src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_double.h"
  48 #include "kernelutil_x86_sse4_1_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse4_1_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse4_1_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     __m128i          ewitab;
  88     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  89     real             *ewtab;
  90     __m128d          dummy_mask,cutoff_mask;
  91     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  92     __m128d          one     = _mm_set1_pd(1.0);
  93     __m128d          two     = _mm_set1_pd(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_pd(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107
 108     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 109     ewtab            = fr->ic->tabq_coul_FDV0;
 110     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 111     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 112
 113     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 114     rcutoff_scalar   = fr->rcoulomb;
 115     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 116     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _mm_setzero_pd();
 144         fiy0             = _mm_setzero_pd();
 145         fiz0             = _mm_setzero_pd();
 146
 147         /* Load parameters for i particles */
 148         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm_setzero_pd();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162
 163             /* load j atom coordinates */
 164             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_pd(ix0,jx0);
 169             dy00             = _mm_sub_pd(iy0,jy0);
 170             dz00             = _mm_sub_pd(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 174
 175             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 176
 177             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 178
 179             /* Load parameters for j particles */
 180             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 181
 182             /**************************
 183              * CALCULATE INTERACTIONS *
 184              **************************/
 185
 186             if (gmx_mm_any_lt(rsq00,rcutoff2))
 187             {
 188
 189             r00              = _mm_mul_pd(rsq00,rinv00);
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm_mul_pd(iq0,jq0);
 193
 194             /* EWALD ELECTROSTATICS */
 195
 196             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 197             ewrt             = _mm_mul_pd(r00,ewtabscale);
 198             ewitab           = _mm_cvttpd_epi32(ewrt);
 199             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 200             ewitab           = _mm_slli_epi32(ewitab,2);
 201             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 202             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 203             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 204             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 205             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 206             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 207             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 208             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 209             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 210             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 211
 212             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 213
 214             /* Update potential sum for this i atom from the interaction with this j atom. */
 215             velec            = _mm_and_pd(velec,cutoff_mask);
 216             velecsum         = _mm_add_pd(velecsum,velec);
 217
 218             fscal            = felec;
 219
 220             fscal            = _mm_and_pd(fscal,cutoff_mask);
 221
 222             /* Calculate temporary vectorial force */
 223             tx               = _mm_mul_pd(fscal,dx00);
 224             ty               = _mm_mul_pd(fscal,dy00);
 225             tz               = _mm_mul_pd(fscal,dz00);
 226
 227             /* Update vectorial force */
 228             fix0             = _mm_add_pd(fix0,tx);
 229             fiy0             = _mm_add_pd(fiy0,ty);
 230             fiz0             = _mm_add_pd(fiz0,tz);
 231
 232             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 233
 234             }
 235
 236             /* Inner loop uses 46 flops */
 237         }
 238
 239         if(jidx<j_index_end)
 240         {
 241
 242             jnrA             = jjnr[jidx];
 243             j_coord_offsetA  = DIM*jnrA;
 244
 245             /* load j atom coordinates */
 246             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 247                                               &jx0,&jy0,&jz0);
 248
 249             /* Calculate displacement vector */
 250             dx00             = _mm_sub_pd(ix0,jx0);
 251             dy00             = _mm_sub_pd(iy0,jy0);
 252             dz00             = _mm_sub_pd(iz0,jz0);
 253
 254             /* Calculate squared distance and things based on it */
 255             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 256
 257             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 258
 259             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 260
 261             /* Load parameters for j particles */
 262             jq0              = _mm_load_sd(charge+jnrA+0);
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             if (gmx_mm_any_lt(rsq00,rcutoff2))
 269             {
 270
 271             r00              = _mm_mul_pd(rsq00,rinv00);
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             qq00             = _mm_mul_pd(iq0,jq0);
 275
 276             /* EWALD ELECTROSTATICS */
 277
 278             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 279             ewrt             = _mm_mul_pd(r00,ewtabscale);
 280             ewitab           = _mm_cvttpd_epi32(ewrt);
 281             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 282             ewitab           = _mm_slli_epi32(ewitab,2);
 283             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 284             ewtabD           = _mm_setzero_pd();
 285             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 286             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 287             ewtabFn          = _mm_setzero_pd();
 288             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 289             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 290             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 291             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 292             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 293
 294             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velec            = _mm_and_pd(velec,cutoff_mask);
 298             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 299             velecsum         = _mm_add_pd(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303             fscal            = _mm_and_pd(fscal,cutoff_mask);
 304
 305             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm_mul_pd(fscal,dx00);
 309             ty               = _mm_mul_pd(fscal,dy00);
 310             tz               = _mm_mul_pd(fscal,dz00);
 311
 312             /* Update vectorial force */
 313             fix0             = _mm_add_pd(fix0,tx);
 314             fiy0             = _mm_add_pd(fiy0,ty);
 315             fiz0             = _mm_add_pd(fiz0,tz);
 316
 317             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 318
 319             }
 320
 321             /* Inner loop uses 46 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
 333         /* Increment number of inner iterations */
 334         inneriter                  += j_index_end - j_index_start;
 335
 336         /* Outer loop uses 8 flops */
 337     }
 338
 339     /* Increment number of outer iterations */
 340     outeriter        += nri;
 341
 342     /* Update outer/inner flops */
 343
 344     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*46);
 345 }
 346 /*
 347  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_double
 348  * Electrostatics interaction: Ewald
 349  * VdW interaction:            None
 350  * Geometry:                   Particle-Particle
 351  * Calculate force/pot:        Force
 352  */
 353 void
 354 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_double
 355                     (t_nblist                    * gmx_restrict       nlist,
 356                      rvec                        * gmx_restrict          xx,
 357                      rvec                        * gmx_restrict          ff,
 358                      t_forcerec                  * gmx_restrict          fr,
 359                      t_mdatoms                   * gmx_restrict     mdatoms,
 360                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 361                      t_nrnb                      * gmx_restrict        nrnb)
 362 {
 363     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 364      * just 0 for non-waters.
 365      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 366      * jnr indices corresponding to data put in the four positions in the SIMD register.
 367      */
 368     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 369     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 370     int              jnrA,jnrB;
 371     int              j_coord_offsetA,j_coord_offsetB;
 372     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 373     real             rcutoff_scalar;
 374     real             *shiftvec,*fshift,*x,*f;
 375     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 376     int              vdwioffset0;
 377     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 378     int              vdwjidx0A,vdwjidx0B;
 379     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 380     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 381     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 382     real             *charge;
 383     __m128i          ewitab;
 384     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 385     real             *ewtab;
 386     __m128d          dummy_mask,cutoff_mask;
 387     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 388     __m128d          one     = _mm_set1_pd(1.0);
 389     __m128d          two     = _mm_set1_pd(2.0);
 390     x                = xx[0];
 391     f                = ff[0];
 392
 393     nri              = nlist->nri;
 394     iinr             = nlist->iinr;
 395     jindex           = nlist->jindex;
 396     jjnr             = nlist->jjnr;
 397     shiftidx         = nlist->shift;
 398     gid              = nlist->gid;
 399     shiftvec         = fr->shift_vec[0];
 400     fshift           = fr->fshift[0];
 401     facel            = _mm_set1_pd(fr->epsfac);
 402     charge           = mdatoms->chargeA;
 403
 404     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 405     ewtab            = fr->ic->tabq_coul_F;
 406     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 407     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 408
 409     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 410     rcutoff_scalar   = fr->rcoulomb;
 411     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 412     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 413
 414     /* Avoid stupid compiler warnings */
 415     jnrA = jnrB = 0;
 416     j_coord_offsetA = 0;
 417     j_coord_offsetB = 0;
 418
 419     outeriter        = 0;
 420     inneriter        = 0;
 421
 422     /* Start outer loop over neighborlists */
 423     for(iidx=0; iidx<nri; iidx++)
 424     {
 425         /* Load shift vector for this list */
 426         i_shift_offset   = DIM*shiftidx[iidx];
 427
 428         /* Load limits for loop over neighbors */
 429         j_index_start    = jindex[iidx];
 430         j_index_end      = jindex[iidx+1];
 431
 432         /* Get outer coordinate index */
 433         inr              = iinr[iidx];
 434         i_coord_offset   = DIM*inr;
 435
 436         /* Load i particle coords and add shift vector */
 437         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 438
 439         fix0             = _mm_setzero_pd();
 440         fiy0             = _mm_setzero_pd();
 441         fiz0             = _mm_setzero_pd();
 442
 443         /* Load parameters for i particles */
 444         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 445
 446         /* Start inner kernel loop */
 447         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 448         {
 449
 450             /* Get j neighbor index, and coordinate index */
 451             jnrA             = jjnr[jidx];
 452             jnrB             = jjnr[jidx+1];
 453             j_coord_offsetA  = DIM*jnrA;
 454             j_coord_offsetB  = DIM*jnrB;
 455
 456             /* load j atom coordinates */
 457             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 458                                               &jx0,&jy0,&jz0);
 459
 460             /* Calculate displacement vector */
 461             dx00             = _mm_sub_pd(ix0,jx0);
 462             dy00             = _mm_sub_pd(iy0,jy0);
 463             dz00             = _mm_sub_pd(iz0,jz0);
 464
 465             /* Calculate squared distance and things based on it */
 466             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 467
 468             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 469
 470             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 471
 472             /* Load parameters for j particles */
 473             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             if (gmx_mm_any_lt(rsq00,rcutoff2))
 480             {
 481
 482             r00              = _mm_mul_pd(rsq00,rinv00);
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq00             = _mm_mul_pd(iq0,jq0);
 486
 487             /* EWALD ELECTROSTATICS */
 488
 489             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 490             ewrt             = _mm_mul_pd(r00,ewtabscale);
 491             ewitab           = _mm_cvttpd_epi32(ewrt);
 492             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 493             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 494                                          &ewtabF,&ewtabFn);
 495             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 496             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 497
 498             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 499
 500             fscal            = felec;
 501
 502             fscal            = _mm_and_pd(fscal,cutoff_mask);
 503
 504             /* Calculate temporary vectorial force */
 505             tx               = _mm_mul_pd(fscal,dx00);
 506             ty               = _mm_mul_pd(fscal,dy00);
 507             tz               = _mm_mul_pd(fscal,dz00);
 508
 509             /* Update vectorial force */
 510             fix0             = _mm_add_pd(fix0,tx);
 511             fiy0             = _mm_add_pd(fiy0,ty);
 512             fiz0             = _mm_add_pd(fiz0,tz);
 513
 514             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 515
 516             }
 517
 518             /* Inner loop uses 39 flops */
 519         }
 520
 521         if(jidx<j_index_end)
 522         {
 523
 524             jnrA             = jjnr[jidx];
 525             j_coord_offsetA  = DIM*jnrA;
 526
 527             /* load j atom coordinates */
 528             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 529                                               &jx0,&jy0,&jz0);
 530
 531             /* Calculate displacement vector */
 532             dx00             = _mm_sub_pd(ix0,jx0);
 533             dy00             = _mm_sub_pd(iy0,jy0);
 534             dz00             = _mm_sub_pd(iz0,jz0);
 535
 536             /* Calculate squared distance and things based on it */
 537             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 538
 539             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 540
 541             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 542
 543             /* Load parameters for j particles */
 544             jq0              = _mm_load_sd(charge+jnrA+0);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             if (gmx_mm_any_lt(rsq00,rcutoff2))
 551             {
 552
 553             r00              = _mm_mul_pd(rsq00,rinv00);
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq00             = _mm_mul_pd(iq0,jq0);
 557
 558             /* EWALD ELECTROSTATICS */
 559
 560             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 561             ewrt             = _mm_mul_pd(r00,ewtabscale);
 562             ewitab           = _mm_cvttpd_epi32(ewrt);
 563             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 564             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 565             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 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             /* Calculate temporary vectorial force */
 577             tx               = _mm_mul_pd(fscal,dx00);
 578             ty               = _mm_mul_pd(fscal,dy00);
 579             tz               = _mm_mul_pd(fscal,dz00);
 580
 581             /* Update vectorial force */
 582             fix0             = _mm_add_pd(fix0,tx);
 583             fiy0             = _mm_add_pd(fiy0,ty);
 584             fiz0             = _mm_add_pd(fiz0,tz);
 585
 586             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 587
 588             }
 589
 590             /* Inner loop uses 39 flops */
 591         }
 592
 593         /* End of innermost loop */
 594
 595         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 596                                               f+i_coord_offset,fshift+i_shift_offset);
 597
 598         /* Increment number of inner iterations */
 599         inneriter                  += j_index_end - j_index_start;
 600
 601         /* Outer loop uses 7 flops */
 602     }
 603
 604     /* Increment number of outer iterations */
 605     outeriter        += nri;
 606
 607     /* Update outer/inner flops */
 608
 609     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*39);
 610 }