src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwNone_GeomP1P1_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_VdwNone_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomP1P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128i          ewitab;
  93     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 114     ewtab            = fr->ic->tabq_coul_FDV0;
 115     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 116     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = jnrC = jnrD = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122     j_coord_offsetC = 0;
 123     j_coord_offsetD = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     for(iidx=0;iidx<4*DIM;iidx++)
 129     {
 130         scratch[iidx] = 0.0;
 131     }
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 149
 150         fix0             = _mm_setzero_ps();
 151         fiy0             = _mm_setzero_ps();
 152         fiz0             = _mm_setzero_ps();
 153
 154         /* Load parameters for i particles */
 155         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_ps();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             jnrC             = jjnr[jidx+2];
 168             jnrD             = jjnr[jidx+3];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171             j_coord_offsetC  = DIM*jnrC;
 172             j_coord_offsetD  = DIM*jnrD;
 173
 174             /* load j atom coordinates */
 175             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               x+j_coord_offsetC,x+j_coord_offsetD,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx00             = _mm_sub_ps(ix0,jx0);
 181             dy00             = _mm_sub_ps(iy0,jy0);
 182             dz00             = _mm_sub_ps(iz0,jz0);
 183
 184             /* Calculate squared distance and things based on it */
 185             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 186
 187             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 188
 189             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 190
 191             /* Load parameters for j particles */
 192             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 193                                                               charge+jnrC+0,charge+jnrD+0);
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             r00              = _mm_mul_ps(rsq00,rinv00);
 200
 201             /* Compute parameters for interactions between i and j atoms */
 202             qq00             = _mm_mul_ps(iq0,jq0);
 203
 204             /* EWALD ELECTROSTATICS */
 205
 206             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 207             ewrt             = _mm_mul_ps(r00,ewtabscale);
 208             ewitab           = _mm_cvttps_epi32(ewrt);
 209             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 210             ewitab           = _mm_slli_epi32(ewitab,2);
 211             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 212             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 213             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 214             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 215             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 216             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 217             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 218             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 219             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 220
 221             /* Update potential sum for this i atom from the interaction with this j atom. */
 222             velecsum         = _mm_add_ps(velecsum,velec);
 223
 224             fscal            = felec;
 225
 226             /* Calculate temporary vectorial force */
 227             tx               = _mm_mul_ps(fscal,dx00);
 228             ty               = _mm_mul_ps(fscal,dy00);
 229             tz               = _mm_mul_ps(fscal,dz00);
 230
 231             /* Update vectorial force */
 232             fix0             = _mm_add_ps(fix0,tx);
 233             fiy0             = _mm_add_ps(fiy0,ty);
 234             fiz0             = _mm_add_ps(fiz0,tz);
 235
 236             fjptrA             = f+j_coord_offsetA;
 237             fjptrB             = f+j_coord_offsetB;
 238             fjptrC             = f+j_coord_offsetC;
 239             fjptrD             = f+j_coord_offsetD;
 240             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 241
 242             /* Inner loop uses 41 flops */
 243         }
 244
 245         if(jidx<j_index_end)
 246         {
 247
 248             /* Get j neighbor index, and coordinate index */
 249             jnrlistA         = jjnr[jidx];
 250             jnrlistB         = jjnr[jidx+1];
 251             jnrlistC         = jjnr[jidx+2];
 252             jnrlistD         = jjnr[jidx+3];
 253             /* Sign of each element will be negative for non-real atoms.
 254              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 255              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 256              */
 257             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 258             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 259             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 260             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 261             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 262             j_coord_offsetA  = DIM*jnrA;
 263             j_coord_offsetB  = DIM*jnrB;
 264             j_coord_offsetC  = DIM*jnrC;
 265             j_coord_offsetD  = DIM*jnrD;
 266
 267             /* load j atom coordinates */
 268             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 269                                               x+j_coord_offsetC,x+j_coord_offsetD,
 270                                               &jx0,&jy0,&jz0);
 271
 272             /* Calculate displacement vector */
 273             dx00             = _mm_sub_ps(ix0,jx0);
 274             dy00             = _mm_sub_ps(iy0,jy0);
 275             dz00             = _mm_sub_ps(iz0,jz0);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 279
 280             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 281
 282             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 283
 284             /* Load parameters for j particles */
 285             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 286                                                               charge+jnrC+0,charge+jnrD+0);
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             r00              = _mm_mul_ps(rsq00,rinv00);
 293             r00              = _mm_andnot_ps(dummy_mask,r00);
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             qq00             = _mm_mul_ps(iq0,jq0);
 297
 298             /* EWALD ELECTROSTATICS */
 299
 300             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 301             ewrt             = _mm_mul_ps(r00,ewtabscale);
 302             ewitab           = _mm_cvttps_epi32(ewrt);
 303             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 304             ewitab           = _mm_slli_epi32(ewitab,2);
 305             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 306             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 307             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 308             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 309             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 310             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 311             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 312             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 313             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _mm_andnot_ps(dummy_mask,velec);
 317             velecsum         = _mm_add_ps(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm_mul_ps(fscal,dx00);
 325             ty               = _mm_mul_ps(fscal,dy00);
 326             tz               = _mm_mul_ps(fscal,dz00);
 327
 328             /* Update vectorial force */
 329             fix0             = _mm_add_ps(fix0,tx);
 330             fiy0             = _mm_add_ps(fiy0,ty);
 331             fiz0             = _mm_add_ps(fiz0,tz);
 332
 333             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 334             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 335             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 336             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 337             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 338
 339             /* Inner loop uses 42 flops */
 340         }
 341
 342         /* End of innermost loop */
 343
 344         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 345                                               f+i_coord_offset,fshift+i_shift_offset);
 346
 347         ggid                        = gid[iidx];
 348         /* Update potential energies */
 349         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 350
 351         /* Increment number of inner iterations */
 352         inneriter                  += j_index_end - j_index_start;
 353
 354         /* Outer loop uses 8 flops */
 355     }
 356
 357     /* Increment number of outer iterations */
 358     outeriter        += nri;
 359
 360     /* Update outer/inner flops */
 361
 362     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*42);
 363 }
 364 /*
 365  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_single
 366  * Electrostatics interaction: Ewald
 367  * VdW interaction:            None
 368  * Geometry:                   Particle-Particle
 369  * Calculate force/pot:        Force
 370  */
 371 void
 372 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_single
 373                     (t_nblist                    * gmx_restrict       nlist,
 374                      rvec                        * gmx_restrict          xx,
 375                      rvec                        * gmx_restrict          ff,
 376                      t_forcerec                  * gmx_restrict          fr,
 377                      t_mdatoms                   * gmx_restrict     mdatoms,
 378                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 379                      t_nrnb                      * gmx_restrict        nrnb)
 380 {
 381     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 382      * just 0 for non-waters.
 383      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 384      * jnr indices corresponding to data put in the four positions in the SIMD register.
 385      */
 386     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 387     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 388     int              jnrA,jnrB,jnrC,jnrD;
 389     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 390     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 391     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 392     real             rcutoff_scalar;
 393     real             *shiftvec,*fshift,*x,*f;
 394     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 395     real             scratch[4*DIM];
 396     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 397     int              vdwioffset0;
 398     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 399     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 400     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 401     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 402     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 403     real             *charge;
 404     __m128i          ewitab;
 405     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 406     real             *ewtab;
 407     __m128           dummy_mask,cutoff_mask;
 408     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 409     __m128           one     = _mm_set1_ps(1.0);
 410     __m128           two     = _mm_set1_ps(2.0);
 411     x                = xx[0];
 412     f                = ff[0];
 413
 414     nri              = nlist->nri;
 415     iinr             = nlist->iinr;
 416     jindex           = nlist->jindex;
 417     jjnr             = nlist->jjnr;
 418     shiftidx         = nlist->shift;
 419     gid              = nlist->gid;
 420     shiftvec         = fr->shift_vec[0];
 421     fshift           = fr->fshift[0];
 422     facel            = _mm_set1_ps(fr->epsfac);
 423     charge           = mdatoms->chargeA;
 424
 425     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 426     ewtab            = fr->ic->tabq_coul_F;
 427     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 428     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 429
 430     /* Avoid stupid compiler warnings */
 431     jnrA = jnrB = jnrC = jnrD = 0;
 432     j_coord_offsetA = 0;
 433     j_coord_offsetB = 0;
 434     j_coord_offsetC = 0;
 435     j_coord_offsetD = 0;
 436
 437     outeriter        = 0;
 438     inneriter        = 0;
 439
 440     for(iidx=0;iidx<4*DIM;iidx++)
 441     {
 442         scratch[iidx] = 0.0;
 443     }
 444
 445     /* Start outer loop over neighborlists */
 446     for(iidx=0; iidx<nri; iidx++)
 447     {
 448         /* Load shift vector for this list */
 449         i_shift_offset   = DIM*shiftidx[iidx];
 450
 451         /* Load limits for loop over neighbors */
 452         j_index_start    = jindex[iidx];
 453         j_index_end      = jindex[iidx+1];
 454
 455         /* Get outer coordinate index */
 456         inr              = iinr[iidx];
 457         i_coord_offset   = DIM*inr;
 458
 459         /* Load i particle coords and add shift vector */
 460         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 461
 462         fix0             = _mm_setzero_ps();
 463         fiy0             = _mm_setzero_ps();
 464         fiz0             = _mm_setzero_ps();
 465
 466         /* Load parameters for i particles */
 467         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 468
 469         /* Start inner kernel loop */
 470         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 471         {
 472
 473             /* Get j neighbor index, and coordinate index */
 474             jnrA             = jjnr[jidx];
 475             jnrB             = jjnr[jidx+1];
 476             jnrC             = jjnr[jidx+2];
 477             jnrD             = jjnr[jidx+3];
 478             j_coord_offsetA  = DIM*jnrA;
 479             j_coord_offsetB  = DIM*jnrB;
 480             j_coord_offsetC  = DIM*jnrC;
 481             j_coord_offsetD  = DIM*jnrD;
 482
 483             /* load j atom coordinates */
 484             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 485                                               x+j_coord_offsetC,x+j_coord_offsetD,
 486                                               &jx0,&jy0,&jz0);
 487
 488             /* Calculate displacement vector */
 489             dx00             = _mm_sub_ps(ix0,jx0);
 490             dy00             = _mm_sub_ps(iy0,jy0);
 491             dz00             = _mm_sub_ps(iz0,jz0);
 492
 493             /* Calculate squared distance and things based on it */
 494             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 495
 496             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 497
 498             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 499
 500             /* Load parameters for j particles */
 501             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 502                                                               charge+jnrC+0,charge+jnrD+0);
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             r00              = _mm_mul_ps(rsq00,rinv00);
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             qq00             = _mm_mul_ps(iq0,jq0);
 512
 513             /* EWALD ELECTROSTATICS */
 514
 515             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 516             ewrt             = _mm_mul_ps(r00,ewtabscale);
 517             ewitab           = _mm_cvttps_epi32(ewrt);
 518             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 519             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 520                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 521                                          &ewtabF,&ewtabFn);
 522             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 523             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 524
 525             fscal            = felec;
 526
 527             /* Calculate temporary vectorial force */
 528             tx               = _mm_mul_ps(fscal,dx00);
 529             ty               = _mm_mul_ps(fscal,dy00);
 530             tz               = _mm_mul_ps(fscal,dz00);
 531
 532             /* Update vectorial force */
 533             fix0             = _mm_add_ps(fix0,tx);
 534             fiy0             = _mm_add_ps(fiy0,ty);
 535             fiz0             = _mm_add_ps(fiz0,tz);
 536
 537             fjptrA             = f+j_coord_offsetA;
 538             fjptrB             = f+j_coord_offsetB;
 539             fjptrC             = f+j_coord_offsetC;
 540             fjptrD             = f+j_coord_offsetD;
 541             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 542
 543             /* Inner loop uses 36 flops */
 544         }
 545
 546         if(jidx<j_index_end)
 547         {
 548
 549             /* Get j neighbor index, and coordinate index */
 550             jnrlistA         = jjnr[jidx];
 551             jnrlistB         = jjnr[jidx+1];
 552             jnrlistC         = jjnr[jidx+2];
 553             jnrlistD         = jjnr[jidx+3];
 554             /* Sign of each element will be negative for non-real atoms.
 555              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 556              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 557              */
 558             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 559             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 560             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 561             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 562             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 563             j_coord_offsetA  = DIM*jnrA;
 564             j_coord_offsetB  = DIM*jnrB;
 565             j_coord_offsetC  = DIM*jnrC;
 566             j_coord_offsetD  = DIM*jnrD;
 567
 568             /* load j atom coordinates */
 569             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 570                                               x+j_coord_offsetC,x+j_coord_offsetD,
 571                                               &jx0,&jy0,&jz0);
 572
 573             /* Calculate displacement vector */
 574             dx00             = _mm_sub_ps(ix0,jx0);
 575             dy00             = _mm_sub_ps(iy0,jy0);
 576             dz00             = _mm_sub_ps(iz0,jz0);
 577
 578             /* Calculate squared distance and things based on it */
 579             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 580
 581             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 582
 583             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 584
 585             /* Load parameters for j particles */
 586             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 587                                                               charge+jnrC+0,charge+jnrD+0);
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             r00              = _mm_mul_ps(rsq00,rinv00);
 594             r00              = _mm_andnot_ps(dummy_mask,r00);
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq00             = _mm_mul_ps(iq0,jq0);
 598
 599             /* EWALD ELECTROSTATICS */
 600
 601             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 602             ewrt             = _mm_mul_ps(r00,ewtabscale);
 603             ewitab           = _mm_cvttps_epi32(ewrt);
 604             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 605             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 606                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 607                                          &ewtabF,&ewtabFn);
 608             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 609             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm_mul_ps(fscal,dx00);
 617             ty               = _mm_mul_ps(fscal,dy00);
 618             tz               = _mm_mul_ps(fscal,dz00);
 619
 620             /* Update vectorial force */
 621             fix0             = _mm_add_ps(fix0,tx);
 622             fiy0             = _mm_add_ps(fiy0,ty);
 623             fiz0             = _mm_add_ps(fiz0,tz);
 624
 625             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 626             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 627             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 628             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 629             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 630
 631             /* Inner loop uses 37 flops */
 632         }
 633
 634         /* End of innermost loop */
 635
 636         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 637                                               f+i_coord_offset,fshift+i_shift_offset);
 638
 639         /* Increment number of inner iterations */
 640         inneriter                  += j_index_end - j_index_start;
 641
 642         /* Outer loop uses 7 flops */
 643     }
 644
 645     /* Increment number of outer iterations */
 646     outeriter        += nri;
 647
 648     /* Update outer/inner flops */
 649
 650     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*37);
 651 }