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) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_double
  51  * Electrostatics interaction: GeneralizedBorn
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     __m128i          gbitab;
  87     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  88     __m128d          minushalf = _mm_set1_pd(-0.5);
  89     real             *invsqrta,*dvda,*gbtab;
  90     __m128i          vfitab;
  91     __m128i          ifour       = _mm_set1_epi32(4);
  92     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  93     real             *vftab;
  94     __m128d          dummy_mask,cutoff_mask;
  95     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  96     __m128d          one     = _mm_set1_pd(1.0);
  97     __m128d          two     = _mm_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm_set1_pd(fr->ic->epsfac);
 110     charge           = mdatoms->chargeA;
 111
 112     invsqrta         = fr->invsqrta;
 113     dvda             = fr->dvda;
 114     gbtabscale       = _mm_set1_pd(fr->gbtab->scale);
 115     gbtab            = fr->gbtab->data;
 116     gbinvepsdiff     = _mm_set1_pd((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _mm_setzero_pd();
 144         fiy0             = _mm_setzero_pd();
 145         fiz0             = _mm_setzero_pd();
 146
 147         /* Load parameters for i particles */
 148         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 149         isai0            = _mm_load1_pd(invsqrta+inr+0);
 150
 151         /* Reset potential sums */
 152         velecsum         = _mm_setzero_pd();
 153         vgbsum           = _mm_setzero_pd();
 154         dvdasum          = _mm_setzero_pd();
 155
 156         /* Start inner kernel loop */
 157         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 158         {
 159
 160             /* Get j neighbor index, and coordinate index */
 161             jnrA             = jjnr[jidx];
 162             jnrB             = jjnr[jidx+1];
 163             j_coord_offsetA  = DIM*jnrA;
 164             j_coord_offsetB  = DIM*jnrB;
 165
 166             /* load j atom coordinates */
 167             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                               &jx0,&jy0,&jz0);
 169
 170             /* Calculate displacement vector */
 171             dx00             = _mm_sub_pd(ix0,jx0);
 172             dy00             = _mm_sub_pd(iy0,jy0);
 173             dz00             = _mm_sub_pd(iz0,jz0);
 174
 175             /* Calculate squared distance and things based on it */
 176             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 177
 178             rinv00           = sse41_invsqrt_d(rsq00);
 179
 180             /* Load parameters for j particles */
 181             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 182             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 183
 184             /**************************
 185              * CALCULATE INTERACTIONS *
 186              **************************/
 187
 188             r00              = _mm_mul_pd(rsq00,rinv00);
 189
 190             /* Compute parameters for interactions between i and j atoms */
 191             qq00             = _mm_mul_pd(iq0,jq0);
 192
 193             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 194             isaprod          = _mm_mul_pd(isai0,isaj0);
 195             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 196             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 197
 198             /* Calculate generalized born table index - this is a separate table from the normal one,
 199              * but we use the same procedure by multiplying r with scale and truncating to integer.
 200              */
 201             rt               = _mm_mul_pd(r00,gbscale);
 202             gbitab           = _mm_cvttpd_epi32(rt);
 203             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 204             gbitab           = _mm_slli_epi32(gbitab,2);
 205
 206             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 207             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 208             GMX_MM_TRANSPOSE2_PD(Y,F);
 209             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 210             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 211             GMX_MM_TRANSPOSE2_PD(G,H);
 212             Heps             = _mm_mul_pd(gbeps,H);
 213             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 214             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 215             vgb              = _mm_mul_pd(gbqqfactor,VV);
 216
 217             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 218             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 219             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 220             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 221             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 222             velec            = _mm_mul_pd(qq00,rinv00);
 223             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 224
 225             /* Update potential sum for this i atom from the interaction with this j atom. */
 226             velecsum         = _mm_add_pd(velecsum,velec);
 227             vgbsum           = _mm_add_pd(vgbsum,vgb);
 228
 229             fscal            = felec;
 230
 231             /* Calculate temporary vectorial force */
 232             tx               = _mm_mul_pd(fscal,dx00);
 233             ty               = _mm_mul_pd(fscal,dy00);
 234             tz               = _mm_mul_pd(fscal,dz00);
 235
 236             /* Update vectorial force */
 237             fix0             = _mm_add_pd(fix0,tx);
 238             fiy0             = _mm_add_pd(fiy0,ty);
 239             fiz0             = _mm_add_pd(fiz0,tz);
 240
 241             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 242
 243             /* Inner loop uses 58 flops */
 244         }
 245
 246         if(jidx<j_index_end)
 247         {
 248
 249             jnrA             = jjnr[jidx];
 250             j_coord_offsetA  = DIM*jnrA;
 251
 252             /* load j atom coordinates */
 253             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 254                                               &jx0,&jy0,&jz0);
 255
 256             /* Calculate displacement vector */
 257             dx00             = _mm_sub_pd(ix0,jx0);
 258             dy00             = _mm_sub_pd(iy0,jy0);
 259             dz00             = _mm_sub_pd(iz0,jz0);
 260
 261             /* Calculate squared distance and things based on it */
 262             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 263
 264             rinv00           = sse41_invsqrt_d(rsq00);
 265
 266             /* Load parameters for j particles */
 267             jq0              = _mm_load_sd(charge+jnrA+0);
 268             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r00              = _mm_mul_pd(rsq00,rinv00);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq00             = _mm_mul_pd(iq0,jq0);
 278
 279             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 280             isaprod          = _mm_mul_pd(isai0,isaj0);
 281             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 282             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 283
 284             /* Calculate generalized born table index - this is a separate table from the normal one,
 285              * but we use the same procedure by multiplying r with scale and truncating to integer.
 286              */
 287             rt               = _mm_mul_pd(r00,gbscale);
 288             gbitab           = _mm_cvttpd_epi32(rt);
 289             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 290             gbitab           = _mm_slli_epi32(gbitab,2);
 291
 292             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 293             F                = _mm_setzero_pd();
 294             GMX_MM_TRANSPOSE2_PD(Y,F);
 295             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 296             H                = _mm_setzero_pd();
 297             GMX_MM_TRANSPOSE2_PD(G,H);
 298             Heps             = _mm_mul_pd(gbeps,H);
 299             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 300             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 301             vgb              = _mm_mul_pd(gbqqfactor,VV);
 302
 303             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 304             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 305             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 306             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 307             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 308             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 309             velec            = _mm_mul_pd(qq00,rinv00);
 310             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 314             velecsum         = _mm_add_pd(velecsum,velec);
 315             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 316             vgbsum           = _mm_add_pd(vgbsum,vgb);
 317
 318             fscal            = felec;
 319
 320             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm_mul_pd(fscal,dx00);
 324             ty               = _mm_mul_pd(fscal,dy00);
 325             tz               = _mm_mul_pd(fscal,dz00);
 326
 327             /* Update vectorial force */
 328             fix0             = _mm_add_pd(fix0,tx);
 329             fiy0             = _mm_add_pd(fiy0,ty);
 330             fiz0             = _mm_add_pd(fiz0,tz);
 331
 332             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 333
 334             /* Inner loop uses 58 flops */
 335         }
 336
 337         /* End of innermost loop */
 338
 339         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 340                                               f+i_coord_offset,fshift+i_shift_offset);
 341
 342         ggid                        = gid[iidx];
 343         /* Update potential energies */
 344         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 345         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 346         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 347         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 348
 349         /* Increment number of inner iterations */
 350         inneriter                  += j_index_end - j_index_start;
 351
 352         /* Outer loop uses 9 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*9 + inneriter*58);
 361 }
 362 /*
 363  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
 364  * Electrostatics interaction: GeneralizedBorn
 365  * VdW interaction:            None
 366  * Geometry:                   Particle-Particle
 367  * Calculate force/pot:        Force
 368  */
 369 void
 370 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
 371                     (t_nblist                    * gmx_restrict       nlist,
 372                      rvec                        * gmx_restrict          xx,
 373                      rvec                        * gmx_restrict          ff,
 374                      struct 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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
 387     int              j_coord_offsetA,j_coord_offsetB;
 388     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 389     real             rcutoff_scalar;
 390     real             *shiftvec,*fshift,*x,*f;
 391     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 392     int              vdwioffset0;
 393     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 394     int              vdwjidx0A,vdwjidx0B;
 395     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 396     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 397     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 398     real             *charge;
 399     __m128i          gbitab;
 400     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 401     __m128d          minushalf = _mm_set1_pd(-0.5);
 402     real             *invsqrta,*dvda,*gbtab;
 403     __m128i          vfitab;
 404     __m128i          ifour       = _mm_set1_epi32(4);
 405     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 406     real             *vftab;
 407     __m128d          dummy_mask,cutoff_mask;
 408     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 409     __m128d          one     = _mm_set1_pd(1.0);
 410     __m128d          two     = _mm_set1_pd(2.0);
 411     x                = xx[0];
 412     f                = ff[0];
 413
 414     nri              = nlist->nri;
 415     iinr             = nlist->iinr;
 416     jindex           = nlist->jindex;
 417     jjnr             = nlist->jjnr;
 418     shiftidx         = nlist->shift;
 419     gid              = nlist->gid;
 420     shiftvec         = fr->shift_vec[0];
 421     fshift           = fr->fshift[0];
 422     facel            = _mm_set1_pd(fr->ic->epsfac);
 423     charge           = mdatoms->chargeA;
 424
 425     invsqrta         = fr->invsqrta;
 426     dvda             = fr->dvda;
 427     gbtabscale       = _mm_set1_pd(fr->gbtab->scale);
 428     gbtab            = fr->gbtab->data;
 429     gbinvepsdiff     = _mm_set1_pd((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 430
 431     /* Avoid stupid compiler warnings */
 432     jnrA = jnrB = 0;
 433     j_coord_offsetA = 0;
 434     j_coord_offsetB = 0;
 435
 436     outeriter        = 0;
 437     inneriter        = 0;
 438
 439     /* Start outer loop over neighborlists */
 440     for(iidx=0; iidx<nri; iidx++)
 441     {
 442         /* Load shift vector for this list */
 443         i_shift_offset   = DIM*shiftidx[iidx];
 444
 445         /* Load limits for loop over neighbors */
 446         j_index_start    = jindex[iidx];
 447         j_index_end      = jindex[iidx+1];
 448
 449         /* Get outer coordinate index */
 450         inr              = iinr[iidx];
 451         i_coord_offset   = DIM*inr;
 452
 453         /* Load i particle coords and add shift vector */
 454         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 455
 456         fix0             = _mm_setzero_pd();
 457         fiy0             = _mm_setzero_pd();
 458         fiz0             = _mm_setzero_pd();
 459
 460         /* Load parameters for i particles */
 461         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 462         isai0            = _mm_load1_pd(invsqrta+inr+0);
 463
 464         dvdasum          = _mm_setzero_pd();
 465
 466         /* Start inner kernel loop */
 467         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 468         {
 469
 470             /* Get j neighbor index, and coordinate index */
 471             jnrA             = jjnr[jidx];
 472             jnrB             = jjnr[jidx+1];
 473             j_coord_offsetA  = DIM*jnrA;
 474             j_coord_offsetB  = DIM*jnrB;
 475
 476             /* load j atom coordinates */
 477             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 478                                               &jx0,&jy0,&jz0);
 479
 480             /* Calculate displacement vector */
 481             dx00             = _mm_sub_pd(ix0,jx0);
 482             dy00             = _mm_sub_pd(iy0,jy0);
 483             dz00             = _mm_sub_pd(iz0,jz0);
 484
 485             /* Calculate squared distance and things based on it */
 486             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 487
 488             rinv00           = sse41_invsqrt_d(rsq00);
 489
 490             /* Load parameters for j particles */
 491             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 492             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r00              = _mm_mul_pd(rsq00,rinv00);
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq00             = _mm_mul_pd(iq0,jq0);
 502
 503             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 504             isaprod          = _mm_mul_pd(isai0,isaj0);
 505             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 506             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 507
 508             /* Calculate generalized born table index - this is a separate table from the normal one,
 509              * but we use the same procedure by multiplying r with scale and truncating to integer.
 510              */
 511             rt               = _mm_mul_pd(r00,gbscale);
 512             gbitab           = _mm_cvttpd_epi32(rt);
 513             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 514             gbitab           = _mm_slli_epi32(gbitab,2);
 515
 516             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 517             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 518             GMX_MM_TRANSPOSE2_PD(Y,F);
 519             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 520             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 521             GMX_MM_TRANSPOSE2_PD(G,H);
 522             Heps             = _mm_mul_pd(gbeps,H);
 523             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 524             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 525             vgb              = _mm_mul_pd(gbqqfactor,VV);
 526
 527             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 528             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 529             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 530             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 531             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 532             velec            = _mm_mul_pd(qq00,rinv00);
 533             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 534
 535             fscal            = felec;
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm_mul_pd(fscal,dx00);
 539             ty               = _mm_mul_pd(fscal,dy00);
 540             tz               = _mm_mul_pd(fscal,dz00);
 541
 542             /* Update vectorial force */
 543             fix0             = _mm_add_pd(fix0,tx);
 544             fiy0             = _mm_add_pd(fiy0,ty);
 545             fiz0             = _mm_add_pd(fiz0,tz);
 546
 547             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 548
 549             /* Inner loop uses 56 flops */
 550         }
 551
 552         if(jidx<j_index_end)
 553         {
 554
 555             jnrA             = jjnr[jidx];
 556             j_coord_offsetA  = DIM*jnrA;
 557
 558             /* load j atom coordinates */
 559             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 560                                               &jx0,&jy0,&jz0);
 561
 562             /* Calculate displacement vector */
 563             dx00             = _mm_sub_pd(ix0,jx0);
 564             dy00             = _mm_sub_pd(iy0,jy0);
 565             dz00             = _mm_sub_pd(iz0,jz0);
 566
 567             /* Calculate squared distance and things based on it */
 568             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 569
 570             rinv00           = sse41_invsqrt_d(rsq00);
 571
 572             /* Load parameters for j particles */
 573             jq0              = _mm_load_sd(charge+jnrA+0);
 574             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             r00              = _mm_mul_pd(rsq00,rinv00);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq00             = _mm_mul_pd(iq0,jq0);
 584
 585             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 586             isaprod          = _mm_mul_pd(isai0,isaj0);
 587             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 588             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 589
 590             /* Calculate generalized born table index - this is a separate table from the normal one,
 591              * but we use the same procedure by multiplying r with scale and truncating to integer.
 592              */
 593             rt               = _mm_mul_pd(r00,gbscale);
 594             gbitab           = _mm_cvttpd_epi32(rt);
 595             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 596             gbitab           = _mm_slli_epi32(gbitab,2);
 597
 598             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 599             F                = _mm_setzero_pd();
 600             GMX_MM_TRANSPOSE2_PD(Y,F);
 601             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 602             H                = _mm_setzero_pd();
 603             GMX_MM_TRANSPOSE2_PD(G,H);
 604             Heps             = _mm_mul_pd(gbeps,H);
 605             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 606             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 607             vgb              = _mm_mul_pd(gbqqfactor,VV);
 608
 609             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 610             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 611             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 612             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 613             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 614             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 615             velec            = _mm_mul_pd(qq00,rinv00);
 616             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 621
 622             /* Calculate temporary vectorial force */
 623             tx               = _mm_mul_pd(fscal,dx00);
 624             ty               = _mm_mul_pd(fscal,dy00);
 625             tz               = _mm_mul_pd(fscal,dz00);
 626
 627             /* Update vectorial force */
 628             fix0             = _mm_add_pd(fix0,tx);
 629             fiy0             = _mm_add_pd(fiy0,ty);
 630             fiz0             = _mm_add_pd(fiz0,tz);
 631
 632             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 633
 634             /* Inner loop uses 56 flops */
 635         }
 636
 637         /* End of innermost loop */
 638
 639         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 640                                               f+i_coord_offset,fshift+i_shift_offset);
 641
 642         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 643         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 644
 645         /* Increment number of inner iterations */
 646         inneriter                  += j_index_end - j_index_start;
 647
 648         /* Outer loop uses 7 flops */
 649     }
 650
 651     /* Increment number of outer iterations */
 652     outeriter        += nri;
 653
 654     /* Update outer/inner flops */
 655
 656     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*56);
 657 }