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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_double
  54  * Electrostatics interaction: GeneralizedBorn
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          gbitab;
  90     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  91     __m128d          minushalf = _mm_set1_pd(-0.5);
  92     real             *invsqrta,*dvda,*gbtab;
  93     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  96     real             *vftab;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114
 115     invsqrta         = fr->invsqrta;
 116     dvda             = fr->dvda;
 117     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 118     gbtab            = fr->gbtab.data;
 119     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149
 150         /* Load parameters for i particles */
 151         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 152         isai0            = _mm_load1_pd(invsqrta+inr+0);
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_pd();
 156         vgbsum           = _mm_setzero_pd();
 157         dvdasum          = _mm_setzero_pd();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm_sub_pd(ix0,jx0);
 175             dy00             = _mm_sub_pd(iy0,jy0);
 176             dz00             = _mm_sub_pd(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 185             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 186
 187             /**************************
 188              * CALCULATE INTERACTIONS *
 189              **************************/
 190
 191             r00              = _mm_mul_pd(rsq00,rinv00);
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm_mul_pd(iq0,jq0);
 195
 196             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 197             isaprod          = _mm_mul_pd(isai0,isaj0);
 198             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 199             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 200
 201             /* Calculate generalized born table index - this is a separate table from the normal one,
 202              * but we use the same procedure by multiplying r with scale and truncating to integer.
 203              */
 204             rt               = _mm_mul_pd(r00,gbscale);
 205             gbitab           = _mm_cvttpd_epi32(rt);
 206             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 207             gbitab           = _mm_slli_epi32(gbitab,2);
 208
 209             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 210             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 211             GMX_MM_TRANSPOSE2_PD(Y,F);
 212             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 213             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 214             GMX_MM_TRANSPOSE2_PD(G,H);
 215             Heps             = _mm_mul_pd(gbeps,H);
 216             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 217             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 218             vgb              = _mm_mul_pd(gbqqfactor,VV);
 219
 220             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 221             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 222             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 223             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 224             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 225             velec            = _mm_mul_pd(qq00,rinv00);
 226             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 227
 228             /* Update potential sum for this i atom from the interaction with this j atom. */
 229             velecsum         = _mm_add_pd(velecsum,velec);
 230             vgbsum           = _mm_add_pd(vgbsum,vgb);
 231
 232             fscal            = felec;
 233
 234             /* Calculate temporary vectorial force */
 235             tx               = _mm_mul_pd(fscal,dx00);
 236             ty               = _mm_mul_pd(fscal,dy00);
 237             tz               = _mm_mul_pd(fscal,dz00);
 238
 239             /* Update vectorial force */
 240             fix0             = _mm_add_pd(fix0,tx);
 241             fiy0             = _mm_add_pd(fiy0,ty);
 242             fiz0             = _mm_add_pd(fiz0,tz);
 243
 244             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 245
 246             /* Inner loop uses 58 flops */
 247         }
 248
 249         if(jidx<j_index_end)
 250         {
 251
 252             jnrA             = jjnr[jidx];
 253             j_coord_offsetA  = DIM*jnrA;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 257                                               &jx0,&jy0,&jz0);
 258
 259             /* Calculate displacement vector */
 260             dx00             = _mm_sub_pd(ix0,jx0);
 261             dy00             = _mm_sub_pd(iy0,jy0);
 262             dz00             = _mm_sub_pd(iz0,jz0);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 266
 267             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 268
 269             /* Load parameters for j particles */
 270             jq0              = _mm_load_sd(charge+jnrA+0);
 271             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             r00              = _mm_mul_pd(rsq00,rinv00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq00             = _mm_mul_pd(iq0,jq0);
 281
 282             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 283             isaprod          = _mm_mul_pd(isai0,isaj0);
 284             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 285             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 286
 287             /* Calculate generalized born table index - this is a separate table from the normal one,
 288              * but we use the same procedure by multiplying r with scale and truncating to integer.
 289              */
 290             rt               = _mm_mul_pd(r00,gbscale);
 291             gbitab           = _mm_cvttpd_epi32(rt);
 292             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 293             gbitab           = _mm_slli_epi32(gbitab,2);
 294
 295             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 296             F                = _mm_setzero_pd();
 297             GMX_MM_TRANSPOSE2_PD(Y,F);
 298             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 299             H                = _mm_setzero_pd();
 300             GMX_MM_TRANSPOSE2_PD(G,H);
 301             Heps             = _mm_mul_pd(gbeps,H);
 302             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 303             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 304             vgb              = _mm_mul_pd(gbqqfactor,VV);
 305
 306             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 307             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 308             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 309             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 310             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 311             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 312             velec            = _mm_mul_pd(qq00,rinv00);
 313             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 317             velecsum         = _mm_add_pd(velecsum,velec);
 318             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 319             vgbsum           = _mm_add_pd(vgbsum,vgb);
 320
 321             fscal            = felec;
 322
 323             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm_mul_pd(fscal,dx00);
 327             ty               = _mm_mul_pd(fscal,dy00);
 328             tz               = _mm_mul_pd(fscal,dz00);
 329
 330             /* Update vectorial force */
 331             fix0             = _mm_add_pd(fix0,tx);
 332             fiy0             = _mm_add_pd(fiy0,ty);
 333             fiz0             = _mm_add_pd(fiz0,tz);
 334
 335             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 336
 337             /* Inner loop uses 58 flops */
 338         }
 339
 340         /* End of innermost loop */
 341
 342         gmx_mm_update_iforce_1atom_swizzle_pd(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_pd(velecsum,kernel_data->energygrp_elec+ggid);
 348         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 349         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 350         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 351
 352         /* Increment number of inner iterations */
 353         inneriter                  += j_index_end - j_index_start;
 354
 355         /* Outer loop uses 9 flops */
 356     }
 357
 358     /* Increment number of outer iterations */
 359     outeriter        += nri;
 360
 361     /* Update outer/inner flops */
 362
 363     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*58);
 364 }
 365 /*
 366  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
 367  * Electrostatics interaction: GeneralizedBorn
 368  * VdW interaction:            None
 369  * Geometry:                   Particle-Particle
 370  * Calculate force/pot:        Force
 371  */
 372 void
 373 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
 374                     (t_nblist                    * gmx_restrict       nlist,
 375                      rvec                        * gmx_restrict          xx,
 376                      rvec                        * gmx_restrict          ff,
 377                      t_forcerec                  * gmx_restrict          fr,
 378                      t_mdatoms                   * gmx_restrict     mdatoms,
 379                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 380                      t_nrnb                      * gmx_restrict        nrnb)
 381 {
 382     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 383      * just 0 for non-waters.
 384      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 385      * jnr indices corresponding to data put in the four positions in the SIMD register.
 386      */
 387     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 388     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 389     int              jnrA,jnrB;
 390     int              j_coord_offsetA,j_coord_offsetB;
 391     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 392     real             rcutoff_scalar;
 393     real             *shiftvec,*fshift,*x,*f;
 394     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 395     int              vdwioffset0;
 396     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 397     int              vdwjidx0A,vdwjidx0B;
 398     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 399     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 400     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 401     real             *charge;
 402     __m128i          gbitab;
 403     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 404     __m128d          minushalf = _mm_set1_pd(-0.5);
 405     real             *invsqrta,*dvda,*gbtab;
 406     __m128i          vfitab;
 407     __m128i          ifour       = _mm_set1_epi32(4);
 408     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 409     real             *vftab;
 410     __m128d          dummy_mask,cutoff_mask;
 411     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 412     __m128d          one     = _mm_set1_pd(1.0);
 413     __m128d          two     = _mm_set1_pd(2.0);
 414     x                = xx[0];
 415     f                = ff[0];
 416
 417     nri              = nlist->nri;
 418     iinr             = nlist->iinr;
 419     jindex           = nlist->jindex;
 420     jjnr             = nlist->jjnr;
 421     shiftidx         = nlist->shift;
 422     gid              = nlist->gid;
 423     shiftvec         = fr->shift_vec[0];
 424     fshift           = fr->fshift[0];
 425     facel            = _mm_set1_pd(fr->epsfac);
 426     charge           = mdatoms->chargeA;
 427
 428     invsqrta         = fr->invsqrta;
 429     dvda             = fr->dvda;
 430     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 431     gbtab            = fr->gbtab.data;
 432     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 433
 434     /* Avoid stupid compiler warnings */
 435     jnrA = jnrB = 0;
 436     j_coord_offsetA = 0;
 437     j_coord_offsetB = 0;
 438
 439     outeriter        = 0;
 440     inneriter        = 0;
 441
 442     /* Start outer loop over neighborlists */
 443     for(iidx=0; iidx<nri; iidx++)
 444     {
 445         /* Load shift vector for this list */
 446         i_shift_offset   = DIM*shiftidx[iidx];
 447
 448         /* Load limits for loop over neighbors */
 449         j_index_start    = jindex[iidx];
 450         j_index_end      = jindex[iidx+1];
 451
 452         /* Get outer coordinate index */
 453         inr              = iinr[iidx];
 454         i_coord_offset   = DIM*inr;
 455
 456         /* Load i particle coords and add shift vector */
 457         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 458
 459         fix0             = _mm_setzero_pd();
 460         fiy0             = _mm_setzero_pd();
 461         fiz0             = _mm_setzero_pd();
 462
 463         /* Load parameters for i particles */
 464         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 465         isai0            = _mm_load1_pd(invsqrta+inr+0);
 466
 467         dvdasum          = _mm_setzero_pd();
 468
 469         /* Start inner kernel loop */
 470         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 471         {
 472
 473             /* Get j neighbor index, and coordinate index */
 474             jnrA             = jjnr[jidx];
 475             jnrB             = jjnr[jidx+1];
 476             j_coord_offsetA  = DIM*jnrA;
 477             j_coord_offsetB  = DIM*jnrB;
 478
 479             /* load j atom coordinates */
 480             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 481                                               &jx0,&jy0,&jz0);
 482
 483             /* Calculate displacement vector */
 484             dx00             = _mm_sub_pd(ix0,jx0);
 485             dy00             = _mm_sub_pd(iy0,jy0);
 486             dz00             = _mm_sub_pd(iz0,jz0);
 487
 488             /* Calculate squared distance and things based on it */
 489             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 490
 491             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 492
 493             /* Load parameters for j particles */
 494             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 495             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 496
 497             /**************************
 498              * CALCULATE INTERACTIONS *
 499              **************************/
 500
 501             r00              = _mm_mul_pd(rsq00,rinv00);
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             qq00             = _mm_mul_pd(iq0,jq0);
 505
 506             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 507             isaprod          = _mm_mul_pd(isai0,isaj0);
 508             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 509             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 510
 511             /* Calculate generalized born table index - this is a separate table from the normal one,
 512              * but we use the same procedure by multiplying r with scale and truncating to integer.
 513              */
 514             rt               = _mm_mul_pd(r00,gbscale);
 515             gbitab           = _mm_cvttpd_epi32(rt);
 516             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 517             gbitab           = _mm_slli_epi32(gbitab,2);
 518
 519             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 520             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 521             GMX_MM_TRANSPOSE2_PD(Y,F);
 522             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 523             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 524             GMX_MM_TRANSPOSE2_PD(G,H);
 525             Heps             = _mm_mul_pd(gbeps,H);
 526             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 527             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 528             vgb              = _mm_mul_pd(gbqqfactor,VV);
 529
 530             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 531             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 532             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 533             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 534             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 535             velec            = _mm_mul_pd(qq00,rinv00);
 536             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 537
 538             fscal            = felec;
 539
 540             /* Calculate temporary vectorial force */
 541             tx               = _mm_mul_pd(fscal,dx00);
 542             ty               = _mm_mul_pd(fscal,dy00);
 543             tz               = _mm_mul_pd(fscal,dz00);
 544
 545             /* Update vectorial force */
 546             fix0             = _mm_add_pd(fix0,tx);
 547             fiy0             = _mm_add_pd(fiy0,ty);
 548             fiz0             = _mm_add_pd(fiz0,tz);
 549
 550             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 551
 552             /* Inner loop uses 56 flops */
 553         }
 554
 555         if(jidx<j_index_end)
 556         {
 557
 558             jnrA             = jjnr[jidx];
 559             j_coord_offsetA  = DIM*jnrA;
 560
 561             /* load j atom coordinates */
 562             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 563                                               &jx0,&jy0,&jz0);
 564
 565             /* Calculate displacement vector */
 566             dx00             = _mm_sub_pd(ix0,jx0);
 567             dy00             = _mm_sub_pd(iy0,jy0);
 568             dz00             = _mm_sub_pd(iz0,jz0);
 569
 570             /* Calculate squared distance and things based on it */
 571             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 572
 573             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 574
 575             /* Load parameters for j particles */
 576             jq0              = _mm_load_sd(charge+jnrA+0);
 577             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             r00              = _mm_mul_pd(rsq00,rinv00);
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             qq00             = _mm_mul_pd(iq0,jq0);
 587
 588             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 589             isaprod          = _mm_mul_pd(isai0,isaj0);
 590             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 591             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 592
 593             /* Calculate generalized born table index - this is a separate table from the normal one,
 594              * but we use the same procedure by multiplying r with scale and truncating to integer.
 595              */
 596             rt               = _mm_mul_pd(r00,gbscale);
 597             gbitab           = _mm_cvttpd_epi32(rt);
 598             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 599             gbitab           = _mm_slli_epi32(gbitab,2);
 600
 601             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 602             F                = _mm_setzero_pd();
 603             GMX_MM_TRANSPOSE2_PD(Y,F);
 604             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 605             H                = _mm_setzero_pd();
 606             GMX_MM_TRANSPOSE2_PD(G,H);
 607             Heps             = _mm_mul_pd(gbeps,H);
 608             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 609             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 610             vgb              = _mm_mul_pd(gbqqfactor,VV);
 611
 612             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 613             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 614             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 615             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 616             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 617             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 618             velec            = _mm_mul_pd(qq00,rinv00);
 619             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 620
 621             fscal            = felec;
 622
 623             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm_mul_pd(fscal,dx00);
 627             ty               = _mm_mul_pd(fscal,dy00);
 628             tz               = _mm_mul_pd(fscal,dz00);
 629
 630             /* Update vectorial force */
 631             fix0             = _mm_add_pd(fix0,tx);
 632             fiy0             = _mm_add_pd(fiy0,ty);
 633             fiz0             = _mm_add_pd(fiz0,tz);
 634
 635             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 636
 637             /* Inner loop uses 56 flops */
 638         }
 639
 640         /* End of innermost loop */
 641
 642         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 643                                               f+i_coord_offset,fshift+i_shift_offset);
 644
 645         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 646         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 647
 648         /* Increment number of inner iterations */
 649         inneriter                  += j_index_end - j_index_start;
 650
 651         /* Outer loop uses 7 flops */
 652     }
 653
 654     /* Increment number of outer iterations */
 655     outeriter        += nri;
 656
 657     /* Update outer/inner flops */
 658
 659     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*56);
 660 }