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) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_sse4_1_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          ewitab;
  90     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     real             *ewtab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->ic->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = jnrC = jnrD = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119     j_coord_offsetC = 0;
 120     j_coord_offsetD = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     for(iidx=0;iidx<4*DIM;iidx++)
 126     {
 127         scratch[iidx] = 0.0;
 128     }
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _mm_setzero_ps();
 148         fiy0             = _mm_setzero_ps();
 149         fiz0             = _mm_setzero_ps();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_ps();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             jnrC             = jjnr[jidx+2];
 165             jnrD             = jjnr[jidx+3];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168             j_coord_offsetC  = DIM*jnrC;
 169             j_coord_offsetD  = DIM*jnrD;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               x+j_coord_offsetC,x+j_coord_offsetD,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _mm_sub_ps(ix0,jx0);
 178             dy00             = _mm_sub_ps(iy0,jy0);
 179             dz00             = _mm_sub_ps(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 183
 184             rinv00           = sse41_invsqrt_f(rsq00);
 185
 186             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 190                                                               charge+jnrC+0,charge+jnrD+0);
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             r00              = _mm_mul_ps(rsq00,rinv00);
 197
 198             /* Compute parameters for interactions between i and j atoms */
 199             qq00             = _mm_mul_ps(iq0,jq0);
 200
 201             /* EWALD ELECTROSTATICS */
 202
 203             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 204             ewrt             = _mm_mul_ps(r00,ewtabscale);
 205             ewitab           = _mm_cvttps_epi32(ewrt);
 206             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 207             ewitab           = _mm_slli_epi32(ewitab,2);
 208             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 209             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 210             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 211             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 212             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 213             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 214             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 215             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 216             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 217
 218             /* Update potential sum for this i atom from the interaction with this j atom. */
 219             velecsum         = _mm_add_ps(velecsum,velec);
 220
 221             fscal            = felec;
 222
 223             /* Calculate temporary vectorial force */
 224             tx               = _mm_mul_ps(fscal,dx00);
 225             ty               = _mm_mul_ps(fscal,dy00);
 226             tz               = _mm_mul_ps(fscal,dz00);
 227
 228             /* Update vectorial force */
 229             fix0             = _mm_add_ps(fix0,tx);
 230             fiy0             = _mm_add_ps(fiy0,ty);
 231             fiz0             = _mm_add_ps(fiz0,tz);
 232
 233             fjptrA             = f+j_coord_offsetA;
 234             fjptrB             = f+j_coord_offsetB;
 235             fjptrC             = f+j_coord_offsetC;
 236             fjptrD             = f+j_coord_offsetD;
 237             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 238
 239             /* Inner loop uses 41 flops */
 240         }
 241
 242         if(jidx<j_index_end)
 243         {
 244
 245             /* Get j neighbor index, and coordinate index */
 246             jnrlistA         = jjnr[jidx];
 247             jnrlistB         = jjnr[jidx+1];
 248             jnrlistC         = jjnr[jidx+2];
 249             jnrlistD         = jjnr[jidx+3];
 250             /* Sign of each element will be negative for non-real atoms.
 251              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 252              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 253              */
 254             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 255             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 256             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 257             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 258             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 259             j_coord_offsetA  = DIM*jnrA;
 260             j_coord_offsetB  = DIM*jnrB;
 261             j_coord_offsetC  = DIM*jnrC;
 262             j_coord_offsetD  = DIM*jnrD;
 263
 264             /* load j atom coordinates */
 265             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 266                                               x+j_coord_offsetC,x+j_coord_offsetD,
 267                                               &jx0,&jy0,&jz0);
 268
 269             /* Calculate displacement vector */
 270             dx00             = _mm_sub_ps(ix0,jx0);
 271             dy00             = _mm_sub_ps(iy0,jy0);
 272             dz00             = _mm_sub_ps(iz0,jz0);
 273
 274             /* Calculate squared distance and things based on it */
 275             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 276
 277             rinv00           = sse41_invsqrt_f(rsq00);
 278
 279             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 280
 281             /* Load parameters for j particles */
 282             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 283                                                               charge+jnrC+0,charge+jnrD+0);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             r00              = _mm_mul_ps(rsq00,rinv00);
 290             r00              = _mm_andnot_ps(dummy_mask,r00);
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq00             = _mm_mul_ps(iq0,jq0);
 294
 295             /* EWALD ELECTROSTATICS */
 296
 297             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 298             ewrt             = _mm_mul_ps(r00,ewtabscale);
 299             ewitab           = _mm_cvttps_epi32(ewrt);
 300             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 301             ewitab           = _mm_slli_epi32(ewitab,2);
 302             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 303             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 304             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 305             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 306             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 307             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 308             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 309             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 310             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velec            = _mm_andnot_ps(dummy_mask,velec);
 314             velecsum         = _mm_add_ps(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm_mul_ps(fscal,dx00);
 322             ty               = _mm_mul_ps(fscal,dy00);
 323             tz               = _mm_mul_ps(fscal,dz00);
 324
 325             /* Update vectorial force */
 326             fix0             = _mm_add_ps(fix0,tx);
 327             fiy0             = _mm_add_ps(fiy0,ty);
 328             fiz0             = _mm_add_ps(fiz0,tz);
 329
 330             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 331             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 332             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 333             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 334             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 335
 336             /* Inner loop uses 42 flops */
 337         }
 338
 339         /* End of innermost loop */
 340
 341         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 342                                               f+i_coord_offset,fshift+i_shift_offset);
 343
 344         ggid                        = gid[iidx];
 345         /* Update potential energies */
 346         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 347
 348         /* Increment number of inner iterations */
 349         inneriter                  += j_index_end - j_index_start;
 350
 351         /* Outer loop uses 8 flops */
 352     }
 353
 354     /* Increment number of outer iterations */
 355     outeriter        += nri;
 356
 357     /* Update outer/inner flops */
 358
 359     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*42);
 360 }
 361 /*
 362  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_single
 363  * Electrostatics interaction: Ewald
 364  * VdW interaction:            None
 365  * Geometry:                   Particle-Particle
 366  * Calculate force/pot:        Force
 367  */
 368 void
 369 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_single
 370                     (t_nblist                    * gmx_restrict       nlist,
 371                      rvec                        * gmx_restrict          xx,
 372                      rvec                        * gmx_restrict          ff,
 373                      struct t_forcerec           * gmx_restrict          fr,
 374                      t_mdatoms                   * gmx_restrict     mdatoms,
 375                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 376                      t_nrnb                      * gmx_restrict        nrnb)
 377 {
 378     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 379      * just 0 for non-waters.
 380      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 381      * jnr indices corresponding to data put in the four positions in the SIMD register.
 382      */
 383     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 384     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 385     int              jnrA,jnrB,jnrC,jnrD;
 386     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 387     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 388     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 389     real             rcutoff_scalar;
 390     real             *shiftvec,*fshift,*x,*f;
 391     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 392     real             scratch[4*DIM];
 393     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 394     int              vdwioffset0;
 395     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 396     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 397     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 398     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 399     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 400     real             *charge;
 401     __m128i          ewitab;
 402     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 403     real             *ewtab;
 404     __m128           dummy_mask,cutoff_mask;
 405     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 406     __m128           one     = _mm_set1_ps(1.0);
 407     __m128           two     = _mm_set1_ps(2.0);
 408     x                = xx[0];
 409     f                = ff[0];
 410
 411     nri              = nlist->nri;
 412     iinr             = nlist->iinr;
 413     jindex           = nlist->jindex;
 414     jjnr             = nlist->jjnr;
 415     shiftidx         = nlist->shift;
 416     gid              = nlist->gid;
 417     shiftvec         = fr->shift_vec[0];
 418     fshift           = fr->fshift[0];
 419     facel            = _mm_set1_ps(fr->ic->epsfac);
 420     charge           = mdatoms->chargeA;
 421
 422     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 423     ewtab            = fr->ic->tabq_coul_F;
 424     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 425     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 426
 427     /* Avoid stupid compiler warnings */
 428     jnrA = jnrB = jnrC = jnrD = 0;
 429     j_coord_offsetA = 0;
 430     j_coord_offsetB = 0;
 431     j_coord_offsetC = 0;
 432     j_coord_offsetD = 0;
 433
 434     outeriter        = 0;
 435     inneriter        = 0;
 436
 437     for(iidx=0;iidx<4*DIM;iidx++)
 438     {
 439         scratch[iidx] = 0.0;
 440     }
 441
 442     /* Start outer loop over neighborlists */
 443     for(iidx=0; iidx<nri; iidx++)
 444     {
 445         /* Load shift vector for this list */
 446         i_shift_offset   = DIM*shiftidx[iidx];
 447
 448         /* Load limits for loop over neighbors */
 449         j_index_start    = jindex[iidx];
 450         j_index_end      = jindex[iidx+1];
 451
 452         /* Get outer coordinate index */
 453         inr              = iinr[iidx];
 454         i_coord_offset   = DIM*inr;
 455
 456         /* Load i particle coords and add shift vector */
 457         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 458
 459         fix0             = _mm_setzero_ps();
 460         fiy0             = _mm_setzero_ps();
 461         fiz0             = _mm_setzero_ps();
 462
 463         /* Load parameters for i particles */
 464         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 465
 466         /* Start inner kernel loop */
 467         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 468         {
 469
 470             /* Get j neighbor index, and coordinate index */
 471             jnrA             = jjnr[jidx];
 472             jnrB             = jjnr[jidx+1];
 473             jnrC             = jjnr[jidx+2];
 474             jnrD             = jjnr[jidx+3];
 475             j_coord_offsetA  = DIM*jnrA;
 476             j_coord_offsetB  = DIM*jnrB;
 477             j_coord_offsetC  = DIM*jnrC;
 478             j_coord_offsetD  = DIM*jnrD;
 479
 480             /* load j atom coordinates */
 481             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 482                                               x+j_coord_offsetC,x+j_coord_offsetD,
 483                                               &jx0,&jy0,&jz0);
 484
 485             /* Calculate displacement vector */
 486             dx00             = _mm_sub_ps(ix0,jx0);
 487             dy00             = _mm_sub_ps(iy0,jy0);
 488             dz00             = _mm_sub_ps(iz0,jz0);
 489
 490             /* Calculate squared distance and things based on it */
 491             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 492
 493             rinv00           = sse41_invsqrt_f(rsq00);
 494
 495             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 496
 497             /* Load parameters for j particles */
 498             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 499                                                               charge+jnrC+0,charge+jnrD+0);
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             r00              = _mm_mul_ps(rsq00,rinv00);
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             qq00             = _mm_mul_ps(iq0,jq0);
 509
 510             /* EWALD ELECTROSTATICS */
 511
 512             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 513             ewrt             = _mm_mul_ps(r00,ewtabscale);
 514             ewitab           = _mm_cvttps_epi32(ewrt);
 515             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 516             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 517                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 518                                          &ewtabF,&ewtabFn);
 519             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 520             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 521
 522             fscal            = felec;
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm_mul_ps(fscal,dx00);
 526             ty               = _mm_mul_ps(fscal,dy00);
 527             tz               = _mm_mul_ps(fscal,dz00);
 528
 529             /* Update vectorial force */
 530             fix0             = _mm_add_ps(fix0,tx);
 531             fiy0             = _mm_add_ps(fiy0,ty);
 532             fiz0             = _mm_add_ps(fiz0,tz);
 533
 534             fjptrA             = f+j_coord_offsetA;
 535             fjptrB             = f+j_coord_offsetB;
 536             fjptrC             = f+j_coord_offsetC;
 537             fjptrD             = f+j_coord_offsetD;
 538             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 539
 540             /* Inner loop uses 36 flops */
 541         }
 542
 543         if(jidx<j_index_end)
 544         {
 545
 546             /* Get j neighbor index, and coordinate index */
 547             jnrlistA         = jjnr[jidx];
 548             jnrlistB         = jjnr[jidx+1];
 549             jnrlistC         = jjnr[jidx+2];
 550             jnrlistD         = jjnr[jidx+3];
 551             /* Sign of each element will be negative for non-real atoms.
 552              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 553              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 554              */
 555             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 556             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 557             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 558             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 559             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 560             j_coord_offsetA  = DIM*jnrA;
 561             j_coord_offsetB  = DIM*jnrB;
 562             j_coord_offsetC  = DIM*jnrC;
 563             j_coord_offsetD  = DIM*jnrD;
 564
 565             /* load j atom coordinates */
 566             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 567                                               x+j_coord_offsetC,x+j_coord_offsetD,
 568                                               &jx0,&jy0,&jz0);
 569
 570             /* Calculate displacement vector */
 571             dx00             = _mm_sub_ps(ix0,jx0);
 572             dy00             = _mm_sub_ps(iy0,jy0);
 573             dz00             = _mm_sub_ps(iz0,jz0);
 574
 575             /* Calculate squared distance and things based on it */
 576             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 577
 578             rinv00           = sse41_invsqrt_f(rsq00);
 579
 580             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 581
 582             /* Load parameters for j particles */
 583             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 584                                                               charge+jnrC+0,charge+jnrD+0);
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             r00              = _mm_mul_ps(rsq00,rinv00);
 591             r00              = _mm_andnot_ps(dummy_mask,r00);
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq00             = _mm_mul_ps(iq0,jq0);
 595
 596             /* EWALD ELECTROSTATICS */
 597
 598             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 599             ewrt             = _mm_mul_ps(r00,ewtabscale);
 600             ewitab           = _mm_cvttps_epi32(ewrt);
 601             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 602             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 603                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 604                                          &ewtabF,&ewtabFn);
 605             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 606             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 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,dx00);
 614             ty               = _mm_mul_ps(fscal,dy00);
 615             tz               = _mm_mul_ps(fscal,dz00);
 616
 617             /* Update vectorial force */
 618             fix0             = _mm_add_ps(fix0,tx);
 619             fiy0             = _mm_add_ps(fiy0,ty);
 620             fiz0             = _mm_add_ps(fiz0,tz);
 621
 622             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 623             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 624             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 625             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 626             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 627
 628             /* Inner loop uses 37 flops */
 629         }
 630
 631         /* End of innermost loop */
 632
 633         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 634                                               f+i_coord_offset,fshift+i_shift_offset);
 635
 636         /* Increment number of inner iterations */
 637         inneriter                  += j_index_end - j_index_start;
 638
 639         /* Outer loop uses 7 flops */
 640     }
 641
 642     /* Increment number of outer iterations */
 643     outeriter        += nri;
 644
 645     /* Update outer/inner flops */
 646
 647     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*37);
 648 }