src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     __m128i          ewitab;
  91     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  92     real             *ewtab;
  93     __m128           dummy_mask,cutoff_mask;
  94     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  95     __m128           one     = _mm_set1_ps(1.0);
  96     __m128           two     = _mm_set1_ps(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm_set1_ps(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110
 111     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 112     ewtab            = fr->ic->tabq_coul_FDV0;
 113     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 114     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 115
 116     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 117     rcutoff_scalar   = fr->rcoulomb;
 118     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 119     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = jnrC = jnrD = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125     j_coord_offsetC = 0;
 126     j_coord_offsetD = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm_setzero_ps();
 154         fiy0             = _mm_setzero_ps();
 155         fiz0             = _mm_setzero_ps();
 156
 157         /* Load parameters for i particles */
 158         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 159
 160         /* Reset potential sums */
 161         velecsum         = _mm_setzero_ps();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             jnrC             = jjnr[jidx+2];
 171             jnrD             = jjnr[jidx+3];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               x+j_coord_offsetC,x+j_coord_offsetD,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_ps(ix0,jx0);
 184             dy00             = _mm_sub_ps(iy0,jy0);
 185             dz00             = _mm_sub_ps(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 189
 190             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 191
 192             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 196                                                               charge+jnrC+0,charge+jnrD+0);
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_mm_any_lt(rsq00,rcutoff2))
 203             {
 204
 205             r00              = _mm_mul_ps(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _mm_mul_ps(iq0,jq0);
 209
 210             /* EWALD ELECTROSTATICS */
 211
 212             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 213             ewrt             = _mm_mul_ps(r00,ewtabscale);
 214             ewitab           = _mm_cvttps_epi32(ewrt);
 215             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 216             ewitab           = _mm_slli_epi32(ewitab,2);
 217             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 218             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 219             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 220             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 221             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 222             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 223             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 224             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 225             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 226
 227             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             velec            = _mm_and_ps(velec,cutoff_mask);
 231             velecsum         = _mm_add_ps(velecsum,velec);
 232
 233             fscal            = felec;
 234
 235             fscal            = _mm_and_ps(fscal,cutoff_mask);
 236
 237             /* Calculate temporary vectorial force */
 238             tx               = _mm_mul_ps(fscal,dx00);
 239             ty               = _mm_mul_ps(fscal,dy00);
 240             tz               = _mm_mul_ps(fscal,dz00);
 241
 242             /* Update vectorial force */
 243             fix0             = _mm_add_ps(fix0,tx);
 244             fiy0             = _mm_add_ps(fiy0,ty);
 245             fiz0             = _mm_add_ps(fiz0,tz);
 246
 247             fjptrA             = f+j_coord_offsetA;
 248             fjptrB             = f+j_coord_offsetB;
 249             fjptrC             = f+j_coord_offsetC;
 250             fjptrD             = f+j_coord_offsetD;
 251             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 252
 253             }
 254
 255             /* Inner loop uses 46 flops */
 256         }
 257
 258         if(jidx<j_index_end)
 259         {
 260
 261             /* Get j neighbor index, and coordinate index */
 262             jnrlistA         = jjnr[jidx];
 263             jnrlistB         = jjnr[jidx+1];
 264             jnrlistC         = jjnr[jidx+2];
 265             jnrlistD         = jjnr[jidx+3];
 266             /* Sign of each element will be negative for non-real atoms.
 267              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 268              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 269              */
 270             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 271             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 272             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 273             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 274             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 275             j_coord_offsetA  = DIM*jnrA;
 276             j_coord_offsetB  = DIM*jnrB;
 277             j_coord_offsetC  = DIM*jnrC;
 278             j_coord_offsetD  = DIM*jnrD;
 279
 280             /* load j atom coordinates */
 281             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 282                                               x+j_coord_offsetC,x+j_coord_offsetD,
 283                                               &jx0,&jy0,&jz0);
 284
 285             /* Calculate displacement vector */
 286             dx00             = _mm_sub_ps(ix0,jx0);
 287             dy00             = _mm_sub_ps(iy0,jy0);
 288             dz00             = _mm_sub_ps(iz0,jz0);
 289
 290             /* Calculate squared distance and things based on it */
 291             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 292
 293             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 294
 295             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 296
 297             /* Load parameters for j particles */
 298             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 299                                                               charge+jnrC+0,charge+jnrD+0);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             if (gmx_mm_any_lt(rsq00,rcutoff2))
 306             {
 307
 308             r00              = _mm_mul_ps(rsq00,rinv00);
 309             r00              = _mm_andnot_ps(dummy_mask,r00);
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq00             = _mm_mul_ps(iq0,jq0);
 313
 314             /* EWALD ELECTROSTATICS */
 315
 316             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 317             ewrt             = _mm_mul_ps(r00,ewtabscale);
 318             ewitab           = _mm_cvttps_epi32(ewrt);
 319             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 320             ewitab           = _mm_slli_epi32(ewitab,2);
 321             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 322             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 323             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 324             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 325             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 326             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 327             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 328             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 329             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 330
 331             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velec            = _mm_and_ps(velec,cutoff_mask);
 335             velec            = _mm_andnot_ps(dummy_mask,velec);
 336             velecsum         = _mm_add_ps(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340             fscal            = _mm_and_ps(fscal,cutoff_mask);
 341
 342             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm_mul_ps(fscal,dx00);
 346             ty               = _mm_mul_ps(fscal,dy00);
 347             tz               = _mm_mul_ps(fscal,dz00);
 348
 349             /* Update vectorial force */
 350             fix0             = _mm_add_ps(fix0,tx);
 351             fiy0             = _mm_add_ps(fiy0,ty);
 352             fiz0             = _mm_add_ps(fiz0,tz);
 353
 354             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 355             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 356             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 357             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 358             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 359
 360             }
 361
 362             /* Inner loop uses 47 flops */
 363         }
 364
 365         /* End of innermost loop */
 366
 367         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 368                                               f+i_coord_offset,fshift+i_shift_offset);
 369
 370         ggid                        = gid[iidx];
 371         /* Update potential energies */
 372         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 373
 374         /* Increment number of inner iterations */
 375         inneriter                  += j_index_end - j_index_start;
 376
 377         /* Outer loop uses 8 flops */
 378     }
 379
 380     /* Increment number of outer iterations */
 381     outeriter        += nri;
 382
 383     /* Update outer/inner flops */
 384
 385     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*47);
 386 }
 387 /*
 388  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_single
 389  * Electrostatics interaction: Ewald
 390  * VdW interaction:            None
 391  * Geometry:                   Particle-Particle
 392  * Calculate force/pot:        Force
 393  */
 394 void
 395 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_single
 396                     (t_nblist                    * gmx_restrict       nlist,
 397                      rvec                        * gmx_restrict          xx,
 398                      rvec                        * gmx_restrict          ff,
 399                      t_forcerec                  * gmx_restrict          fr,
 400                      t_mdatoms                   * gmx_restrict     mdatoms,
 401                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 402                      t_nrnb                      * gmx_restrict        nrnb)
 403 {
 404     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 405      * just 0 for non-waters.
 406      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 407      * jnr indices corresponding to data put in the four positions in the SIMD register.
 408      */
 409     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 410     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 411     int              jnrA,jnrB,jnrC,jnrD;
 412     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 413     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 414     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 415     real             rcutoff_scalar;
 416     real             *shiftvec,*fshift,*x,*f;
 417     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 418     real             scratch[4*DIM];
 419     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 420     int              vdwioffset0;
 421     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 422     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 423     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 424     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 425     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 426     real             *charge;
 427     __m128i          ewitab;
 428     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 429     real             *ewtab;
 430     __m128           dummy_mask,cutoff_mask;
 431     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 432     __m128           one     = _mm_set1_ps(1.0);
 433     __m128           two     = _mm_set1_ps(2.0);
 434     x                = xx[0];
 435     f                = ff[0];
 436
 437     nri              = nlist->nri;
 438     iinr             = nlist->iinr;
 439     jindex           = nlist->jindex;
 440     jjnr             = nlist->jjnr;
 441     shiftidx         = nlist->shift;
 442     gid              = nlist->gid;
 443     shiftvec         = fr->shift_vec[0];
 444     fshift           = fr->fshift[0];
 445     facel            = _mm_set1_ps(fr->epsfac);
 446     charge           = mdatoms->chargeA;
 447
 448     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 449     ewtab            = fr->ic->tabq_coul_F;
 450     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 451     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 452
 453     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 454     rcutoff_scalar   = fr->rcoulomb;
 455     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 456     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 457
 458     /* Avoid stupid compiler warnings */
 459     jnrA = jnrB = jnrC = jnrD = 0;
 460     j_coord_offsetA = 0;
 461     j_coord_offsetB = 0;
 462     j_coord_offsetC = 0;
 463     j_coord_offsetD = 0;
 464
 465     outeriter        = 0;
 466     inneriter        = 0;
 467
 468     for(iidx=0;iidx<4*DIM;iidx++)
 469     {
 470         scratch[iidx] = 0.0;
 471     }
 472
 473     /* Start outer loop over neighborlists */
 474     for(iidx=0; iidx<nri; iidx++)
 475     {
 476         /* Load shift vector for this list */
 477         i_shift_offset   = DIM*shiftidx[iidx];
 478
 479         /* Load limits for loop over neighbors */
 480         j_index_start    = jindex[iidx];
 481         j_index_end      = jindex[iidx+1];
 482
 483         /* Get outer coordinate index */
 484         inr              = iinr[iidx];
 485         i_coord_offset   = DIM*inr;
 486
 487         /* Load i particle coords and add shift vector */
 488         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 489
 490         fix0             = _mm_setzero_ps();
 491         fiy0             = _mm_setzero_ps();
 492         fiz0             = _mm_setzero_ps();
 493
 494         /* Load parameters for i particles */
 495         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 496
 497         /* Start inner kernel loop */
 498         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 499         {
 500
 501             /* Get j neighbor index, and coordinate index */
 502             jnrA             = jjnr[jidx];
 503             jnrB             = jjnr[jidx+1];
 504             jnrC             = jjnr[jidx+2];
 505             jnrD             = jjnr[jidx+3];
 506             j_coord_offsetA  = DIM*jnrA;
 507             j_coord_offsetB  = DIM*jnrB;
 508             j_coord_offsetC  = DIM*jnrC;
 509             j_coord_offsetD  = DIM*jnrD;
 510
 511             /* load j atom coordinates */
 512             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 513                                               x+j_coord_offsetC,x+j_coord_offsetD,
 514                                               &jx0,&jy0,&jz0);
 515
 516             /* Calculate displacement vector */
 517             dx00             = _mm_sub_ps(ix0,jx0);
 518             dy00             = _mm_sub_ps(iy0,jy0);
 519             dz00             = _mm_sub_ps(iz0,jz0);
 520
 521             /* Calculate squared distance and things based on it */
 522             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 523
 524             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 525
 526             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 527
 528             /* Load parameters for j particles */
 529             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 530                                                               charge+jnrC+0,charge+jnrD+0);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm_any_lt(rsq00,rcutoff2))
 537             {
 538
 539             r00              = _mm_mul_ps(rsq00,rinv00);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq00             = _mm_mul_ps(iq0,jq0);
 543
 544             /* EWALD ELECTROSTATICS */
 545
 546             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 547             ewrt             = _mm_mul_ps(r00,ewtabscale);
 548             ewitab           = _mm_cvttps_epi32(ewrt);
 549             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 550             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 551                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 552                                          &ewtabF,&ewtabFn);
 553             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 554             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 555
 556             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 557
 558             fscal            = felec;
 559
 560             fscal            = _mm_and_ps(fscal,cutoff_mask);
 561
 562             /* Calculate temporary vectorial force */
 563             tx               = _mm_mul_ps(fscal,dx00);
 564             ty               = _mm_mul_ps(fscal,dy00);
 565             tz               = _mm_mul_ps(fscal,dz00);
 566
 567             /* Update vectorial force */
 568             fix0             = _mm_add_ps(fix0,tx);
 569             fiy0             = _mm_add_ps(fiy0,ty);
 570             fiz0             = _mm_add_ps(fiz0,tz);
 571
 572             fjptrA             = f+j_coord_offsetA;
 573             fjptrB             = f+j_coord_offsetB;
 574             fjptrC             = f+j_coord_offsetC;
 575             fjptrD             = f+j_coord_offsetD;
 576             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 577
 578             }
 579
 580             /* Inner loop uses 39 flops */
 581         }
 582
 583         if(jidx<j_index_end)
 584         {
 585
 586             /* Get j neighbor index, and coordinate index */
 587             jnrlistA         = jjnr[jidx];
 588             jnrlistB         = jjnr[jidx+1];
 589             jnrlistC         = jjnr[jidx+2];
 590             jnrlistD         = jjnr[jidx+3];
 591             /* Sign of each element will be negative for non-real atoms.
 592              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 593              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 594              */
 595             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 596             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 597             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 598             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 599             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 600             j_coord_offsetA  = DIM*jnrA;
 601             j_coord_offsetB  = DIM*jnrB;
 602             j_coord_offsetC  = DIM*jnrC;
 603             j_coord_offsetD  = DIM*jnrD;
 604
 605             /* load j atom coordinates */
 606             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 607                                               x+j_coord_offsetC,x+j_coord_offsetD,
 608                                               &jx0,&jy0,&jz0);
 609
 610             /* Calculate displacement vector */
 611             dx00             = _mm_sub_ps(ix0,jx0);
 612             dy00             = _mm_sub_ps(iy0,jy0);
 613             dz00             = _mm_sub_ps(iz0,jz0);
 614
 615             /* Calculate squared distance and things based on it */
 616             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 617
 618             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 619
 620             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 621
 622             /* Load parameters for j particles */
 623             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 624                                                               charge+jnrC+0,charge+jnrD+0);
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             if (gmx_mm_any_lt(rsq00,rcutoff2))
 631             {
 632
 633             r00              = _mm_mul_ps(rsq00,rinv00);
 634             r00              = _mm_andnot_ps(dummy_mask,r00);
 635
 636             /* Compute parameters for interactions between i and j atoms */
 637             qq00             = _mm_mul_ps(iq0,jq0);
 638
 639             /* EWALD ELECTROSTATICS */
 640
 641             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 642             ewrt             = _mm_mul_ps(r00,ewtabscale);
 643             ewitab           = _mm_cvttps_epi32(ewrt);
 644             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 645             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 646                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 647                                          &ewtabF,&ewtabFn);
 648             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 649             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 650
 651             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 652
 653             fscal            = felec;
 654
 655             fscal            = _mm_and_ps(fscal,cutoff_mask);
 656
 657             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm_mul_ps(fscal,dx00);
 661             ty               = _mm_mul_ps(fscal,dy00);
 662             tz               = _mm_mul_ps(fscal,dz00);
 663
 664             /* Update vectorial force */
 665             fix0             = _mm_add_ps(fix0,tx);
 666             fiy0             = _mm_add_ps(fiy0,ty);
 667             fiz0             = _mm_add_ps(fiz0,tz);
 668
 669             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 670             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 671             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 672             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 673             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 674
 675             }
 676
 677             /* Inner loop uses 40 flops */
 678         }
 679
 680         /* End of innermost loop */
 681
 682         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 683                                               f+i_coord_offset,fshift+i_shift_offset);
 684
 685         /* Increment number of inner iterations */
 686         inneriter                  += j_index_end - j_index_start;
 687
 688         /* Outer loop uses 7 flops */
 689     }
 690
 691     /* Increment number of outer iterations */
 692     outeriter        += nri;
 693
 694     /* Update outer/inner flops */
 695
 696     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*40);
 697 }