 882
 883             /* Update vectorial force */
 884             fix0             = _mm_add_ps(fix0,tx);
 885             fiy0             = _mm_add_ps(fiy0,ty);
 886             fiz0             = _mm_add_ps(fiz0,tz);
 887
 888             fjx0             = _mm_add_ps(fjx0,tx);
 889             fjy0             = _mm_add_ps(fjy0,ty);
 890             fjz0             = _mm_add_ps(fjz0,tz);
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             r10              = _mm_mul_ps(rsq10,rinv10);
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             qq10             = _mm_mul_ps(iq1,jq0);
 900
 901             /* EWALD ELECTROSTATICS */
 902
 903             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 904             ewrt             = _mm_mul_ps(r10,ewtabscale);
 905             ewitab           = _mm_cvttps_epi32(ewrt);
 906             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 907             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 908                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 909                                          &ewtabF,&ewtabFn);
 910             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 911             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 912
 913             fscal            = felec;
 914
 915             /* Calculate temporary vectorial force */
 916             tx               = _mm_mul_ps(fscal,dx10);
 917             ty               = _mm_mul_ps(fscal,dy10);
 918             tz               = _mm_mul_ps(fscal,dz10);
 919
 920             /* Update vectorial force */
 921             fix1             = _mm_add_ps(fix1,tx);
 922             fiy1             = _mm_add_ps(fiy1,ty);
 923             fiz1             = _mm_add_ps(fiz1,tz);
 924
 925             fjx0             = _mm_add_ps(fjx0,tx);
 926             fjy0             = _mm_add_ps(fjy0,ty);
 927             fjz0             = _mm_add_ps(fjz0,tz);
 928
 929             /**************************
 930              * CALCULATE INTERACTIONS *
 931              **************************/
 932
 933             r20              = _mm_mul_ps(rsq20,rinv20);
 934
 935             /* Compute parameters for interactions between i and j atoms */
 936             qq20             = _mm_mul_ps(iq2,jq0);
 937
 938             /* EWALD ELECTROSTATICS */
 939
 940             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 941             ewrt             = _mm_mul_ps(r20,ewtabscale);
 942             ewitab           = _mm_cvttps_epi32(ewrt);
 943             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 944             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 945                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 946                                          &ewtabF,&ewtabFn);
 947             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 948             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 949
 950             fscal            = felec;
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm_mul_ps(fscal,dx20);
 954             ty               = _mm_mul_ps(fscal,dy20);
 955             tz               = _mm_mul_ps(fscal,dz20);
 956
 957             /* Update vectorial force */
 958             fix2             = _mm_add_ps(fix2,tx);
 959             fiy2             = _mm_add_ps(fiy2,ty);
 960             fiz2             = _mm_add_ps(fiz2,tz);
 961
 962             fjx0             = _mm_add_ps(fjx0,tx);
 963             fjy0             = _mm_add_ps(fjy0,ty);
 964             fjz0             = _mm_add_ps(fjz0,tz);
 965
 966             fjptrA             = f+j_coord_offsetA;
 967             fjptrB             = f+j_coord_offsetB;
 968             fjptrC             = f+j_coord_offsetC;
 969             fjptrD             = f+j_coord_offsetD;
 970
 971             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 972
 973             /* Inner loop uses 115 flops */
 974         }
 975
 976         if(jidx<j_index_end)
 977         {
 978
 979             /* Get j neighbor index, and coordinate index */
 980             jnrlistA         = jjnr[jidx];
 981             jnrlistB         = jjnr[jidx+1];
 982             jnrlistC         = jjnr[jidx+2];
 983             jnrlistD         = jjnr[jidx+3];
 984             /* Sign of each element will be negative for non-real atoms.
 985              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 986              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 987              */
 988             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 989             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 990             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 991             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 992             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 993             j_coord_offsetA  = DIM*jnrA;
 994             j_coord_offsetB  = DIM*jnrB;
 995             j_coord_offsetC  = DIM*jnrC;
 996             j_coord_offsetD  = DIM*jnrD;
 997
 998             /* load j atom coordinates */
 999             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1000                                               x+j_coord_offsetC,x+j_coord_offsetD,
1001                                               &jx0,&jy0,&jz0);
1002
1003             /* Calculate displacement vector */
1004             dx00             = _mm_sub_ps(ix0,jx0);
1005             dy00             = _mm_sub_ps(iy0,jy0);
1006             dz00             = _mm_sub_ps(iz0,jz0);
1007             dx10             = _mm_sub_ps(ix1,jx0);
1008             dy10             = _mm_sub_ps(iy1,jy0);
1009             dz10             = _mm_sub_ps(iz1,jz0);
1010             dx20             = _mm_sub_ps(ix2,jx0);
1011             dy20             = _mm_sub_ps(iy2,jy0);
1012             dz20             = _mm_sub_ps(iz2,jz0);
1013
1014             /* Calculate squared distance and things based on it */
1015             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1016             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1017             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1018
1019             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1020             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1021             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1022
1023             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1024             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1025             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1026
1027             /* Load parameters for j particles */
1028             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1029                                                               charge+jnrC+0,charge+jnrD+0);
1030             vdwjidx0A        = 2*vdwtype[jnrA+0];
1031             vdwjidx0B        = 2*vdwtype[jnrB+0];
1032             vdwjidx0C        = 2*vdwtype[jnrC+0];
1033             vdwjidx0D        = 2*vdwtype[jnrD+0];
1034
1035             fjx0             = _mm_setzero_ps();
1036             fjy0             = _mm_setzero_ps();
1037             fjz0             = _mm_setzero_ps();
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             r00              = _mm_mul_ps(rsq00,rinv00);
1044             r00              = _mm_andnot_ps(dummy_mask,r00);
1045
1046             /* Compute parameters for interactions between i and j atoms */
1047             qq00             = _mm_mul_ps(iq0,jq0);
1048             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1049                                          vdwparam+vdwioffset0+vdwjidx0B,
1050                                          vdwparam+vdwioffset0+vdwjidx0C,
1051                                          vdwparam+vdwioffset0+vdwjidx0D,
1052                                          &c6_00,&c12_00);
1053
1054             /* EWALD ELECTROSTATICS */
1055
1056             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1057             ewrt             = _mm_mul_ps(r00,ewtabscale);
1058             ewitab           = _mm_cvttps_epi32(ewrt);
1059             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1060             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1061                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1062                                          &ewtabF,&ewtabFn);
1063             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1064             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1065
1066             /* LENNARD-JONES DISPERSION/REPULSION */
1067
1068             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1069             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1070
1071             fscal            = _mm_add_ps(felec,fvdw);
1072
1073             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1074
1075             /* Calculate temporary vectorial force */
1076             tx               = _mm_mul_ps(fscal,dx00);
1077             ty               = _mm_mul_ps(fscal,dy00);
1078             tz               = _mm_mul_ps(fscal,dz00);
1079
1080             /* Update vectorial force */
1081             fix0             = _mm_add_ps(fix0,tx);
1082             fiy0             = _mm_add_ps(fiy0,ty);
1083             fiz0             = _mm_add_ps(fiz0,tz);
1084
1085             fjx0             = _mm_add_ps(fjx0,tx);
1086             fjy0             = _mm_add_ps(fjy0,ty);
1087             fjz0             = _mm_add_ps(fjz0,tz);
1088
1089             /**************************
1090              * CALCULATE INTERACTIONS *
1091              **************************/
1092
1093             r10              = _mm_mul_ps(rsq10,rinv10);
1094             r10              = _mm_andnot_ps(dummy_mask,r10);
1095
1096             /* Compute parameters for interactions between i and j atoms */
1097             qq10             = _mm_mul_ps(iq1,jq0);
1098
1099             /* EWALD ELECTROSTATICS */
1100
1101             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1102             ewrt             = _mm_mul_ps(r10,ewtabscale);
1103             ewitab           = _mm_cvttps_epi32(ewrt);
1104             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1105             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1106                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1107                                          &ewtabF,&ewtabFn);
1108             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1109             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1110
1111             fscal            = felec;
1112
1113             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1114
1115             /* Calculate temporary vectorial force */
1116             tx               = _mm_mul_ps(fscal,dx10);
1117             ty               = _mm_mul_ps(fscal,dy10);
1118             tz               = _mm_mul_ps(fscal,dz10);
1119
1120             /* Update vectorial force */
1121             fix1             = _mm_add_ps(fix1,tx);
1122             fiy1             = _mm_add_ps(fiy1,ty);
1123             fiz1             = _mm_add_ps(fiz1,tz);
1124
1125             fjx0             = _mm_add_ps(fjx0,tx);
1126             fjy0             = _mm_add_ps(fjy0,ty);
1127             fjz0             = _mm_add_ps(fjz0,tz);
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             r20              = _mm_mul_ps(rsq20,rinv20);
1134             r20              = _mm_andnot_ps(dummy_mask,r20);
1135
1136             /* Compute parameters for interactions between i and j atoms */
1137             qq20             = _mm_mul_ps(iq2,jq0);
1138
1139             /* EWALD ELECTROSTATICS */
1140
1141             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1142             ewrt             = _mm_mul_ps(r20,ewtabscale);
1143             ewitab           = _mm_cvttps_epi32(ewrt);
1144             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1145             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1146                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1147                                          &ewtabF,&ewtabFn);
1148             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1149             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1150
1151             fscal            = felec;
1152
1153             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1154
1155             /* Calculate temporary vectorial force */
1156             tx               = _mm_mul_ps(fscal,dx20);
1157             ty               = _mm_mul_ps(fscal,dy20);
1158             tz               = _mm_mul_ps(fscal,dz20);
1159
1160             /* Update vectorial force */
1161             fix2             = _mm_add_ps(fix2,tx);
1162             fiy2             = _mm_add_ps(fiy2,ty);
1163             fiz2             = _mm_add_ps(fiz2,tz);
1164
1165             fjx0             = _mm_add_ps(fjx0,tx);
1166             fjy0             = _mm_add_ps(fjy0,ty);
1167             fjz0             = _mm_add_ps(fjz0,tz);
1168
1169             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1170             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1171             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1172             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1173
1174             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1175
1176             /* Inner loop uses 118 flops */
1177         }
1178
1179         /* End of innermost loop */
1180
1181         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1182                                               f+i_coord_offset,fshift+i_shift_offset);
1183
1184         /* Increment number of inner iterations */
1185         inneriter                  += j_index_end - j_index_start;
1186
1187         /* Outer loop uses 18 flops */
1188     }
1189
1190     /* Increment number of outer iterations */
1191     outeriter        += nri;
1192
1193     /* Update outer/inner flops */
1194
1195     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1196 }