src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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,twoeweps,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 #ifdef __XOP__
 186             eweps            = _mm_frcz_pd(ewrt);
 187 #else
 188             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 189 #endif
 190             twoeweps         = _mm_add_pd(eweps,eweps);
 191             ewitab           = _mm_slli_epi32(ewitab,2);
 192             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 193             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 194             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 195             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 196             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 197             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 198             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 199             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 200             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 201             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 202
 203             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 204
 205             /* Update potential sum for this i atom from the interaction with this j atom. */
 206             velec            = _mm_and_pd(velec,cutoff_mask);
 207             velecsum         = _mm_add_pd(velecsum,velec);
 208
 209             fscal            = felec;
 210
 211             fscal            = _mm_and_pd(fscal,cutoff_mask);
 212
 213             /* Update vectorial force */
 214             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 215             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 216             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 217
 218             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 219                                                    _mm_mul_pd(dx00,fscal),
 220                                                    _mm_mul_pd(dy00,fscal),
 221                                                    _mm_mul_pd(dz00,fscal));
 222
 223             }
 224
 225             /* Inner loop uses 49 flops */
 226         }
 227
 228         if(jidx<j_index_end)
 229         {
 230
 231             jnrA             = jjnr[jidx];
 232             j_coord_offsetA  = DIM*jnrA;
 233
 234             /* load j atom coordinates */
 235             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 236                                               &jx0,&jy0,&jz0);
 237
 238             /* Calculate displacement vector */
 239             dx00             = _mm_sub_pd(ix0,jx0);
 240             dy00             = _mm_sub_pd(iy0,jy0);
 241             dz00             = _mm_sub_pd(iz0,jz0);
 242
 243             /* Calculate squared distance and things based on it */
 244             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 245
 246             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 247
 248             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 249
 250             /* Load parameters for j particles */
 251             jq0              = _mm_load_sd(charge+jnrA+0);
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             if (gmx_mm_any_lt(rsq00,rcutoff2))
 258             {
 259
 260             r00              = _mm_mul_pd(rsq00,rinv00);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             qq00             = _mm_mul_pd(iq0,jq0);
 264
 265             /* EWALD ELECTROSTATICS */
 266
 267             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 268             ewrt             = _mm_mul_pd(r00,ewtabscale);
 269             ewitab           = _mm_cvttpd_epi32(ewrt);
 270 #ifdef __XOP__
 271             eweps            = _mm_frcz_pd(ewrt);
 272 #else
 273             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 274 #endif
 275             twoeweps         = _mm_add_pd(eweps,eweps);
 276             ewitab           = _mm_slli_epi32(ewitab,2);
 277             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 278             ewtabD           = _mm_setzero_pd();
 279             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 280             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 281             ewtabFn          = _mm_setzero_pd();
 282             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 283             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 284             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 285             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 286             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 287
 288             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velec            = _mm_and_pd(velec,cutoff_mask);
 292             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 293             velecsum         = _mm_add_pd(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297             fscal            = _mm_and_pd(fscal,cutoff_mask);
 298
 299             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 300
 301             /* Update vectorial force */
 302             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 303             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 304             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 305
 306             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 307                                                    _mm_mul_pd(dx00,fscal),
 308                                                    _mm_mul_pd(dy00,fscal),
 309                                                    _mm_mul_pd(dz00,fscal));
 310
 311             }
 312
 313             /* Inner loop uses 49 flops */
 314         }
 315
 316         /* End of innermost loop */
 317
 318         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 319                                               f+i_coord_offset,fshift+i_shift_offset);
 320
 321         ggid                        = gid[iidx];
 322         /* Update potential energies */
 323         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 324
 325         /* Increment number of inner iterations */
 326         inneriter                  += j_index_end - j_index_start;
 327
 328         /* Outer loop uses 8 flops */
 329     }
 330
 331     /* Increment number of outer iterations */
 332     outeriter        += nri;
 333
 334     /* Update outer/inner flops */
 335
 336     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*49);
 337 }
 338 /*
 339  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_double
 340  * Electrostatics interaction: Ewald
 341  * VdW interaction:            None
 342  * Geometry:                   Particle-Particle
 343  * Calculate force/pot:        Force
 344  */
 345 void
 346 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_double
 347                     (t_nblist * gmx_restrict                nlist,
 348                      rvec * gmx_restrict                    xx,
 349                      rvec * gmx_restrict                    ff,
 350                      t_forcerec * gmx_restrict              fr,
 351                      t_mdatoms * gmx_restrict               mdatoms,
 352                      nb_kernel_data_t * gmx_restrict        kernel_data,
 353                      t_nrnb * gmx_restrict                  nrnb)
 354 {
 355     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 356      * just 0 for non-waters.
 357      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 358      * jnr indices corresponding to data put in the four positions in the SIMD register.
 359      */
 360     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 361     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 362     int              jnrA,jnrB;
 363     int              j_coord_offsetA,j_coord_offsetB;
 364     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 365     real             rcutoff_scalar;
 366     real             *shiftvec,*fshift,*x,*f;
 367     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 368     int              vdwioffset0;
 369     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 370     int              vdwjidx0A,vdwjidx0B;
 371     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 372     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 373     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 374     real             *charge;
 375     __m128i          ewitab;
 376     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 377     real             *ewtab;
 378     __m128d          dummy_mask,cutoff_mask;
 379     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 380     __m128d          one     = _mm_set1_pd(1.0);
 381     __m128d          two     = _mm_set1_pd(2.0);
 382     x                = xx[0];
 383     f                = ff[0];
 384
 385     nri              = nlist->nri;
 386     iinr             = nlist->iinr;
 387     jindex           = nlist->jindex;
 388     jjnr             = nlist->jjnr;
 389     shiftidx         = nlist->shift;
 390     gid              = nlist->gid;
 391     shiftvec         = fr->shift_vec[0];
 392     fshift           = fr->fshift[0];
 393     facel            = _mm_set1_pd(fr->epsfac);
 394     charge           = mdatoms->chargeA;
 395
 396     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 397     ewtab            = fr->ic->tabq_coul_F;
 398     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 399     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 400
 401     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 402     rcutoff_scalar   = fr->rcoulomb;
 403     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 404     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 405
 406     /* Avoid stupid compiler warnings */
 407     jnrA = jnrB = 0;
 408     j_coord_offsetA = 0;
 409     j_coord_offsetB = 0;
 410
 411     outeriter        = 0;
 412     inneriter        = 0;
 413
 414     /* Start outer loop over neighborlists */
 415     for(iidx=0; iidx<nri; iidx++)
 416     {
 417         /* Load shift vector for this list */
 418         i_shift_offset   = DIM*shiftidx[iidx];
 419
 420         /* Load limits for loop over neighbors */
 421         j_index_start    = jindex[iidx];
 422         j_index_end      = jindex[iidx+1];
 423
 424         /* Get outer coordinate index */
 425         inr              = iinr[iidx];
 426         i_coord_offset   = DIM*inr;
 427
 428         /* Load i particle coords and add shift vector */
 429         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 430
 431         fix0             = _mm_setzero_pd();
 432         fiy0             = _mm_setzero_pd();
 433         fiz0             = _mm_setzero_pd();
 434
 435         /* Load parameters for i particles */
 436         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 437
 438         /* Start inner kernel loop */
 439         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 440         {
 441
 442             /* Get j neighbor index, and coordinate index */
 443             jnrA             = jjnr[jidx];
 444             jnrB             = jjnr[jidx+1];
 445             j_coord_offsetA  = DIM*jnrA;
 446             j_coord_offsetB  = DIM*jnrB;
 447
 448             /* load j atom coordinates */
 449             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 450                                               &jx0,&jy0,&jz0);
 451
 452             /* Calculate displacement vector */
 453             dx00             = _mm_sub_pd(ix0,jx0);
 454             dy00             = _mm_sub_pd(iy0,jy0);
 455             dz00             = _mm_sub_pd(iz0,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 459
 460             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 461
 462             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 463
 464             /* Load parameters for j particles */
 465             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             if (gmx_mm_any_lt(rsq00,rcutoff2))
 472             {
 473
 474             r00              = _mm_mul_pd(rsq00,rinv00);
 475
 476             /* Compute parameters for interactions between i and j atoms */
 477             qq00             = _mm_mul_pd(iq0,jq0);
 478
 479             /* EWALD ELECTROSTATICS */
 480
 481             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 482             ewrt             = _mm_mul_pd(r00,ewtabscale);
 483             ewitab           = _mm_cvttpd_epi32(ewrt);
 484 #ifdef __XOP__
 485             eweps            = _mm_frcz_pd(ewrt);
 486 #else
 487             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 488 #endif
 489             twoeweps         = _mm_add_pd(eweps,eweps);
 490             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 491                                          &ewtabF,&ewtabFn);
 492             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 493             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 494
 495             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 496
 497             fscal            = felec;
 498
 499             fscal            = _mm_and_pd(fscal,cutoff_mask);
 500
 501             /* Update vectorial force */
 502             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 503             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 504             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 505
 506             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 507                                                    _mm_mul_pd(dx00,fscal),
 508                                                    _mm_mul_pd(dy00,fscal),
 509                                                    _mm_mul_pd(dz00,fscal));
 510
 511             }
 512
 513             /* Inner loop uses 42 flops */
 514         }
 515
 516         if(jidx<j_index_end)
 517         {
 518
 519             jnrA             = jjnr[jidx];
 520             j_coord_offsetA  = DIM*jnrA;
 521
 522             /* load j atom coordinates */
 523             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 524                                               &jx0,&jy0,&jz0);
 525
 526             /* Calculate displacement vector */
 527             dx00             = _mm_sub_pd(ix0,jx0);
 528             dy00             = _mm_sub_pd(iy0,jy0);
 529             dz00             = _mm_sub_pd(iz0,jz0);
 530
 531             /* Calculate squared distance and things based on it */
 532             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 533
 534             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 535
 536             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 537
 538             /* Load parameters for j particles */
 539             jq0              = _mm_load_sd(charge+jnrA+0);
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             if (gmx_mm_any_lt(rsq00,rcutoff2))
 546             {
 547
 548             r00              = _mm_mul_pd(rsq00,rinv00);
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq00             = _mm_mul_pd(iq0,jq0);
 552
 553             /* EWALD ELECTROSTATICS */
 554
 555             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 556             ewrt             = _mm_mul_pd(r00,ewtabscale);
 557             ewitab           = _mm_cvttpd_epi32(ewrt);
 558 #ifdef __XOP__
 559             eweps            = _mm_frcz_pd(ewrt);
 560 #else
 561             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 562 #endif
 563             twoeweps         = _mm_add_pd(eweps,eweps);
 564             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 565             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 566             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 567
 568             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm_and_pd(fscal,cutoff_mask);
 573
 574             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 575
 576             /* Update vectorial force */
 577             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 578             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 579             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 580
 581             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 582                                                    _mm_mul_pd(dx00,fscal),
 583                                                    _mm_mul_pd(dy00,fscal),
 584                                                    _mm_mul_pd(dz00,fscal));
 585
 586             }
 587
 588             /* Inner loop uses 42 flops */
 589         }
 590
 591         /* End of innermost loop */
 592
 593         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 594                                               f+i_coord_offset,fshift+i_shift_offset);
 595
 596         /* Increment number of inner iterations */
 597         inneriter                  += j_index_end - j_index_start;
 598
 599         /* Outer loop uses 7 flops */
 600     }
 601
 602     /* Increment number of outer iterations */
 603     outeriter        += nri;
 604
 605     /* Update outer/inner flops */
 606
 607     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*42);
 608 }