src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_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_ElecEwSh_VdwNone_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_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          ewitab;
  74     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  75     real             *ewtab;
  76     __m128d          dummy_mask,cutoff_mask;
  77     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  78     __m128d          one     = _mm_set1_pd(1.0);
  79     __m128d          two     = _mm_set1_pd(2.0);
  80     x                = xx[0];
  81     f                = ff[0];
  82
  83     nri              = nlist->nri;
  84     iinr             = nlist->iinr;
  85     jindex           = nlist->jindex;
  86     jjnr             = nlist->jjnr;
  87     shiftidx         = nlist->shift;
  88     gid              = nlist->gid;
  89     shiftvec         = fr->shift_vec[0];
  90     fshift           = fr->fshift[0];
  91     facel            = _mm_set1_pd(fr->epsfac);
  92     charge           = mdatoms->chargeA;
  93
  94     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
  95     ewtab            = fr->ic->tabq_coul_FDV0;
  96     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
  97     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
  98
  99     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 100     rcutoff_scalar   = fr->rcoulomb;
 101     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 102     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 103
 104     /* Avoid stupid compiler warnings */
 105     jnrA = jnrB = 0;
 106     j_coord_offsetA = 0;
 107     j_coord_offsetB = 0;
 108
 109     outeriter        = 0;
 110     inneriter        = 0;
 111
 112     /* Start outer loop over neighborlists */
 113     for(iidx=0; iidx<nri; iidx++)
 114     {
 115         /* Load shift vector for this list */
 116         i_shift_offset   = DIM*shiftidx[iidx];
 117
 118         /* Load limits for loop over neighbors */
 119         j_index_start    = jindex[iidx];
 120         j_index_end      = jindex[iidx+1];
 121
 122         /* Get outer coordinate index */
 123         inr              = iinr[iidx];
 124         i_coord_offset   = DIM*inr;
 125
 126         /* Load i particle coords and add shift vector */
 127         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 128
 129         fix0             = _mm_setzero_pd();
 130         fiy0             = _mm_setzero_pd();
 131         fiz0             = _mm_setzero_pd();
 132
 133         /* Load parameters for i particles */
 134         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 135
 136         /* Reset potential sums */
 137         velecsum         = _mm_setzero_pd();
 138
 139         /* Start inner kernel loop */
 140         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 141         {
 142
 143             /* Get j neighbor index, and coordinate index */
 144             jnrA             = jjnr[jidx];
 145             jnrB             = jjnr[jidx+1];
 146             j_coord_offsetA  = DIM*jnrA;
 147             j_coord_offsetB  = DIM*jnrB;
 148
 149             /* load j atom coordinates */
 150             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 151                                               &jx0,&jy0,&jz0);
 152
 153             /* Calculate displacement vector */
 154             dx00             = _mm_sub_pd(ix0,jx0);
 155             dy00             = _mm_sub_pd(iy0,jy0);
 156             dz00             = _mm_sub_pd(iz0,jz0);
 157
 158             /* Calculate squared distance and things based on it */
 159             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 160
 161             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 162
 163             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 164
 165             /* Load parameters for j particles */
 166             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 167
 168             /**************************
 169              * CALCULATE INTERACTIONS *
 170              **************************/
 171
 172             if (gmx_mm_any_lt(rsq00,rcutoff2))
 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             /* EWALD ELECTROSTATICS */
 181
 182             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 183             ewrt             = _mm_mul_pd(r00,ewtabscale);
 184             ewitab           = _mm_cvttpd_epi32(ewrt);
 185             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 186             ewitab           = _mm_slli_epi32(ewitab,2);
 187             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 188             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 189             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 190             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 191             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 192             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 193             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 194             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 195             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 196             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 197
 198             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 199
 200             /* Update potential sum for this i atom from the interaction with this j atom. */
 201             velec            = _mm_and_pd(velec,cutoff_mask);
 202             velecsum         = _mm_add_pd(velecsum,velec);
 203
 204             fscal            = felec;
 205
 206             fscal            = _mm_and_pd(fscal,cutoff_mask);
 207
 208             /* Calculate temporary vectorial force */
 209             tx               = _mm_mul_pd(fscal,dx00);
 210             ty               = _mm_mul_pd(fscal,dy00);
 211             tz               = _mm_mul_pd(fscal,dz00);
 212
 213             /* Update vectorial force */
 214             fix0             = _mm_add_pd(fix0,tx);
 215             fiy0             = _mm_add_pd(fiy0,ty);
 216             fiz0             = _mm_add_pd(fiz0,tz);
 217
 218             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 219
 220             }
 221
 222             /* Inner loop uses 46 flops */
 223         }
 224
 225         if(jidx<j_index_end)
 226         {
 227
 228             jnrA             = jjnr[jidx];
 229             j_coord_offsetA  = DIM*jnrA;
 230
 231             /* load j atom coordinates */
 232             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 233                                               &jx0,&jy0,&jz0);
 234
 235             /* Calculate displacement vector */
 236             dx00             = _mm_sub_pd(ix0,jx0);
 237             dy00             = _mm_sub_pd(iy0,jy0);
 238             dz00             = _mm_sub_pd(iz0,jz0);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 242
 243             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 244
 245             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 246
 247             /* Load parameters for j particles */
 248             jq0              = _mm_load_sd(charge+jnrA+0);
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             if (gmx_mm_any_lt(rsq00,rcutoff2))
 255             {
 256
 257             r00              = _mm_mul_pd(rsq00,rinv00);
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq00             = _mm_mul_pd(iq0,jq0);
 261
 262             /* EWALD ELECTROSTATICS */
 263
 264             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 265             ewrt             = _mm_mul_pd(r00,ewtabscale);
 266             ewitab           = _mm_cvttpd_epi32(ewrt);
 267             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 268             ewitab           = _mm_slli_epi32(ewitab,2);
 269             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 270             ewtabD           = _mm_setzero_pd();
 271             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 272             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 273             ewtabFn          = _mm_setzero_pd();
 274             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 275             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 276             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 277             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 278             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 279
 280             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velec            = _mm_and_pd(velec,cutoff_mask);
 284             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 285             velecsum         = _mm_add_pd(velecsum,velec);
 286
 287             fscal            = felec;
 288
 289             fscal            = _mm_and_pd(fscal,cutoff_mask);
 290
 291             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 292
 293             /* Calculate temporary vectorial force */
 294             tx               = _mm_mul_pd(fscal,dx00);
 295             ty               = _mm_mul_pd(fscal,dy00);
 296             tz               = _mm_mul_pd(fscal,dz00);
 297
 298             /* Update vectorial force */
 299             fix0             = _mm_add_pd(fix0,tx);
 300             fiy0             = _mm_add_pd(fiy0,ty);
 301             fiz0             = _mm_add_pd(fiz0,tz);
 302
 303             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 304
 305             }
 306
 307             /* Inner loop uses 46 flops */
 308         }
 309
 310         /* End of innermost loop */
 311
 312         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 313                                               f+i_coord_offset,fshift+i_shift_offset);
 314
 315         ggid                        = gid[iidx];
 316         /* Update potential energies */
 317         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 318
 319         /* Increment number of inner iterations */
 320         inneriter                  += j_index_end - j_index_start;
 321
 322         /* Outer loop uses 8 flops */
 323     }
 324
 325     /* Increment number of outer iterations */
 326     outeriter        += nri;
 327
 328     /* Update outer/inner flops */
 329
 330     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*46);
 331 }
 332 /*
 333  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_double
 334  * Electrostatics interaction: Ewald
 335  * VdW interaction:            None
 336  * Geometry:                   Particle-Particle
 337  * Calculate force/pot:        Force
 338  */
 339 void
 340 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_double
 341                     (t_nblist * gmx_restrict                nlist,
 342                      rvec * gmx_restrict                    xx,
 343                      rvec * gmx_restrict                    ff,
 344                      t_forcerec * gmx_restrict              fr,
 345                      t_mdatoms * gmx_restrict               mdatoms,
 346                      nb_kernel_data_t * gmx_restrict        kernel_data,
 347                      t_nrnb * gmx_restrict                  nrnb)
 348 {
 349     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 350      * just 0 for non-waters.
 351      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 352      * jnr indices corresponding to data put in the four positions in the SIMD register.
 353      */
 354     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 355     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 356     int              jnrA,jnrB;
 357     int              j_coord_offsetA,j_coord_offsetB;
 358     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 359     real             rcutoff_scalar;
 360     real             *shiftvec,*fshift,*x,*f;
 361     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 362     int              vdwioffset0;
 363     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 364     int              vdwjidx0A,vdwjidx0B;
 365     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 366     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 367     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 368     real             *charge;
 369     __m128i          ewitab;
 370     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 371     real             *ewtab;
 372     __m128d          dummy_mask,cutoff_mask;
 373     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 374     __m128d          one     = _mm_set1_pd(1.0);
 375     __m128d          two     = _mm_set1_pd(2.0);
 376     x                = xx[0];
 377     f                = ff[0];
 378
 379     nri              = nlist->nri;
 380     iinr             = nlist->iinr;
 381     jindex           = nlist->jindex;
 382     jjnr             = nlist->jjnr;
 383     shiftidx         = nlist->shift;
 384     gid              = nlist->gid;
 385     shiftvec         = fr->shift_vec[0];
 386     fshift           = fr->fshift[0];
 387     facel            = _mm_set1_pd(fr->epsfac);
 388     charge           = mdatoms->chargeA;
 389
 390     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 391     ewtab            = fr->ic->tabq_coul_F;
 392     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 393     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 394
 395     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 396     rcutoff_scalar   = fr->rcoulomb;
 397     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 398     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 399
 400     /* Avoid stupid compiler warnings */
 401     jnrA = jnrB = 0;
 402     j_coord_offsetA = 0;
 403     j_coord_offsetB = 0;
 404
 405     outeriter        = 0;
 406     inneriter        = 0;
 407
 408     /* Start outer loop over neighborlists */
 409     for(iidx=0; iidx<nri; iidx++)
 410     {
 411         /* Load shift vector for this list */
 412         i_shift_offset   = DIM*shiftidx[iidx];
 413
 414         /* Load limits for loop over neighbors */
 415         j_index_start    = jindex[iidx];
 416         j_index_end      = jindex[iidx+1];
 417
 418         /* Get outer coordinate index */
 419         inr              = iinr[iidx];
 420         i_coord_offset   = DIM*inr;
 421
 422         /* Load i particle coords and add shift vector */
 423         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 424
 425         fix0             = _mm_setzero_pd();
 426         fiy0             = _mm_setzero_pd();
 427         fiz0             = _mm_setzero_pd();
 428
 429         /* Load parameters for i particles */
 430         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 431
 432         /* Start inner kernel loop */
 433         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 434         {
 435
 436             /* Get j neighbor index, and coordinate index */
 437             jnrA             = jjnr[jidx];
 438             jnrB             = jjnr[jidx+1];
 439             j_coord_offsetA  = DIM*jnrA;
 440             j_coord_offsetB  = DIM*jnrB;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                               &jx0,&jy0,&jz0);
 445
 446             /* Calculate displacement vector */
 447             dx00             = _mm_sub_pd(ix0,jx0);
 448             dy00             = _mm_sub_pd(iy0,jy0);
 449             dz00             = _mm_sub_pd(iz0,jz0);
 450
 451             /* Calculate squared distance and things based on it */
 452             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 453
 454             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 455
 456             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 457
 458             /* Load parameters for j particles */
 459             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             if (gmx_mm_any_lt(rsq00,rcutoff2))
 466             {
 467
 468             r00              = _mm_mul_pd(rsq00,rinv00);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq00             = _mm_mul_pd(iq0,jq0);
 472
 473             /* EWALD ELECTROSTATICS */
 474
 475             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 476             ewrt             = _mm_mul_pd(r00,ewtabscale);
 477             ewitab           = _mm_cvttpd_epi32(ewrt);
 478             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 479             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 480                                          &ewtabF,&ewtabFn);
 481             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 482             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 483
 484             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 485
 486             fscal            = felec;
 487
 488             fscal            = _mm_and_pd(fscal,cutoff_mask);
 489
 490             /* Calculate temporary vectorial force */
 491             tx               = _mm_mul_pd(fscal,dx00);
 492             ty               = _mm_mul_pd(fscal,dy00);
 493             tz               = _mm_mul_pd(fscal,dz00);
 494
 495             /* Update vectorial force */
 496             fix0             = _mm_add_pd(fix0,tx);
 497             fiy0             = _mm_add_pd(fiy0,ty);
 498             fiz0             = _mm_add_pd(fiz0,tz);
 499
 500             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 501
 502             }
 503
 504             /* Inner loop uses 39 flops */
 505         }
 506
 507         if(jidx<j_index_end)
 508         {
 509
 510             jnrA             = jjnr[jidx];
 511             j_coord_offsetA  = DIM*jnrA;
 512
 513             /* load j atom coordinates */
 514             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 515                                               &jx0,&jy0,&jz0);
 516
 517             /* Calculate displacement vector */
 518             dx00             = _mm_sub_pd(ix0,jx0);
 519             dy00             = _mm_sub_pd(iy0,jy0);
 520             dz00             = _mm_sub_pd(iz0,jz0);
 521
 522             /* Calculate squared distance and things based on it */
 523             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 524
 525             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 526
 527             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 528
 529             /* Load parameters for j particles */
 530             jq0              = _mm_load_sd(charge+jnrA+0);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm_any_lt(rsq00,rcutoff2))
 537             {
 538
 539             r00              = _mm_mul_pd(rsq00,rinv00);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq00             = _mm_mul_pd(iq0,jq0);
 543
 544             /* EWALD ELECTROSTATICS */
 545
 546             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 547             ewrt             = _mm_mul_pd(r00,ewtabscale);
 548             ewitab           = _mm_cvttpd_epi32(ewrt);
 549             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 550             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 551             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 552             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 553
 554             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm_and_pd(fscal,cutoff_mask);
 559
 560             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 561
 562             /* Calculate temporary vectorial force */
 563             tx               = _mm_mul_pd(fscal,dx00);
 564             ty               = _mm_mul_pd(fscal,dy00);
 565             tz               = _mm_mul_pd(fscal,dz00);
 566
 567             /* Update vectorial force */
 568             fix0             = _mm_add_pd(fix0,tx);
 569             fiy0             = _mm_add_pd(fiy0,ty);
 570             fiz0             = _mm_add_pd(fiz0,tz);
 571
 572             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 573
 574             }
 575
 576             /* Inner loop uses 39 flops */
 577         }
 578
 579         /* End of innermost loop */
 580
 581         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 582                                               f+i_coord_offset,fshift+i_shift_offset);
 583
 584         /* Increment number of inner iterations */
 585         inneriter                  += j_index_end - j_index_start;
 586
 587         /* Outer loop uses 7 flops */
 588     }
 589
 590     /* Increment number of outer iterations */
 591     outeriter        += nri;
 592
 593     /* Update outer/inner flops */
 594
 595     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 596 }