src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecGB_VdwNone_GeomP1P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: GeneralizedBorn
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sse4_1_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     __m128i          gbitab;
  74     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
  75     __m128d          minushalf = _mm_set1_pd(-0.5);
  76     real             *invsqrta,*dvda,*gbtab;
  77     __m128i          vfitab;
  78     __m128i          ifour       = _mm_set1_epi32(4);
  79     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  80     real             *vftab;
  81     __m128d          dummy_mask,cutoff_mask;
  82     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  83     __m128d          one     = _mm_set1_pd(1.0);
  84     __m128d          two     = _mm_set1_pd(2.0);
  85     x                = xx[0];
  86     f                = ff[0];
  87
  88     nri              = nlist->nri;
  89     iinr             = nlist->iinr;
  90     jindex           = nlist->jindex;
  91     jjnr             = nlist->jjnr;
  92     shiftidx         = nlist->shift;
  93     gid              = nlist->gid;
  94     shiftvec         = fr->shift_vec[0];
  95     fshift           = fr->fshift[0];
  96     facel            = _mm_set1_pd(fr->epsfac);
  97     charge           = mdatoms->chargeA;
  98
  99     invsqrta         = fr->invsqrta;
 100     dvda             = fr->dvda;
 101     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 102     gbtab            = fr->gbtab.data;
 103     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 104
 105     /* Avoid stupid compiler warnings */
 106     jnrA = jnrB = 0;
 107     j_coord_offsetA = 0;
 108     j_coord_offsetB = 0;
 109
 110     outeriter        = 0;
 111     inneriter        = 0;
 112
 113     /* Start outer loop over neighborlists */
 114     for(iidx=0; iidx<nri; iidx++)
 115     {
 116         /* Load shift vector for this list */
 117         i_shift_offset   = DIM*shiftidx[iidx];
 118
 119         /* Load limits for loop over neighbors */
 120         j_index_start    = jindex[iidx];
 121         j_index_end      = jindex[iidx+1];
 122
 123         /* Get outer coordinate index */
 124         inr              = iinr[iidx];
 125         i_coord_offset   = DIM*inr;
 126
 127         /* Load i particle coords and add shift vector */
 128         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 129
 130         fix0             = _mm_setzero_pd();
 131         fiy0             = _mm_setzero_pd();
 132         fiz0             = _mm_setzero_pd();
 133
 134         /* Load parameters for i particles */
 135         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 136         isai0            = _mm_load1_pd(invsqrta+inr+0);
 137
 138         /* Reset potential sums */
 139         velecsum         = _mm_setzero_pd();
 140         vgbsum           = _mm_setzero_pd();
 141         dvdasum          = _mm_setzero_pd();
 142
 143         /* Start inner kernel loop */
 144         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 145         {
 146
 147             /* Get j neighbor index, and coordinate index */
 148             jnrA             = jjnr[jidx];
 149             jnrB             = jjnr[jidx+1];
 150             j_coord_offsetA  = DIM*jnrA;
 151             j_coord_offsetB  = DIM*jnrB;
 152
 153             /* load j atom coordinates */
 154             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 155                                               &jx0,&jy0,&jz0);
 156
 157             /* Calculate displacement vector */
 158             dx00             = _mm_sub_pd(ix0,jx0);
 159             dy00             = _mm_sub_pd(iy0,jy0);
 160             dz00             = _mm_sub_pd(iz0,jz0);
 161
 162             /* Calculate squared distance and things based on it */
 163             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 164
 165             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 166
 167             /* Load parameters for j particles */
 168             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 169             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 170
 171             /**************************
 172              * CALCULATE INTERACTIONS *
 173              **************************/
 174
 175             r00              = _mm_mul_pd(rsq00,rinv00);
 176
 177             /* Compute parameters for interactions between i and j atoms */
 178             qq00             = _mm_mul_pd(iq0,jq0);
 179
 180             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 181             isaprod          = _mm_mul_pd(isai0,isaj0);
 182             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 183             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 184
 185             /* Calculate generalized born table index - this is a separate table from the normal one,
 186              * but we use the same procedure by multiplying r with scale and truncating to integer.
 187              */
 188             rt               = _mm_mul_pd(r00,gbscale);
 189             gbitab           = _mm_cvttpd_epi32(rt);
 190             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 191             gbitab           = _mm_slli_epi32(gbitab,2);
 192
 193             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 194             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 195             GMX_MM_TRANSPOSE2_PD(Y,F);
 196             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 197             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 198             GMX_MM_TRANSPOSE2_PD(G,H);
 199             Heps             = _mm_mul_pd(gbeps,H);
 200             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 201             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 202             vgb              = _mm_mul_pd(gbqqfactor,VV);
 203
 204             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 205             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 206             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 207             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 208             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 209             velec            = _mm_mul_pd(qq00,rinv00);
 210             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 211
 212             /* Update potential sum for this i atom from the interaction with this j atom. */
 213             velecsum         = _mm_add_pd(velecsum,velec);
 214             vgbsum           = _mm_add_pd(vgbsum,vgb);
 215
 216             fscal            = felec;
 217
 218             /* Calculate temporary vectorial force */
 219             tx               = _mm_mul_pd(fscal,dx00);
 220             ty               = _mm_mul_pd(fscal,dy00);
 221             tz               = _mm_mul_pd(fscal,dz00);
 222
 223             /* Update vectorial force */
 224             fix0             = _mm_add_pd(fix0,tx);
 225             fiy0             = _mm_add_pd(fiy0,ty);
 226             fiz0             = _mm_add_pd(fiz0,tz);
 227
 228             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 229
 230             /* Inner loop uses 58 flops */
 231         }
 232
 233         if(jidx<j_index_end)
 234         {
 235
 236             jnrA             = jjnr[jidx];
 237             j_coord_offsetA  = DIM*jnrA;
 238
 239             /* load j atom coordinates */
 240             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 241                                               &jx0,&jy0,&jz0);
 242
 243             /* Calculate displacement vector */
 244             dx00             = _mm_sub_pd(ix0,jx0);
 245             dy00             = _mm_sub_pd(iy0,jy0);
 246             dz00             = _mm_sub_pd(iz0,jz0);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 250
 251             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 252
 253             /* Load parameters for j particles */
 254             jq0              = _mm_load_sd(charge+jnrA+0);
 255             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             r00              = _mm_mul_pd(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             qq00             = _mm_mul_pd(iq0,jq0);
 265
 266             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 267             isaprod          = _mm_mul_pd(isai0,isaj0);
 268             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 269             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 270
 271             /* Calculate generalized born table index - this is a separate table from the normal one,
 272              * but we use the same procedure by multiplying r with scale and truncating to integer.
 273              */
 274             rt               = _mm_mul_pd(r00,gbscale);
 275             gbitab           = _mm_cvttpd_epi32(rt);
 276             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 277             gbitab           = _mm_slli_epi32(gbitab,2);
 278
 279             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 280             F                = _mm_setzero_pd();
 281             GMX_MM_TRANSPOSE2_PD(Y,F);
 282             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 283             H                = _mm_setzero_pd();
 284             GMX_MM_TRANSPOSE2_PD(G,H);
 285             Heps             = _mm_mul_pd(gbeps,H);
 286             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 287             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 288             vgb              = _mm_mul_pd(gbqqfactor,VV);
 289
 290             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 291             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 292             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 293             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 294             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 295             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 296             velec            = _mm_mul_pd(qq00,rinv00);
 297             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 301             velecsum         = _mm_add_pd(velecsum,velec);
 302             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 303             vgbsum           = _mm_add_pd(vgbsum,vgb);
 304
 305             fscal            = felec;
 306
 307             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm_mul_pd(fscal,dx00);
 311             ty               = _mm_mul_pd(fscal,dy00);
 312             tz               = _mm_mul_pd(fscal,dz00);
 313
 314             /* Update vectorial force */
 315             fix0             = _mm_add_pd(fix0,tx);
 316             fiy0             = _mm_add_pd(fiy0,ty);
 317             fiz0             = _mm_add_pd(fiz0,tz);
 318
 319             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 320
 321             /* Inner loop uses 58 flops */
 322         }
 323
 324         /* End of innermost loop */
 325
 326         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 327                                               f+i_coord_offset,fshift+i_shift_offset);
 328
 329         ggid                        = gid[iidx];
 330         /* Update potential energies */
 331         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 332         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 333         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 334         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 335
 336         /* Increment number of inner iterations */
 337         inneriter                  += j_index_end - j_index_start;
 338
 339         /* Outer loop uses 9 flops */
 340     }
 341
 342     /* Increment number of outer iterations */
 343     outeriter        += nri;
 344
 345     /* Update outer/inner flops */
 346
 347     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*58);
 348 }
 349 /*
 350  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
 351  * Electrostatics interaction: GeneralizedBorn
 352  * VdW interaction:            None
 353  * Geometry:                   Particle-Particle
 354  * Calculate force/pot:        Force
 355  */
 356 void
 357 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sse4_1_double
 358                     (t_nblist * gmx_restrict                nlist,
 359                      rvec * gmx_restrict                    xx,
 360                      rvec * gmx_restrict                    ff,
 361                      t_forcerec * gmx_restrict              fr,
 362                      t_mdatoms * gmx_restrict               mdatoms,
 363                      nb_kernel_data_t * gmx_restrict        kernel_data,
 364                      t_nrnb * gmx_restrict                  nrnb)
 365 {
 366     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 367      * just 0 for non-waters.
 368      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 369      * jnr indices corresponding to data put in the four positions in the SIMD register.
 370      */
 371     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 372     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 373     int              jnrA,jnrB;
 374     int              j_coord_offsetA,j_coord_offsetB;
 375     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 376     real             rcutoff_scalar;
 377     real             *shiftvec,*fshift,*x,*f;
 378     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 379     int              vdwioffset0;
 380     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 381     int              vdwjidx0A,vdwjidx0B;
 382     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 383     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 384     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 385     real             *charge;
 386     __m128i          gbitab;
 387     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 388     __m128d          minushalf = _mm_set1_pd(-0.5);
 389     real             *invsqrta,*dvda,*gbtab;
 390     __m128i          vfitab;
 391     __m128i          ifour       = _mm_set1_epi32(4);
 392     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 393     real             *vftab;
 394     __m128d          dummy_mask,cutoff_mask;
 395     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 396     __m128d          one     = _mm_set1_pd(1.0);
 397     __m128d          two     = _mm_set1_pd(2.0);
 398     x                = xx[0];
 399     f                = ff[0];
 400
 401     nri              = nlist->nri;
 402     iinr             = nlist->iinr;
 403     jindex           = nlist->jindex;
 404     jjnr             = nlist->jjnr;
 405     shiftidx         = nlist->shift;
 406     gid              = nlist->gid;
 407     shiftvec         = fr->shift_vec[0];
 408     fshift           = fr->fshift[0];
 409     facel            = _mm_set1_pd(fr->epsfac);
 410     charge           = mdatoms->chargeA;
 411
 412     invsqrta         = fr->invsqrta;
 413     dvda             = fr->dvda;
 414     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 415     gbtab            = fr->gbtab.data;
 416     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 417
 418     /* Avoid stupid compiler warnings */
 419     jnrA = jnrB = 0;
 420     j_coord_offsetA = 0;
 421     j_coord_offsetB = 0;
 422
 423     outeriter        = 0;
 424     inneriter        = 0;
 425
 426     /* Start outer loop over neighborlists */
 427     for(iidx=0; iidx<nri; iidx++)
 428     {
 429         /* Load shift vector for this list */
 430         i_shift_offset   = DIM*shiftidx[iidx];
 431
 432         /* Load limits for loop over neighbors */
 433         j_index_start    = jindex[iidx];
 434         j_index_end      = jindex[iidx+1];
 435
 436         /* Get outer coordinate index */
 437         inr              = iinr[iidx];
 438         i_coord_offset   = DIM*inr;
 439
 440         /* Load i particle coords and add shift vector */
 441         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 442
 443         fix0             = _mm_setzero_pd();
 444         fiy0             = _mm_setzero_pd();
 445         fiz0             = _mm_setzero_pd();
 446
 447         /* Load parameters for i particles */
 448         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 449         isai0            = _mm_load1_pd(invsqrta+inr+0);
 450
 451         dvdasum          = _mm_setzero_pd();
 452
 453         /* Start inner kernel loop */
 454         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 455         {
 456
 457             /* Get j neighbor index, and coordinate index */
 458             jnrA             = jjnr[jidx];
 459             jnrB             = jjnr[jidx+1];
 460             j_coord_offsetA  = DIM*jnrA;
 461             j_coord_offsetB  = DIM*jnrB;
 462
 463             /* load j atom coordinates */
 464             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 465                                               &jx0,&jy0,&jz0);
 466
 467             /* Calculate displacement vector */
 468             dx00             = _mm_sub_pd(ix0,jx0);
 469             dy00             = _mm_sub_pd(iy0,jy0);
 470             dz00             = _mm_sub_pd(iz0,jz0);
 471
 472             /* Calculate squared distance and things based on it */
 473             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 474
 475             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 476
 477             /* Load parameters for j particles */
 478             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 479             isaj0            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0);
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             r00              = _mm_mul_pd(rsq00,rinv00);
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq00             = _mm_mul_pd(iq0,jq0);
 489
 490             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 491             isaprod          = _mm_mul_pd(isai0,isaj0);
 492             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 493             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 494
 495             /* Calculate generalized born table index - this is a separate table from the normal one,
 496              * but we use the same procedure by multiplying r with scale and truncating to integer.
 497              */
 498             rt               = _mm_mul_pd(r00,gbscale);
 499             gbitab           = _mm_cvttpd_epi32(rt);
 500             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 501             gbitab           = _mm_slli_epi32(gbitab,2);
 502
 503             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 504             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 505             GMX_MM_TRANSPOSE2_PD(Y,F);
 506             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 507             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 508             GMX_MM_TRANSPOSE2_PD(G,H);
 509             Heps             = _mm_mul_pd(gbeps,H);
 510             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 511             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 512             vgb              = _mm_mul_pd(gbqqfactor,VV);
 513
 514             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 515             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 516             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 517             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 518             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 519             velec            = _mm_mul_pd(qq00,rinv00);
 520             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 521
 522             fscal            = felec;
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm_mul_pd(fscal,dx00);
 526             ty               = _mm_mul_pd(fscal,dy00);
 527             tz               = _mm_mul_pd(fscal,dz00);
 528
 529             /* Update vectorial force */
 530             fix0             = _mm_add_pd(fix0,tx);
 531             fiy0             = _mm_add_pd(fiy0,ty);
 532             fiz0             = _mm_add_pd(fiz0,tz);
 533
 534             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 535
 536             /* Inner loop uses 56 flops */
 537         }
 538
 539         if(jidx<j_index_end)
 540         {
 541
 542             jnrA             = jjnr[jidx];
 543             j_coord_offsetA  = DIM*jnrA;
 544
 545             /* load j atom coordinates */
 546             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 547                                               &jx0,&jy0,&jz0);
 548
 549             /* Calculate displacement vector */
 550             dx00             = _mm_sub_pd(ix0,jx0);
 551             dy00             = _mm_sub_pd(iy0,jy0);
 552             dz00             = _mm_sub_pd(iz0,jz0);
 553
 554             /* Calculate squared distance and things based on it */
 555             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 556
 557             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 558
 559             /* Load parameters for j particles */
 560             jq0              = _mm_load_sd(charge+jnrA+0);
 561             isaj0            = _mm_load_sd(invsqrta+jnrA+0);
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             r00              = _mm_mul_pd(rsq00,rinv00);
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq00             = _mm_mul_pd(iq0,jq0);
 571
 572             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 573             isaprod          = _mm_mul_pd(isai0,isaj0);
 574             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq00,_mm_mul_pd(isaprod,gbinvepsdiff)));
 575             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 576
 577             /* Calculate generalized born table index - this is a separate table from the normal one,
 578              * but we use the same procedure by multiplying r with scale and truncating to integer.
 579              */
 580             rt               = _mm_mul_pd(r00,gbscale);
 581             gbitab           = _mm_cvttpd_epi32(rt);
 582             gbeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 583             gbitab           = _mm_slli_epi32(gbitab,2);
 584
 585             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 586             F                = _mm_setzero_pd();
 587             GMX_MM_TRANSPOSE2_PD(Y,F);
 588             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 589             H                = _mm_setzero_pd();
 590             GMX_MM_TRANSPOSE2_PD(G,H);
 591             Heps             = _mm_mul_pd(gbeps,H);
 592             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 593             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 594             vgb              = _mm_mul_pd(gbqqfactor,VV);
 595
 596             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 597             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 598             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r00)));
 599             dvdatmp          = _mm_unpacklo_pd(dvdatmp,_mm_setzero_pd());
 600             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 601             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj0,isaj0)));
 602             velec            = _mm_mul_pd(qq00,rinv00);
 603             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv00),fgb),rinv00);
 604
 605             fscal            = felec;
 606
 607             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 608
 609             /* Calculate temporary vectorial force */
 610             tx               = _mm_mul_pd(fscal,dx00);
 611             ty               = _mm_mul_pd(fscal,dy00);
 612             tz               = _mm_mul_pd(fscal,dz00);
 613
 614             /* Update vectorial force */
 615             fix0             = _mm_add_pd(fix0,tx);
 616             fiy0             = _mm_add_pd(fiy0,ty);
 617             fiz0             = _mm_add_pd(fiz0,tz);
 618
 619             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 620
 621             /* Inner loop uses 56 flops */
 622         }
 623
 624         /* End of innermost loop */
 625
 626         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 627                                               f+i_coord_offset,fshift+i_shift_offset);
 628
 629         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai0,isai0));
 630         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
 631
 632         /* Increment number of inner iterations */
 633         inneriter                  += j_index_end - j_index_start;
 634
 635         /* Outer loop uses 7 flops */
 636     }
 637
 638     /* Increment number of outer iterations */
 639     outeriter        += nri;
 640
 641     /* Update outer/inner flops */
 642
 643     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*56);
 644 }