src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecGB_VdwNone_GeomP1P1_sse4_1_double.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_double 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_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: GeneralizedBorn
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     __m128i          gbitab;
  88     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  89     __m128d          minushalf = _mm_set1_pd(-0.5);
  90     real             *invsqrta,*dvda,*gbtab;
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     invsqrta         = fr->invsqrta;
 114     dvda             = fr->dvda;
 115     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 116     gbtab            = fr->gbtab.data;
 117     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 143
 144         fix0             = _mm_setzero_pd();
 145         fiy0             = _mm_setzero_pd();
 146         fiz0             = _mm_setzero_pd();
 147
 148         /* Load parameters for i particles */
 149         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 150         isai0            = _mm_load1_pd(invsqrta+inr+0);
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_pd();
 154         vgbsum           = _mm_setzero_pd();
 155         dvdasum          = _mm_setzero_pd();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_pd(ix0,jx0);
 173             dy00             = _mm_sub_pd(iy0,jy0);
 174             dz00             = _mm_sub_pd(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 183             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 184
 185             /**************************
 186              * CALCULATE INTERACTIONS *
 187              **************************/
 188
 189             r00              = _mm_mul_pd(rsq00,rinv00);
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm_mul_pd(iq0,jq0);
 193
 194             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 195             isaprod          = _mm_mul_pd(isai0,isaj0);
 196             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 197             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 198
 199             /* Calculate generalized born table index - this is a separate table from the normal one,
 200              * but we use the same procedure by multiplying r with scale and truncating to integer.
 201              */
 202             rt               = _mm_mul_pd(r00,gbscale);
 203             gbitab           = _mm_cvttpd_epi32(rt);
 204             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 205             gbitab           = _mm_slli_epi32(gbitab,2);
 206
 207             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 208             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 209             GMX_MM_TRANSPOSE2_PD(Y,F);
 210             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 211             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 212             GMX_MM_TRANSPOSE2_PD(G,H);
 213             Heps             = _mm_mul_pd(gbeps,H);
 214             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 215             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 216             vgb              = _mm_mul_pd(gbqqfactor,VV);
 217
 218             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 219             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 220             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 221             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 222             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 223             velec            = _mm_mul_pd(qq00,rinv00);
 224             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             velecsum         = _mm_add_pd(velecsum,velec);
 228             vgbsum           = _mm_add_pd(vgbsum,vgb);
 229
 230             fscal            = felec;
 231
 232             /* Calculate temporary vectorial force */
 233             tx               = _mm_mul_pd(fscal,dx00);
 234             ty               = _mm_mul_pd(fscal,dy00);
 235             tz               = _mm_mul_pd(fscal,dz00);
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_add_pd(fix0,tx);
 239             fiy0             = _mm_add_pd(fiy0,ty);
 240             fiz0             = _mm_add_pd(fiz0,tz);
 241
 242             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 243
 244             /* Inner loop uses 58 flops */
 245         }
 246
 247         if(jidx<j_index_end)
 248         {
 249
 250             jnrA             = jjnr[jidx];
 251             j_coord_offsetA  = DIM*jnrA;
 252
 253             /* load j atom coordinates */
 254             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 255                                               &jx0,&jy0,&jz0);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _mm_sub_pd(ix0,jx0);
 259             dy00             = _mm_sub_pd(iy0,jy0);
 260             dz00             = _mm_sub_pd(iz0,jz0);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 264
 265             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 266
 267             /* Load parameters for j particles */
 268             jq0              = _mm_load_sd(charge+jnrA+0);
 269             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             r00              = _mm_mul_pd(rsq00,rinv00);
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             qq00             = _mm_mul_pd(iq0,jq0);
 279
 280             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 281             isaprod          = _mm_mul_pd(isai0,isaj0);
 282             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 283             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 284
 285             /* Calculate generalized born table index - this is a separate table from the normal one,
 286              * but we use the same procedure by multiplying r with scale and truncating to integer.
 287              */
 288             rt               = _mm_mul_pd(r00,gbscale);
 289             gbitab           = _mm_cvttpd_epi32(rt);
 290             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 291             gbitab           = _mm_slli_epi32(gbitab,2);
 292
 293             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 294             F                = _mm_setzero_pd();
 295             GMX_MM_TRANSPOSE2_PD(Y,F);
 296             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 297             H                = _mm_setzero_pd();
 298             GMX_MM_TRANSPOSE2_PD(G,H);
 299             Heps             = _mm_mul_pd(gbeps,H);
 300             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 301             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 302             vgb              = _mm_mul_pd(gbqqfactor,VV);
 303
 304             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 305             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 306             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 307             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 308             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 309             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 310             velec            = _mm_mul_pd(qq00,rinv00);
 311             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 315             velecsum         = _mm_add_pd(velecsum,velec);
 316             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 317             vgbsum           = _mm_add_pd(vgbsum,vgb);
 318
 319             fscal            = felec;
 320
 321             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm_mul_pd(fscal,dx00);
 325             ty               = _mm_mul_pd(fscal,dy00);
 326             tz               = _mm_mul_pd(fscal,dz00);
 327
 328             /* Update vectorial force */
 329             fix0             = _mm_add_pd(fix0,tx);
 330             fiy0             = _mm_add_pd(fiy0,ty);
 331             fiz0             = _mm_add_pd(fiz0,tz);
 332
 333             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 334
 335             /* Inner loop uses 58 flops */
 336         }
 337
 338         /* End of innermost loop */
 339
 340         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 341                                               f+i_coord_offset,fshift+i_shift_offset);
 342
 343         ggid                        = gid[iidx];
 344         /* Update potential energies */
 345         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 346         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 347         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 348         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 349
 350         /* Increment number of inner iterations */
 351         inneriter                  += j_index_end - j_index_start;
 352
 353         /* Outer loop uses 9 flops */
 354     }
 355
 356     /* Increment number of outer iterations */
 357     outeriter        += nri;
 358
 359     /* Update outer/inner flops */
 360
 361     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*58);
 362 }
 363 /*
 364  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
 365  * Electrostatics interaction: GeneralizedBorn
 366  * VdW interaction:            None
 367  * Geometry:                   Particle-Particle
 368  * Calculate force/pot:        Force
 369  */
 370 void
 371 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
 372                     (t_nblist                    * gmx_restrict       nlist,
 373                      rvec                        * gmx_restrict          xx,
 374                      rvec                        * gmx_restrict          ff,
 375                      t_forcerec                  * gmx_restrict          fr,
 376                      t_mdatoms                   * gmx_restrict     mdatoms,
 377                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 378                      t_nrnb                      * gmx_restrict        nrnb)
 379 {
 380     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 381      * just 0 for non-waters.
 382      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 383      * jnr indices corresponding to data put in the four positions in the SIMD register.
 384      */
 385     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 386     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 387     int              jnrA,jnrB;
 388     int              j_coord_offsetA,j_coord_offsetB;
 389     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 390     real             rcutoff_scalar;
 391     real             *shiftvec,*fshift,*x,*f;
 392     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 393     int              vdwioffset0;
 394     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 395     int              vdwjidx0A,vdwjidx0B;
 396     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 397     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 398     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 399     real             *charge;
 400     __m128i          gbitab;
 401     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 402     __m128d          minushalf = _mm_set1_pd(-0.5);
 403     real             *invsqrta,*dvda,*gbtab;
 404     __m128i          vfitab;
 405     __m128i          ifour       = _mm_set1_epi32(4);
 406     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 407     real             *vftab;
 408     __m128d          dummy_mask,cutoff_mask;
 409     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 410     __m128d          one     = _mm_set1_pd(1.0);
 411     __m128d          two     = _mm_set1_pd(2.0);
 412     x                = xx[0];
 413     f                = ff[0];
 414
 415     nri              = nlist->nri;
 416     iinr             = nlist->iinr;
 417     jindex           = nlist->jindex;
 418     jjnr             = nlist->jjnr;
 419     shiftidx         = nlist->shift;
 420     gid              = nlist->gid;
 421     shiftvec         = fr->shift_vec[0];
 422     fshift           = fr->fshift[0];
 423     facel            = _mm_set1_pd(fr->epsfac);
 424     charge           = mdatoms->chargeA;
 425
 426     invsqrta         = fr->invsqrta;
 427     dvda             = fr->dvda;
 428     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 429     gbtab            = fr->gbtab.data;
 430     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 431
 432     /* Avoid stupid compiler warnings */
 433     jnrA = jnrB = 0;
 434     j_coord_offsetA = 0;
 435     j_coord_offsetB = 0;
 436
 437     outeriter        = 0;
 438     inneriter        = 0;
 439
 440     /* Start outer loop over neighborlists */
 441     for(iidx=0; iidx<nri; iidx++)
 442     {
 443         /* Load shift vector for this list */
 444         i_shift_offset   = DIM*shiftidx[iidx];
 445
 446         /* Load limits for loop over neighbors */
 447         j_index_start    = jindex[iidx];
 448         j_index_end      = jindex[iidx+1];
 449
 450         /* Get outer coordinate index */
 451         inr              = iinr[iidx];
 452         i_coord_offset   = DIM*inr;
 453
 454         /* Load i particle coords and add shift vector */
 455         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 456
 457         fix0             = _mm_setzero_pd();
 458         fiy0             = _mm_setzero_pd();
 459         fiz0             = _mm_setzero_pd();
 460
 461         /* Load parameters for i particles */
 462         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 463         isai0            = _mm_load1_pd(invsqrta+inr+0);
 464
 465         dvdasum          = _mm_setzero_pd();
 466
 467         /* Start inner kernel loop */
 468         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 469         {
 470
 471             /* Get j neighbor index, and coordinate index */
 472             jnrA             = jjnr[jidx];
 473             jnrB             = jjnr[jidx+1];
 474             j_coord_offsetA  = DIM*jnrA;
 475             j_coord_offsetB  = DIM*jnrB;
 476
 477             /* load j atom coordinates */
 478             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 479                                               &jx0,&jy0,&jz0);
 480
 481             /* Calculate displacement vector */
 482             dx00             = _mm_sub_pd(ix0,jx0);
 483             dy00             = _mm_sub_pd(iy0,jy0);
 484             dz00             = _mm_sub_pd(iz0,jz0);
 485
 486             /* Calculate squared distance and things based on it */
 487             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 488
 489             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 490
 491             /* Load parameters for j particles */
 492             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 493             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             r00              = _mm_mul_pd(rsq00,rinv00);
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             qq00             = _mm_mul_pd(iq0,jq0);
 503
 504             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 505             isaprod          = _mm_mul_pd(isai0,isaj0);
 506             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 507             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 508
 509             /* Calculate generalized born table index - this is a separate table from the normal one,
 510              * but we use the same procedure by multiplying r with scale and truncating to integer.
 511              */
 512             rt               = _mm_mul_pd(r00,gbscale);
 513             gbitab           = _mm_cvttpd_epi32(rt);
 514             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 515             gbitab           = _mm_slli_epi32(gbitab,2);
 516
 517             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 518             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 519             GMX_MM_TRANSPOSE2_PD(Y,F);
 520             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 521             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 522             GMX_MM_TRANSPOSE2_PD(G,H);
 523             Heps             = _mm_mul_pd(gbeps,H);
 524             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 525             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 526             vgb              = _mm_mul_pd(gbqqfactor,VV);
 527
 528             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 529             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 530             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 531             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 532             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 533             velec            = _mm_mul_pd(qq00,rinv00);
 534             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 535
 536             fscal            = felec;
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm_mul_pd(fscal,dx00);
 540             ty               = _mm_mul_pd(fscal,dy00);
 541             tz               = _mm_mul_pd(fscal,dz00);
 542
 543             /* Update vectorial force */
 544             fix0             = _mm_add_pd(fix0,tx);
 545             fiy0             = _mm_add_pd(fiy0,ty);
 546             fiz0             = _mm_add_pd(fiz0,tz);
 547
 548             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 549
 550             /* Inner loop uses 56 flops */
 551         }
 552
 553         if(jidx<j_index_end)
 554         {
 555
 556             jnrA             = jjnr[jidx];
 557             j_coord_offsetA  = DIM*jnrA;
 558
 559             /* load j atom coordinates */
 560             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 561                                               &jx0,&jy0,&jz0);
 562
 563             /* Calculate displacement vector */
 564             dx00             = _mm_sub_pd(ix0,jx0);
 565             dy00             = _mm_sub_pd(iy0,jy0);
 566             dz00             = _mm_sub_pd(iz0,jz0);
 567
 568             /* Calculate squared distance and things based on it */
 569             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 570
 571             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 572
 573             /* Load parameters for j particles */
 574             jq0              = _mm_load_sd(charge+jnrA+0);
 575             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             r00              = _mm_mul_pd(rsq00,rinv00);
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq00             = _mm_mul_pd(iq0,jq0);
 585
 586             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 587             isaprod          = _mm_mul_pd(isai0,isaj0);
 588             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 589             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 590
 591             /* Calculate generalized born table index - this is a separate table from the normal one,
 592              * but we use the same procedure by multiplying r with scale and truncating to integer.
 593              */
 594             rt               = _mm_mul_pd(r00,gbscale);
 595             gbitab           = _mm_cvttpd_epi32(rt);
 596             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 597             gbitab           = _mm_slli_epi32(gbitab,2);
 598
 599             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 600             F                = _mm_setzero_pd();
 601             GMX_MM_TRANSPOSE2_PD(Y,F);
 602             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 603             H                = _mm_setzero_pd();
 604             GMX_MM_TRANSPOSE2_PD(G,H);
 605             Heps             = _mm_mul_pd(gbeps,H);
 606             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 607             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 608             vgb              = _mm_mul_pd(gbqqfactor,VV);
 609
 610             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 611             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 612             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 613             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 614             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 615             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 616             velec            = _mm_mul_pd(qq00,rinv00);
 617             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm_mul_pd(fscal,dx00);
 625             ty               = _mm_mul_pd(fscal,dy00);
 626             tz               = _mm_mul_pd(fscal,dz00);
 627
 628             /* Update vectorial force */
 629             fix0             = _mm_add_pd(fix0,tx);
 630             fiy0             = _mm_add_pd(fiy0,ty);
 631             fiz0             = _mm_add_pd(fiz0,tz);
 632
 633             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 634
 635             /* Inner loop uses 56 flops */
 636         }
 637
 638         /* End of innermost loop */
 639
 640         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 641                                               f+i_coord_offset,fshift+i_shift_offset);
 642
 643         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 644         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 645
 646         /* Increment number of inner iterations */
 647         inneriter                  += j_index_end - j_index_start;
 648
 649         /* Outer loop uses 7 flops */
 650     }
 651
 652     /* Increment number of outer iterations */
 653     outeriter        += nri;
 654
 655     /* Update outer/inner flops */
 656
 657     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*56);
 658 }