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 "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_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_sse4_1_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwNone_GeomP1P1_VF_sse4_1_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     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = jnrC = jnrD = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120     j_coord_offsetC = 0;
 121     j_coord_offsetD = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     for(iidx=0;iidx<4*DIM;iidx++)
 127     {
 128         scratch[iidx] = 0.0;
 129     }
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 147
 148         fix0             = _mm_setzero_ps();
 149         fiy0             = _mm_setzero_ps();
 150         fiz0             = _mm_setzero_ps();
 151
 152         /* Load parameters for i particles */
 153         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 154
 155         /* Reset potential sums */
 156         velecsum         = _mm_setzero_ps();
 157
 158         /* Start inner kernel loop */
 159         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 160         {
 161
 162             /* Get j neighbor index, and coordinate index */
 163             jnrA             = jjnr[jidx];
 164             jnrB             = jjnr[jidx+1];
 165             jnrC             = jjnr[jidx+2];
 166             jnrD             = jjnr[jidx+3];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169             j_coord_offsetC  = DIM*jnrC;
 170             j_coord_offsetD  = DIM*jnrD;
 171
 172             /* load j atom coordinates */
 173             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               x+j_coord_offsetC,x+j_coord_offsetD,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_ps(ix0,jx0);
 179             dy00             = _mm_sub_ps(iy0,jy0);
 180             dz00             = _mm_sub_ps(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 184
 185             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 186
 187             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 188
 189             /* Load parameters for j particles */
 190             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 191                                                               charge+jnrC+0,charge+jnrD+0);
 192
 193             /**************************
 194              * CALCULATE INTERACTIONS *
 195              **************************/
 196
 197             r00              = _mm_mul_ps(rsq00,rinv00);
 198
 199             /* Compute parameters for interactions between i and j atoms */
 200             qq00             = _mm_mul_ps(iq0,jq0);
 201
 202             /* EWALD ELECTROSTATICS */
 203
 204             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 205             ewrt             = _mm_mul_ps(r00,ewtabscale);
 206             ewitab           = _mm_cvttps_epi32(ewrt);
 207             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 208             ewitab           = _mm_slli_epi32(ewitab,2);
 209             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 210             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 211             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 212             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 213             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 214             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 215             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 216             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 217             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 218
 219             /* Update potential sum for this i atom from the interaction with this j atom. */
 220             velecsum         = _mm_add_ps(velecsum,velec);
 221
 222             fscal            = felec;
 223
 224             /* Calculate temporary vectorial force */
 225             tx               = _mm_mul_ps(fscal,dx00);
 226             ty               = _mm_mul_ps(fscal,dy00);
 227             tz               = _mm_mul_ps(fscal,dz00);
 228
 229             /* Update vectorial force */
 230             fix0             = _mm_add_ps(fix0,tx);
 231             fiy0             = _mm_add_ps(fiy0,ty);
 232             fiz0             = _mm_add_ps(fiz0,tz);
 233
 234             fjptrA             = f+j_coord_offsetA;
 235             fjptrB             = f+j_coord_offsetB;
 236             fjptrC             = f+j_coord_offsetC;
 237             fjptrD             = f+j_coord_offsetD;
 238             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 239
 240             /* Inner loop uses 41 flops */
 241         }
 242
 243         if(jidx<j_index_end)
 244         {
 245
 246             /* Get j neighbor index, and coordinate index */
 247             jnrlistA         = jjnr[jidx];
 248             jnrlistB         = jjnr[jidx+1];
 249             jnrlistC         = jjnr[jidx+2];
 250             jnrlistD         = jjnr[jidx+3];
 251             /* Sign of each element will be negative for non-real atoms.
 252              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 253              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 254              */
 255             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 256             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 257             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 258             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 259             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 260             j_coord_offsetA  = DIM*jnrA;
 261             j_coord_offsetB  = DIM*jnrB;
 262             j_coord_offsetC  = DIM*jnrC;
 263             j_coord_offsetD  = DIM*jnrD;
 264
 265             /* load j atom coordinates */
 266             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 267                                               x+j_coord_offsetC,x+j_coord_offsetD,
 268                                               &jx0,&jy0,&jz0);
 269
 270             /* Calculate displacement vector */
 271             dx00             = _mm_sub_ps(ix0,jx0);
 272             dy00             = _mm_sub_ps(iy0,jy0);
 273             dz00             = _mm_sub_ps(iz0,jz0);
 274
 275             /* Calculate squared distance and things based on it */
 276             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 277
 278             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 279
 280             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 281
 282             /* Load parameters for j particles */
 283             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 284                                                               charge+jnrC+0,charge+jnrD+0);
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             r00              = _mm_mul_ps(rsq00,rinv00);
 291             r00              = _mm_andnot_ps(dummy_mask,r00);
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             qq00             = _mm_mul_ps(iq0,jq0);
 295
 296             /* EWALD ELECTROSTATICS */
 297
 298             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 299             ewrt             = _mm_mul_ps(r00,ewtabscale);
 300             ewitab           = _mm_cvttps_epi32(ewrt);
 301             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 302             ewitab           = _mm_slli_epi32(ewitab,2);
 303             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 304             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 305             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 306             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 307             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 308             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 309             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 310             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 311             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velec            = _mm_andnot_ps(dummy_mask,velec);
 315             velecsum         = _mm_add_ps(velecsum,velec);
 316
 317             fscal            = felec;
 318
 319             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 320
 321             /* Calculate temporary vectorial force */
 322             tx               = _mm_mul_ps(fscal,dx00);
 323             ty               = _mm_mul_ps(fscal,dy00);
 324             tz               = _mm_mul_ps(fscal,dz00);
 325
 326             /* Update vectorial force */
 327             fix0             = _mm_add_ps(fix0,tx);
 328             fiy0             = _mm_add_ps(fiy0,ty);
 329             fiz0             = _mm_add_ps(fiz0,tz);
 330
 331             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 332             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 333             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 334             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 335             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 336
 337             /* Inner loop uses 42 flops */
 338         }
 339
 340         /* End of innermost loop */
 341
 342         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 343                                               f+i_coord_offset,fshift+i_shift_offset);
 344
 345         ggid                        = gid[iidx];
 346         /* Update potential energies */
 347         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 348
 349         /* Increment number of inner iterations */
 350         inneriter                  += j_index_end - j_index_start;
 351
 352         /* Outer loop uses 8 flops */
 353     }
 354
 355     /* Increment number of outer iterations */
 356     outeriter        += nri;
 357
 358     /* Update outer/inner flops */
 359
 360     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*42);
 361 }
 362 /*
 363  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_single
 364  * Electrostatics interaction: Ewald
 365  * VdW interaction:            None
 366  * Geometry:                   Particle-Particle
 367  * Calculate force/pot:        Force
 368  */
 369 void
 370 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_single
 371                     (t_nblist                    * gmx_restrict       nlist,
 372                      rvec                        * gmx_restrict          xx,
 373                      rvec                        * gmx_restrict          ff,
 374                      t_forcerec                  * gmx_restrict          fr,
 375                      t_mdatoms                   * gmx_restrict     mdatoms,
 376                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 377                      t_nrnb                      * gmx_restrict        nrnb)
 378 {
 379     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 380      * just 0 for non-waters.
 381      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 382      * jnr indices corresponding to data put in the four positions in the SIMD register.
 383      */
 384     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 385     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 386     int              jnrA,jnrB,jnrC,jnrD;
 387     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 388     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 389     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 390     real             rcutoff_scalar;
 391     real             *shiftvec,*fshift,*x,*f;
 392     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 393     real             scratch[4*DIM];
 394     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 395     int              vdwioffset0;
 396     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 397     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 398     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 399     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 400     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 401     real             *charge;
 402     __m128i          ewitab;
 403     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 404     real             *ewtab;
 405     __m128           dummy_mask,cutoff_mask;
 406     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 407     __m128           one     = _mm_set1_ps(1.0);
 408     __m128           two     = _mm_set1_ps(2.0);
 409     x                = xx[0];
 410     f                = ff[0];
 411
 412     nri              = nlist->nri;
 413     iinr             = nlist->iinr;
 414     jindex           = nlist->jindex;
 415     jjnr             = nlist->jjnr;
 416     shiftidx         = nlist->shift;
 417     gid              = nlist->gid;
 418     shiftvec         = fr->shift_vec[0];
 419     fshift           = fr->fshift[0];
 420     facel            = _mm_set1_ps(fr->epsfac);
 421     charge           = mdatoms->chargeA;
 422
 423     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 424     ewtab            = fr->ic->tabq_coul_F;
 425     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 426     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 427
 428     /* Avoid stupid compiler warnings */
 429     jnrA = jnrB = jnrC = jnrD = 0;
 430     j_coord_offsetA = 0;
 431     j_coord_offsetB = 0;
 432     j_coord_offsetC = 0;
 433     j_coord_offsetD = 0;
 434
 435     outeriter        = 0;
 436     inneriter        = 0;
 437
 438     for(iidx=0;iidx<4*DIM;iidx++)
 439     {
 440         scratch[iidx] = 0.0;
 441     }
 442
 443     /* Start outer loop over neighborlists */
 444     for(iidx=0; iidx<nri; iidx++)
 445     {
 446         /* Load shift vector for this list */
 447         i_shift_offset   = DIM*shiftidx[iidx];
 448
 449         /* Load limits for loop over neighbors */
 450         j_index_start    = jindex[iidx];
 451         j_index_end      = jindex[iidx+1];
 452
 453         /* Get outer coordinate index */
 454         inr              = iinr[iidx];
 455         i_coord_offset   = DIM*inr;
 456
 457         /* Load i particle coords and add shift vector */
 458         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 459
 460         fix0             = _mm_setzero_ps();
 461         fiy0             = _mm_setzero_ps();
 462         fiz0             = _mm_setzero_ps();
 463
 464         /* Load parameters for i particles */
 465         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 466
 467         /* Start inner kernel loop */
 468         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 469         {
 470
 471             /* Get j neighbor index, and coordinate index */
 472             jnrA             = jjnr[jidx];
 473             jnrB             = jjnr[jidx+1];
 474             jnrC             = jjnr[jidx+2];
 475             jnrD             = jjnr[jidx+3];
 476             j_coord_offsetA  = DIM*jnrA;
 477             j_coord_offsetB  = DIM*jnrB;
 478             j_coord_offsetC  = DIM*jnrC;
 479             j_coord_offsetD  = DIM*jnrD;
 480
 481             /* load j atom coordinates */
 482             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                               x+j_coord_offsetC,x+j_coord_offsetD,
 484                                               &jx0,&jy0,&jz0);
 485
 486             /* Calculate displacement vector */
 487             dx00             = _mm_sub_ps(ix0,jx0);
 488             dy00             = _mm_sub_ps(iy0,jy0);
 489             dz00             = _mm_sub_ps(iz0,jz0);
 490
 491             /* Calculate squared distance and things based on it */
 492             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 493
 494             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 495
 496             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 497
 498             /* Load parameters for j particles */
 499             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 500                                                               charge+jnrC+0,charge+jnrD+0);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r00              = _mm_mul_ps(rsq00,rinv00);
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             qq00             = _mm_mul_ps(iq0,jq0);
 510
 511             /* EWALD ELECTROSTATICS */
 512
 513             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 514             ewrt             = _mm_mul_ps(r00,ewtabscale);
 515             ewitab           = _mm_cvttps_epi32(ewrt);
 516             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 517             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 518                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 519                                          &ewtabF,&ewtabFn);
 520             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 521             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 522
 523             fscal            = felec;
 524
 525             /* Calculate temporary vectorial force */
 526             tx               = _mm_mul_ps(fscal,dx00);
 527             ty               = _mm_mul_ps(fscal,dy00);
 528             tz               = _mm_mul_ps(fscal,dz00);
 529
 530             /* Update vectorial force */
 531             fix0             = _mm_add_ps(fix0,tx);
 532             fiy0             = _mm_add_ps(fiy0,ty);
 533             fiz0             = _mm_add_ps(fiz0,tz);
 534
 535             fjptrA             = f+j_coord_offsetA;
 536             fjptrB             = f+j_coord_offsetB;
 537             fjptrC             = f+j_coord_offsetC;
 538             fjptrD             = f+j_coord_offsetD;
 539             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 540
 541             /* Inner loop uses 36 flops */
 542         }
 543
 544         if(jidx<j_index_end)
 545         {
 546
 547             /* Get j neighbor index, and coordinate index */
 548             jnrlistA         = jjnr[jidx];
 549             jnrlistB         = jjnr[jidx+1];
 550             jnrlistC         = jjnr[jidx+2];
 551             jnrlistD         = jjnr[jidx+3];
 552             /* Sign of each element will be negative for non-real atoms.
 553              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 554              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 555              */
 556             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 557             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 558             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 559             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 560             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 561             j_coord_offsetA  = DIM*jnrA;
 562             j_coord_offsetB  = DIM*jnrB;
 563             j_coord_offsetC  = DIM*jnrC;
 564             j_coord_offsetD  = DIM*jnrD;
 565
 566             /* load j atom coordinates */
 567             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 568                                               x+j_coord_offsetC,x+j_coord_offsetD,
 569                                               &jx0,&jy0,&jz0);
 570
 571             /* Calculate displacement vector */
 572             dx00             = _mm_sub_ps(ix0,jx0);
 573             dy00             = _mm_sub_ps(iy0,jy0);
 574             dz00             = _mm_sub_ps(iz0,jz0);
 575
 576             /* Calculate squared distance and things based on it */
 577             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 578
 579             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 580
 581             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 582
 583             /* Load parameters for j particles */
 584             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 585                                                               charge+jnrC+0,charge+jnrD+0);
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             r00              = _mm_mul_ps(rsq00,rinv00);
 592             r00              = _mm_andnot_ps(dummy_mask,r00);
 593
 594             /* Compute parameters for interactions between i and j atoms */
 595             qq00             = _mm_mul_ps(iq0,jq0);
 596
 597             /* EWALD ELECTROSTATICS */
 598
 599             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 600             ewrt             = _mm_mul_ps(r00,ewtabscale);
 601             ewitab           = _mm_cvttps_epi32(ewrt);
 602             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 603             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 604                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 605                                          &ewtabF,&ewtabFn);
 606             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 607             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_ps(fscal,dx00);
 615             ty               = _mm_mul_ps(fscal,dy00);
 616             tz               = _mm_mul_ps(fscal,dz00);
 617
 618             /* Update vectorial force */
 619             fix0             = _mm_add_ps(fix0,tx);
 620             fiy0             = _mm_add_ps(fiy0,ty);
 621             fiz0             = _mm_add_ps(fiz0,tz);
 622
 623             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 624             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 625             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 626             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 627             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 628
 629             /* Inner loop uses 37 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 635                                               f+i_coord_offset,fshift+i_shift_offset);
 636
 637         /* Increment number of inner iterations */
 638         inneriter                  += j_index_end - j_index_start;
 639
 640         /* Outer loop uses 7 flops */
 641     }
 642
 643     /* Increment number of outer iterations */
 644     outeriter        += nri;
 645
 646     /* Update outer/inner flops */
 647
 648     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*37);
 649 }