src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwNone_GeomW4P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "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_VdwNone_GeomW4P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomW4P1_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              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     __m128i          ewitab;
  99     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     __m128           dummy_mask,cutoff_mask;
 102     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 103     __m128           one     = _mm_set1_ps(1.0);
 104     __m128           two     = _mm_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_ps(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118
 119     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 127     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 128     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 161                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 162
 163         fix1             = _mm_setzero_ps();
 164         fiy1             = _mm_setzero_ps();
 165         fiz1             = _mm_setzero_ps();
 166         fix2             = _mm_setzero_ps();
 167         fiy2             = _mm_setzero_ps();
 168         fiz2             = _mm_setzero_ps();
 169         fix3             = _mm_setzero_ps();
 170         fiy3             = _mm_setzero_ps();
 171         fiz3             = _mm_setzero_ps();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               x+j_coord_offsetC,x+j_coord_offsetD,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx10             = _mm_sub_ps(ix1,jx0);
 197             dy10             = _mm_sub_ps(iy1,jy0);
 198             dz10             = _mm_sub_ps(iz1,jz0);
 199             dx20             = _mm_sub_ps(ix2,jx0);
 200             dy20             = _mm_sub_ps(iy2,jy0);
 201             dz20             = _mm_sub_ps(iz2,jz0);
 202             dx30             = _mm_sub_ps(ix3,jx0);
 203             dy30             = _mm_sub_ps(iy3,jy0);
 204             dz30             = _mm_sub_ps(iz3,jz0);
 205
 206             /* Calculate squared distance and things based on it */
 207             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 208             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 209             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 210
 211             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 212             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 213             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 214
 215             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 216             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 217             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 221                                                               charge+jnrC+0,charge+jnrD+0);
 222
 223             fjx0             = _mm_setzero_ps();
 224             fjy0             = _mm_setzero_ps();
 225             fjz0             = _mm_setzero_ps();
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             r10              = _mm_mul_ps(rsq10,rinv10);
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             qq10             = _mm_mul_ps(iq1,jq0);
 235
 236             /* EWALD ELECTROSTATICS */
 237
 238             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 239             ewrt             = _mm_mul_ps(r10,ewtabscale);
 240             ewitab           = _mm_cvttps_epi32(ewrt);
 241             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 242             ewitab           = _mm_slli_epi32(ewitab,2);
 243             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 244             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 245             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 246             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 247             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 248             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 249             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 250             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 251             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 252
 253             /* Update potential sum for this i atom from the interaction with this j atom. */
 254             velecsum         = _mm_add_ps(velecsum,velec);
 255
 256             fscal            = felec;
 257
 258             /* Calculate temporary vectorial force */
 259             tx               = _mm_mul_ps(fscal,dx10);
 260             ty               = _mm_mul_ps(fscal,dy10);
 261             tz               = _mm_mul_ps(fscal,dz10);
 262
 263             /* Update vectorial force */
 264             fix1             = _mm_add_ps(fix1,tx);
 265             fiy1             = _mm_add_ps(fiy1,ty);
 266             fiz1             = _mm_add_ps(fiz1,tz);
 267
 268             fjx0             = _mm_add_ps(fjx0,tx);
 269             fjy0             = _mm_add_ps(fjy0,ty);
 270             fjz0             = _mm_add_ps(fjz0,tz);
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             r20              = _mm_mul_ps(rsq20,rinv20);
 277
 278             /* Compute parameters for interactions between i and j atoms */
 279             qq20             = _mm_mul_ps(iq2,jq0);
 280
 281             /* EWALD ELECTROSTATICS */
 282
 283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 284             ewrt             = _mm_mul_ps(r20,ewtabscale);
 285             ewitab           = _mm_cvttps_epi32(ewrt);
 286             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 287             ewitab           = _mm_slli_epi32(ewitab,2);
 288             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 289             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 290             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 291             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 292             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 293             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 294             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 295             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 296             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velecsum         = _mm_add_ps(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm_mul_ps(fscal,dx20);
 305             ty               = _mm_mul_ps(fscal,dy20);
 306             tz               = _mm_mul_ps(fscal,dz20);
 307
 308             /* Update vectorial force */
 309             fix2             = _mm_add_ps(fix2,tx);
 310             fiy2             = _mm_add_ps(fiy2,ty);
 311             fiz2             = _mm_add_ps(fiz2,tz);
 312
 313             fjx0             = _mm_add_ps(fjx0,tx);
 314             fjy0             = _mm_add_ps(fjy0,ty);
 315             fjz0             = _mm_add_ps(fjz0,tz);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r30              = _mm_mul_ps(rsq30,rinv30);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq30             = _mm_mul_ps(iq3,jq0);
 325
 326             /* EWALD ELECTROSTATICS */
 327
 328             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 329             ewrt             = _mm_mul_ps(r30,ewtabscale);
 330             ewitab           = _mm_cvttps_epi32(ewrt);
 331             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 332             ewitab           = _mm_slli_epi32(ewitab,2);
 333             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 334             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 335             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 336             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 337             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 338             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 339             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 340             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 341             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_ps(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_ps(fscal,dx30);
 350             ty               = _mm_mul_ps(fscal,dy30);
 351             tz               = _mm_mul_ps(fscal,dz30);
 352
 353             /* Update vectorial force */
 354             fix3             = _mm_add_ps(fix3,tx);
 355             fiy3             = _mm_add_ps(fiy3,ty);
 356             fiz3             = _mm_add_ps(fiz3,tz);
 357
 358             fjx0             = _mm_add_ps(fjx0,tx);
 359             fjy0             = _mm_add_ps(fjy0,ty);
 360             fjz0             = _mm_add_ps(fjz0,tz);
 361
 362             fjptrA             = f+j_coord_offsetA;
 363             fjptrB             = f+j_coord_offsetB;
 364             fjptrC             = f+j_coord_offsetC;
 365             fjptrD             = f+j_coord_offsetD;
 366
 367             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 368
 369             /* Inner loop uses 123 flops */
 370         }
 371
 372         if(jidx<j_index_end)
 373         {
 374
 375             /* Get j neighbor index, and coordinate index */
 376             jnrlistA         = jjnr[jidx];
 377             jnrlistB         = jjnr[jidx+1];
 378             jnrlistC         = jjnr[jidx+2];
 379             jnrlistD         = jjnr[jidx+3];
 380             /* Sign of each element will be negative for non-real atoms.
 381              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 382              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 383              */
 384             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 385             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 386             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 387             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 388             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 389             j_coord_offsetA  = DIM*jnrA;
 390             j_coord_offsetB  = DIM*jnrB;
 391             j_coord_offsetC  = DIM*jnrC;
 392             j_coord_offsetD  = DIM*jnrD;
 393
 394             /* load j atom coordinates */
 395             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 396                                               x+j_coord_offsetC,x+j_coord_offsetD,
 397                                               &jx0,&jy0,&jz0);
 398
 399             /* Calculate displacement vector */
 400             dx10             = _mm_sub_ps(ix1,jx0);
 401             dy10             = _mm_sub_ps(iy1,jy0);
 402             dz10             = _mm_sub_ps(iz1,jz0);
 403             dx20             = _mm_sub_ps(ix2,jx0);
 404             dy20             = _mm_sub_ps(iy2,jy0);
 405             dz20             = _mm_sub_ps(iz2,jz0);
 406             dx30             = _mm_sub_ps(ix3,jx0);
 407             dy30             = _mm_sub_ps(iy3,jy0);
 408             dz30             = _mm_sub_ps(iz3,jz0);
 409
 410             /* Calculate squared distance and things based on it */
 411             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 412             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 413             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 414
 415             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 416             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 417             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 418
 419             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 420             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 421             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 422
 423             /* Load parameters for j particles */
 424             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 425                                                               charge+jnrC+0,charge+jnrD+0);
 426
 427             fjx0             = _mm_setzero_ps();
 428             fjy0             = _mm_setzero_ps();
 429             fjz0             = _mm_setzero_ps();
 430
 431             /**************************
 432              * CALCULATE INTERACTIONS *
 433              **************************/
 434
 435             r10              = _mm_mul_ps(rsq10,rinv10);
 436             r10              = _mm_andnot_ps(dummy_mask,r10);
 437
 438             /* Compute parameters for interactions between i and j atoms */
 439             qq10             = _mm_mul_ps(iq1,jq0);
 440
 441             /* EWALD ELECTROSTATICS */
 442
 443             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 444             ewrt             = _mm_mul_ps(r10,ewtabscale);
 445             ewitab           = _mm_cvttps_epi32(ewrt);
 446             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 447             ewitab           = _mm_slli_epi32(ewitab,2);
 448             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 449             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 450             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 451             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 452             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 453             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 454             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 455             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 456             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm_andnot_ps(dummy_mask,velec);
 460             velecsum         = _mm_add_ps(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm_mul_ps(fscal,dx10);
 468             ty               = _mm_mul_ps(fscal,dy10);
 469             tz               = _mm_mul_ps(fscal,dz10);
 470
 471             /* Update vectorial force */
 472             fix1             = _mm_add_ps(fix1,tx);
 473             fiy1             = _mm_add_ps(fiy1,ty);
 474             fiz1             = _mm_add_ps(fiz1,tz);
 475
 476             fjx0             = _mm_add_ps(fjx0,tx);
 477             fjy0             = _mm_add_ps(fjy0,ty);
 478             fjz0             = _mm_add_ps(fjz0,tz);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r20              = _mm_mul_ps(rsq20,rinv20);
 485             r20              = _mm_andnot_ps(dummy_mask,r20);
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq20             = _mm_mul_ps(iq2,jq0);
 489
 490             /* EWALD ELECTROSTATICS */
 491
 492             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 493             ewrt             = _mm_mul_ps(r20,ewtabscale);
 494             ewitab           = _mm_cvttps_epi32(ewrt);
 495             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 496             ewitab           = _mm_slli_epi32(ewitab,2);
 497             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 498             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 499             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 500             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 501             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 502             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 503             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 504             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 505             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 506
 507             /* Update potential sum for this i atom from the interaction with this j atom. */
 508             velec            = _mm_andnot_ps(dummy_mask,velec);
 509             velecsum         = _mm_add_ps(velecsum,velec);
 510
 511             fscal            = felec;
 512
 513             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 514
 515             /* Calculate temporary vectorial force */
 516             tx               = _mm_mul_ps(fscal,dx20);
 517             ty               = _mm_mul_ps(fscal,dy20);
 518             tz               = _mm_mul_ps(fscal,dz20);
 519
 520             /* Update vectorial force */
 521             fix2             = _mm_add_ps(fix2,tx);
 522             fiy2             = _mm_add_ps(fiy2,ty);
 523             fiz2             = _mm_add_ps(fiz2,tz);
 524
 525             fjx0             = _mm_add_ps(fjx0,tx);
 526             fjy0             = _mm_add_ps(fjy0,ty);
 527             fjz0             = _mm_add_ps(fjz0,tz);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r30              = _mm_mul_ps(rsq30,rinv30);
 534             r30              = _mm_andnot_ps(dummy_mask,r30);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq30             = _mm_mul_ps(iq3,jq0);
 538
 539             /* EWALD ELECTROSTATICS */
 540
 541             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 542             ewrt             = _mm_mul_ps(r30,ewtabscale);
 543             ewitab           = _mm_cvttps_epi32(ewrt);
 544             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 545             ewitab           = _mm_slli_epi32(ewitab,2);
 546             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 547             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 548             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 549             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 550             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 551             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 552             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 553             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 554             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_andnot_ps(dummy_mask,velec);
 558             velecsum         = _mm_add_ps(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_ps(fscal,dx30);
 566             ty               = _mm_mul_ps(fscal,dy30);
 567             tz               = _mm_mul_ps(fscal,dz30);
 568
 569             /* Update vectorial force */
 570             fix3             = _mm_add_ps(fix3,tx);
 571             fiy3             = _mm_add_ps(fiy3,ty);
 572             fiz3             = _mm_add_ps(fiz3,tz);
 573
 574             fjx0             = _mm_add_ps(fjx0,tx);
 575             fjy0             = _mm_add_ps(fjy0,ty);
 576             fjz0             = _mm_add_ps(fjz0,tz);
 577
 578             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 579             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 580             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 581             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 582
 583             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 584
 585             /* Inner loop uses 126 flops */
 586         }
 587
 588         /* End of innermost loop */
 589
 590         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 591                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 592
 593         ggid                        = gid[iidx];
 594         /* Update potential energies */
 595         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 596
 597         /* Increment number of inner iterations */
 598         inneriter                  += j_index_end - j_index_start;
 599
 600         /* Outer loop uses 19 flops */
 601     }
 602
 603     /* Increment number of outer iterations */
 604     outeriter        += nri;
 605
 606     /* Update outer/inner flops */
 607
 608     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*126);
 609 }
 610 /*
 611  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse4_1_single
 612  * Electrostatics interaction: Ewald
 613  * VdW interaction:            None
 614  * Geometry:                   Water4-Particle
 615  * Calculate force/pot:        Force
 616  */
 617 void
 618 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse4_1_single
 619                     (t_nblist                    * gmx_restrict       nlist,
 620                      rvec                        * gmx_restrict          xx,
 621                      rvec                        * gmx_restrict          ff,
 622                      t_forcerec                  * gmx_restrict          fr,
 623                      t_mdatoms                   * gmx_restrict     mdatoms,
 624                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 625                      t_nrnb                      * gmx_restrict        nrnb)
 626 {
 627     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 628      * just 0 for non-waters.
 629      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 630      * jnr indices corresponding to data put in the four positions in the SIMD register.
 631      */
 632     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 633     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 634     int              jnrA,jnrB,jnrC,jnrD;
 635     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 636     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 637     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 638     real             rcutoff_scalar;
 639     real             *shiftvec,*fshift,*x,*f;
 640     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 641     real             scratch[4*DIM];
 642     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 643     int              vdwioffset1;
 644     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 645     int              vdwioffset2;
 646     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 647     int              vdwioffset3;
 648     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 649     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 650     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 651     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 652     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 653     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 654     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 655     real             *charge;
 656     __m128i          ewitab;
 657     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 658     real             *ewtab;
 659     __m128           dummy_mask,cutoff_mask;
 660     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 661     __m128           one     = _mm_set1_ps(1.0);
 662     __m128           two     = _mm_set1_ps(2.0);
 663     x                = xx[0];
 664     f                = ff[0];
 665
 666     nri              = nlist->nri;
 667     iinr             = nlist->iinr;
 668     jindex           = nlist->jindex;
 669     jjnr             = nlist->jjnr;
 670     shiftidx         = nlist->shift;
 671     gid              = nlist->gid;
 672     shiftvec         = fr->shift_vec[0];
 673     fshift           = fr->fshift[0];
 674     facel            = _mm_set1_ps(fr->epsfac);
 675     charge           = mdatoms->chargeA;
 676
 677     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 678     ewtab            = fr->ic->tabq_coul_F;
 679     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 680     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 681
 682     /* Setup water-specific parameters */
 683     inr              = nlist->iinr[0];
 684     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 685     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 686     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 687
 688     /* Avoid stupid compiler warnings */
 689     jnrA = jnrB = jnrC = jnrD = 0;
 690     j_coord_offsetA = 0;
 691     j_coord_offsetB = 0;
 692     j_coord_offsetC = 0;
 693     j_coord_offsetD = 0;
 694
 695     outeriter        = 0;
 696     inneriter        = 0;
 697
 698     for(iidx=0;iidx<4*DIM;iidx++)
 699     {
 700         scratch[iidx] = 0.0;
 701     }
 702
 703     /* Start outer loop over neighborlists */
 704     for(iidx=0; iidx<nri; iidx++)
 705     {
 706         /* Load shift vector for this list */
 707         i_shift_offset   = DIM*shiftidx[iidx];
 708
 709         /* Load limits for loop over neighbors */
 710         j_index_start    = jindex[iidx];
 711         j_index_end      = jindex[iidx+1];
 712
 713         /* Get outer coordinate index */
 714         inr              = iinr[iidx];
 715         i_coord_offset   = DIM*inr;
 716
 717         /* Load i particle coords and add shift vector */
 718         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 719                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 720
 721         fix1             = _mm_setzero_ps();
 722         fiy1             = _mm_setzero_ps();
 723         fiz1             = _mm_setzero_ps();
 724         fix2             = _mm_setzero_ps();
 725         fiy2             = _mm_setzero_ps();
 726         fiz2             = _mm_setzero_ps();
 727         fix3             = _mm_setzero_ps();
 728         fiy3             = _mm_setzero_ps();
 729         fiz3             = _mm_setzero_ps();
 730
 731         /* Start inner kernel loop */
 732         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 733         {
 734
 735             /* Get j neighbor index, and coordinate index */
 736             jnrA             = jjnr[jidx];
 737             jnrB             = jjnr[jidx+1];
 738             jnrC             = jjnr[jidx+2];
 739             jnrD             = jjnr[jidx+3];
 740             j_coord_offsetA  = DIM*jnrA;
 741             j_coord_offsetB  = DIM*jnrB;
 742             j_coord_offsetC  = DIM*jnrC;
 743             j_coord_offsetD  = DIM*jnrD;
 744
 745             /* load j atom coordinates */
 746             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 747                                               x+j_coord_offsetC,x+j_coord_offsetD,
 748                                               &jx0,&jy0,&jz0);
 749
 750             /* Calculate displacement vector */
 751             dx10             = _mm_sub_ps(ix1,jx0);
 752             dy10             = _mm_sub_ps(iy1,jy0);
 753             dz10             = _mm_sub_ps(iz1,jz0);
 754             dx20             = _mm_sub_ps(ix2,jx0);
 755             dy20             = _mm_sub_ps(iy2,jy0);
 756             dz20             = _mm_sub_ps(iz2,jz0);
 757             dx30             = _mm_sub_ps(ix3,jx0);
 758             dy30             = _mm_sub_ps(iy3,jy0);
 759             dz30             = _mm_sub_ps(iz3,jz0);
 760
 761             /* Calculate squared distance and things based on it */
 762             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 763             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 764             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 765
 766             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 767             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 768             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 769
 770             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 771             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 772             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 773
 774             /* Load parameters for j particles */
 775             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 776                                                               charge+jnrC+0,charge+jnrD+0);
 777
 778             fjx0             = _mm_setzero_ps();
 779             fjy0             = _mm_setzero_ps();
 780             fjz0             = _mm_setzero_ps();
 781
 782             /**************************
 783              * CALCULATE INTERACTIONS *
 784              **************************/
 785
 786             r10              = _mm_mul_ps(rsq10,rinv10);
 787
 788             /* Compute parameters for interactions between i and j atoms */
 789             qq10             = _mm_mul_ps(iq1,jq0);
 790
 791             /* EWALD ELECTROSTATICS */
 792
 793             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 794             ewrt             = _mm_mul_ps(r10,ewtabscale);
 795             ewitab           = _mm_cvttps_epi32(ewrt);
 796             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 797             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 798                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 799                                          &ewtabF,&ewtabFn);
 800             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 801             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 802
 803             fscal            = felec;
 804
 805             /* Calculate temporary vectorial force */
 806             tx               = _mm_mul_ps(fscal,dx10);
 807             ty               = _mm_mul_ps(fscal,dy10);
 808             tz               = _mm_mul_ps(fscal,dz10);
 809
 810             /* Update vectorial force */
 811             fix1             = _mm_add_ps(fix1,tx);
 812             fiy1             = _mm_add_ps(fiy1,ty);
 813             fiz1             = _mm_add_ps(fiz1,tz);
 814
 815             fjx0             = _mm_add_ps(fjx0,tx);
 816             fjy0             = _mm_add_ps(fjy0,ty);
 817             fjz0             = _mm_add_ps(fjz0,tz);
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             r20              = _mm_mul_ps(rsq20,rinv20);
 824
 825             /* Compute parameters for interactions between i and j atoms */
 826             qq20             = _mm_mul_ps(iq2,jq0);
 827
 828             /* EWALD ELECTROSTATICS */
 829
 830             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 831             ewrt             = _mm_mul_ps(r20,ewtabscale);
 832             ewitab           = _mm_cvttps_epi32(ewrt);
 833             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 834             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 835                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 836                                          &ewtabF,&ewtabFn);
 837             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 838             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 839
 840             fscal            = felec;
 841
 842             /* Calculate temporary vectorial force */
 843             tx               = _mm_mul_ps(fscal,dx20);
 844             ty               = _mm_mul_ps(fscal,dy20);
 845             tz               = _mm_mul_ps(fscal,dz20);
 846
 847             /* Update vectorial force */
 848             fix2             = _mm_add_ps(fix2,tx);
 849             fiy2             = _mm_add_ps(fiy2,ty);
 850             fiz2             = _mm_add_ps(fiz2,tz);
 851
 852             fjx0             = _mm_add_ps(fjx0,tx);
 853             fjy0             = _mm_add_ps(fjy0,ty);
 854             fjz0             = _mm_add_ps(fjz0,tz);
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             r30              = _mm_mul_ps(rsq30,rinv30);
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             qq30             = _mm_mul_ps(iq3,jq0);
 864
 865             /* EWALD ELECTROSTATICS */
 866
 867             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 868             ewrt             = _mm_mul_ps(r30,ewtabscale);
 869             ewitab           = _mm_cvttps_epi32(ewrt);
 870             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 871             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 872                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 873                                          &ewtabF,&ewtabFn);
 874             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 875             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 876
 877             fscal            = felec;
 878
 879             /* Calculate temporary vectorial force */
 880             tx               = _mm_mul_ps(fscal,dx30);
 881             ty               = _mm_mul_ps(fscal,dy30);
 882             tz               = _mm_mul_ps(fscal,dz30);
 883
 884             /* Update vectorial force */
 885             fix3             = _mm_add_ps(fix3,tx);
 886             fiy3             = _mm_add_ps(fiy3,ty);
 887             fiz3             = _mm_add_ps(fiz3,tz);
 888
 889             fjx0             = _mm_add_ps(fjx0,tx);
 890             fjy0             = _mm_add_ps(fjy0,ty);
 891             fjz0             = _mm_add_ps(fjz0,tz);
 892
 893             fjptrA             = f+j_coord_offsetA;
 894             fjptrB             = f+j_coord_offsetB;
 895             fjptrC             = f+j_coord_offsetC;
 896             fjptrD             = f+j_coord_offsetD;
 897
 898             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 899
 900             /* Inner loop uses 108 flops */
 901         }
 902
 903         if(jidx<j_index_end)
 904         {
 905
 906             /* Get j neighbor index, and coordinate index */
 907             jnrlistA         = jjnr[jidx];
 908             jnrlistB         = jjnr[jidx+1];
 909             jnrlistC         = jjnr[jidx+2];
 910             jnrlistD         = jjnr[jidx+3];
 911             /* Sign of each element will be negative for non-real atoms.
 912              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 913              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 914              */
 915             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 916             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 917             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 918             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 919             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 920             j_coord_offsetA  = DIM*jnrA;
 921             j_coord_offsetB  = DIM*jnrB;
 922             j_coord_offsetC  = DIM*jnrC;
 923             j_coord_offsetD  = DIM*jnrD;
 924
 925             /* load j atom coordinates */
 926             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 927                                               x+j_coord_offsetC,x+j_coord_offsetD,
 928                                               &jx0,&jy0,&jz0);
 929
 930             /* Calculate displacement vector */
 931             dx10             = _mm_sub_ps(ix1,jx0);
 932             dy10             = _mm_sub_ps(iy1,jy0);
 933             dz10             = _mm_sub_ps(iz1,jz0);
 934             dx20             = _mm_sub_ps(ix2,jx0);
 935             dy20             = _mm_sub_ps(iy2,jy0);
 936             dz20             = _mm_sub_ps(iz2,jz0);
 937             dx30             = _mm_sub_ps(ix3,jx0);
 938             dy30             = _mm_sub_ps(iy3,jy0);
 939             dz30             = _mm_sub_ps(iz3,jz0);
 940
 941             /* Calculate squared distance and things based on it */
 942             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 943             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 944             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 945
 946             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 947             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 948             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 949
 950             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 951             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 952             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 953
 954             /* Load parameters for j particles */
 955             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 956                                                               charge+jnrC+0,charge+jnrD+0);
 957
 958             fjx0             = _mm_setzero_ps();
 959             fjy0             = _mm_setzero_ps();
 960             fjz0             = _mm_setzero_ps();
 961
 962             /**************************
 963              * CALCULATE INTERACTIONS *
 964              **************************/
 965
 966             r10              = _mm_mul_ps(rsq10,rinv10);
 967             r10              = _mm_andnot_ps(dummy_mask,r10);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq10             = _mm_mul_ps(iq1,jq0);
 971
 972             /* EWALD ELECTROSTATICS */
 973
 974             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 975             ewrt             = _mm_mul_ps(r10,ewtabscale);
 976             ewitab           = _mm_cvttps_epi32(ewrt);
 977             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 978             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 979                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 980                                          &ewtabF,&ewtabFn);
 981             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 982             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 983
 984             fscal            = felec;
 985
 986             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 987
 988             /* Calculate temporary vectorial force */
 989             tx               = _mm_mul_ps(fscal,dx10);
 990             ty               = _mm_mul_ps(fscal,dy10);
 991             tz               = _mm_mul_ps(fscal,dz10);
 992
 993             /* Update vectorial force */
 994             fix1             = _mm_add_ps(fix1,tx);
 995             fiy1             = _mm_add_ps(fiy1,ty);
 996             fiz1             = _mm_add_ps(fiz1,tz);
 997
 998             fjx0             = _mm_add_ps(fjx0,tx);
 999             fjy0             = _mm_add_ps(fjy0,ty);
1000             fjz0             = _mm_add_ps(fjz0,tz);
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             r20              = _mm_mul_ps(rsq20,rinv20);
1007             r20              = _mm_andnot_ps(dummy_mask,r20);
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             qq20             = _mm_mul_ps(iq2,jq0);
1011
1012             /* EWALD ELECTROSTATICS */
1013
1014             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1015             ewrt             = _mm_mul_ps(r20,ewtabscale);
1016             ewitab           = _mm_cvttps_epi32(ewrt);
1017             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1018             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1019                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1020                                          &ewtabF,&ewtabFn);
1021             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1022             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1027
1028             /* Calculate temporary vectorial force */
1029             tx               = _mm_mul_ps(fscal,dx20);
1030             ty               = _mm_mul_ps(fscal,dy20);
1031             tz               = _mm_mul_ps(fscal,dz20);
1032
1033             /* Update vectorial force */
1034             fix2             = _mm_add_ps(fix2,tx);
1035             fiy2             = _mm_add_ps(fiy2,ty);
1036             fiz2             = _mm_add_ps(fiz2,tz);
1037
1038             fjx0             = _mm_add_ps(fjx0,tx);
1039             fjy0             = _mm_add_ps(fjy0,ty);
1040             fjz0             = _mm_add_ps(fjz0,tz);
1041
1042             /**************************
1043              * CALCULATE INTERACTIONS *
1044              **************************/
1045
1046             r30              = _mm_mul_ps(rsq30,rinv30);
1047             r30              = _mm_andnot_ps(dummy_mask,r30);
1048
1049             /* Compute parameters for interactions between i and j atoms */
1050             qq30             = _mm_mul_ps(iq3,jq0);
1051
1052             /* EWALD ELECTROSTATICS */
1053
1054             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1055             ewrt             = _mm_mul_ps(r30,ewtabscale);
1056             ewitab           = _mm_cvttps_epi32(ewrt);
1057             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
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(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1067
1068             /* Calculate temporary vectorial force */
1069             tx               = _mm_mul_ps(fscal,dx30);
1070             ty               = _mm_mul_ps(fscal,dy30);
1071             tz               = _mm_mul_ps(fscal,dz30);
1072
1073             /* Update vectorial force */
1074             fix3             = _mm_add_ps(fix3,tx);
1075             fiy3             = _mm_add_ps(fiy3,ty);
1076             fiz3             = _mm_add_ps(fiz3,tz);
1077
1078             fjx0             = _mm_add_ps(fjx0,tx);
1079             fjy0             = _mm_add_ps(fjy0,ty);
1080             fjz0             = _mm_add_ps(fjz0,tz);
1081
1082             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1083             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1084             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1085             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1086
1087             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1088
1089             /* Inner loop uses 111 flops */
1090         }
1091
1092         /* End of innermost loop */
1093
1094         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1095                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1096
1097         /* Increment number of inner iterations */
1098         inneriter                  += j_index_end - j_index_start;
1099
1100         /* Outer loop uses 18 flops */
1101     }
1102
1103     /* Increment number of outer iterations */
1104     outeriter        += nri;
1105
1106     /* Update outer/inner flops */
1107
1108     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*111);
1109 }