src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_avx_128_fma_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 avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128i          ewitab;
  90     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     real             *ewtab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109
 110     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 114
 115     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 116     rcutoff_scalar   = fr->rcoulomb;
 117     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 118     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_pd();
 146         fiy0             = _mm_setzero_pd();
 147         fiz0             = _mm_setzero_pd();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_pd();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             j_coord_offsetA  = DIM*jnrA;
 163             j_coord_offsetB  = DIM*jnrB;
 164
 165             /* load j atom coordinates */
 166             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 167                                               &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx00             = _mm_sub_pd(ix0,jx0);
 171             dy00             = _mm_sub_pd(iy0,jy0);
 172             dz00             = _mm_sub_pd(iz0,jz0);
 173
 174             /* Calculate squared distance and things based on it */
 175             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 176
 177             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 178
 179             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 183
 184             /**************************
 185              * CALCULATE INTERACTIONS *
 186              **************************/
 187
 188             if (gmx_mm_any_lt(rsq00,rcutoff2))
 189             {
 190
 191             r00              = _mm_mul_pd(rsq00,rinv00);
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm_mul_pd(iq0,jq0);
 195
 196             /* EWALD ELECTROSTATICS */
 197
 198             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 199             ewrt             = _mm_mul_pd(r00,ewtabscale);
 200             ewitab           = _mm_cvttpd_epi32(ewrt);
 201 #ifdef __XOP__
 202             eweps            = _mm_frcz_pd(ewrt);
 203 #else
 204             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 205 #endif
 206             twoeweps         = _mm_add_pd(eweps,eweps);
 207             ewitab           = _mm_slli_epi32(ewitab,2);
 208             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 209             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 210             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 211             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 212             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 213             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 214             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 215             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 216             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 217             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 218
 219             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 220
 221             /* Update potential sum for this i atom from the interaction with this j atom. */
 222             velec            = _mm_and_pd(velec,cutoff_mask);
 223             velecsum         = _mm_add_pd(velecsum,velec);
 224
 225             fscal            = felec;
 226
 227             fscal            = _mm_and_pd(fscal,cutoff_mask);
 228
 229             /* Update vectorial force */
 230             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 231             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 232             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 233
 234             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 235                                                    _mm_mul_pd(dx00,fscal),
 236                                                    _mm_mul_pd(dy00,fscal),
 237                                                    _mm_mul_pd(dz00,fscal));
 238
 239             }
 240
 241             /* Inner loop uses 49 flops */
 242         }
 243
 244         if(jidx<j_index_end)
 245         {
 246
 247             jnrA             = jjnr[jidx];
 248             j_coord_offsetA  = DIM*jnrA;
 249
 250             /* load j atom coordinates */
 251             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 252                                               &jx0,&jy0,&jz0);
 253
 254             /* Calculate displacement vector */
 255             dx00             = _mm_sub_pd(ix0,jx0);
 256             dy00             = _mm_sub_pd(iy0,jy0);
 257             dz00             = _mm_sub_pd(iz0,jz0);
 258
 259             /* Calculate squared distance and things based on it */
 260             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 261
 262             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 263
 264             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 265
 266             /* Load parameters for j particles */
 267             jq0              = _mm_load_sd(charge+jnrA+0);
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             if (gmx_mm_any_lt(rsq00,rcutoff2))
 274             {
 275
 276             r00              = _mm_mul_pd(rsq00,rinv00);
 277
 278             /* Compute parameters for interactions between i and j atoms */
 279             qq00             = _mm_mul_pd(iq0,jq0);
 280
 281             /* EWALD ELECTROSTATICS */
 282
 283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 284             ewrt             = _mm_mul_pd(r00,ewtabscale);
 285             ewitab           = _mm_cvttpd_epi32(ewrt);
 286 #ifdef __XOP__
 287             eweps            = _mm_frcz_pd(ewrt);
 288 #else
 289             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 290 #endif
 291             twoeweps         = _mm_add_pd(eweps,eweps);
 292             ewitab           = _mm_slli_epi32(ewitab,2);
 293             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 294             ewtabD           = _mm_setzero_pd();
 295             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 296             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 297             ewtabFn          = _mm_setzero_pd();
 298             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 299             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 300             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 301             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 302             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 303
 304             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velec            = _mm_and_pd(velec,cutoff_mask);
 308             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 309             velecsum         = _mm_add_pd(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             fscal            = _mm_and_pd(fscal,cutoff_mask);
 314
 315             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 316
 317             /* Update vectorial force */
 318             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 319             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 320             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 321
 322             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 323                                                    _mm_mul_pd(dx00,fscal),
 324                                                    _mm_mul_pd(dy00,fscal),
 325                                                    _mm_mul_pd(dz00,fscal));
 326
 327             }
 328
 329             /* Inner loop uses 49 flops */
 330         }
 331
 332         /* End of innermost loop */
 333
 334         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 335                                               f+i_coord_offset,fshift+i_shift_offset);
 336
 337         ggid                        = gid[iidx];
 338         /* Update potential energies */
 339         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 340
 341         /* Increment number of inner iterations */
 342         inneriter                  += j_index_end - j_index_start;
 343
 344         /* Outer loop uses 8 flops */
 345     }
 346
 347     /* Increment number of outer iterations */
 348     outeriter        += nri;
 349
 350     /* Update outer/inner flops */
 351
 352     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*49);
 353 }
 354 /*
 355  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_double
 356  * Electrostatics interaction: Ewald
 357  * VdW interaction:            None
 358  * Geometry:                   Particle-Particle
 359  * Calculate force/pot:        Force
 360  */
 361 void
 362 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_double
 363                     (t_nblist                    * gmx_restrict       nlist,
 364                      rvec                        * gmx_restrict          xx,
 365                      rvec                        * gmx_restrict          ff,
 366                      t_forcerec                  * gmx_restrict          fr,
 367                      t_mdatoms                   * gmx_restrict     mdatoms,
 368                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 369                      t_nrnb                      * gmx_restrict        nrnb)
 370 {
 371     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 372      * just 0 for non-waters.
 373      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 374      * jnr indices corresponding to data put in the four positions in the SIMD register.
 375      */
 376     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 377     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 378     int              jnrA,jnrB;
 379     int              j_coord_offsetA,j_coord_offsetB;
 380     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 381     real             rcutoff_scalar;
 382     real             *shiftvec,*fshift,*x,*f;
 383     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 384     int              vdwioffset0;
 385     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 386     int              vdwjidx0A,vdwjidx0B;
 387     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 388     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 389     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 390     real             *charge;
 391     __m128i          ewitab;
 392     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 393     real             *ewtab;
 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     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 413     ewtab            = fr->ic->tabq_coul_F;
 414     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 415     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 416
 417     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 418     rcutoff_scalar   = fr->rcoulomb;
 419     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 420     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 421
 422     /* Avoid stupid compiler warnings */
 423     jnrA = jnrB = 0;
 424     j_coord_offsetA = 0;
 425     j_coord_offsetB = 0;
 426
 427     outeriter        = 0;
 428     inneriter        = 0;
 429
 430     /* Start outer loop over neighborlists */
 431     for(iidx=0; iidx<nri; iidx++)
 432     {
 433         /* Load shift vector for this list */
 434         i_shift_offset   = DIM*shiftidx[iidx];
 435
 436         /* Load limits for loop over neighbors */
 437         j_index_start    = jindex[iidx];
 438         j_index_end      = jindex[iidx+1];
 439
 440         /* Get outer coordinate index */
 441         inr              = iinr[iidx];
 442         i_coord_offset   = DIM*inr;
 443
 444         /* Load i particle coords and add shift vector */
 445         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 446
 447         fix0             = _mm_setzero_pd();
 448         fiy0             = _mm_setzero_pd();
 449         fiz0             = _mm_setzero_pd();
 450
 451         /* Load parameters for i particles */
 452         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 453
 454         /* Start inner kernel loop */
 455         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 456         {
 457
 458             /* Get j neighbor index, and coordinate index */
 459             jnrA             = jjnr[jidx];
 460             jnrB             = jjnr[jidx+1];
 461             j_coord_offsetA  = DIM*jnrA;
 462             j_coord_offsetB  = DIM*jnrB;
 463
 464             /* load j atom coordinates */
 465             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 466                                               &jx0,&jy0,&jz0);
 467
 468             /* Calculate displacement vector */
 469             dx00             = _mm_sub_pd(ix0,jx0);
 470             dy00             = _mm_sub_pd(iy0,jy0);
 471             dz00             = _mm_sub_pd(iz0,jz0);
 472
 473             /* Calculate squared distance and things based on it */
 474             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 475
 476             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 477
 478             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 479
 480             /* Load parameters for j particles */
 481             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             if (gmx_mm_any_lt(rsq00,rcutoff2))
 488             {
 489
 490             r00              = _mm_mul_pd(rsq00,rinv00);
 491
 492             /* Compute parameters for interactions between i and j atoms */
 493             qq00             = _mm_mul_pd(iq0,jq0);
 494
 495             /* EWALD ELECTROSTATICS */
 496
 497             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 498             ewrt             = _mm_mul_pd(r00,ewtabscale);
 499             ewitab           = _mm_cvttpd_epi32(ewrt);
 500 #ifdef __XOP__
 501             eweps            = _mm_frcz_pd(ewrt);
 502 #else
 503             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 504 #endif
 505             twoeweps         = _mm_add_pd(eweps,eweps);
 506             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 507                                          &ewtabF,&ewtabFn);
 508             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 509             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 510
 511             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 512
 513             fscal            = felec;
 514
 515             fscal            = _mm_and_pd(fscal,cutoff_mask);
 516
 517             /* Update vectorial force */
 518             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 519             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 520             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 521
 522             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 523                                                    _mm_mul_pd(dx00,fscal),
 524                                                    _mm_mul_pd(dy00,fscal),
 525                                                    _mm_mul_pd(dz00,fscal));
 526
 527             }
 528
 529             /* Inner loop uses 42 flops */
 530         }
 531
 532         if(jidx<j_index_end)
 533         {
 534
 535             jnrA             = jjnr[jidx];
 536             j_coord_offsetA  = DIM*jnrA;
 537
 538             /* load j atom coordinates */
 539             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 540                                               &jx0,&jy0,&jz0);
 541
 542             /* Calculate displacement vector */
 543             dx00             = _mm_sub_pd(ix0,jx0);
 544             dy00             = _mm_sub_pd(iy0,jy0);
 545             dz00             = _mm_sub_pd(iz0,jz0);
 546
 547             /* Calculate squared distance and things based on it */
 548             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 549
 550             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 551
 552             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 553
 554             /* Load parameters for j particles */
 555             jq0              = _mm_load_sd(charge+jnrA+0);
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             if (gmx_mm_any_lt(rsq00,rcutoff2))
 562             {
 563
 564             r00              = _mm_mul_pd(rsq00,rinv00);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq00             = _mm_mul_pd(iq0,jq0);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 572             ewrt             = _mm_mul_pd(r00,ewtabscale);
 573             ewitab           = _mm_cvttpd_epi32(ewrt);
 574 #ifdef __XOP__
 575             eweps            = _mm_frcz_pd(ewrt);
 576 #else
 577             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 578 #endif
 579             twoeweps         = _mm_add_pd(eweps,eweps);
 580             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 581             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 582             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 583
 584             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm_and_pd(fscal,cutoff_mask);
 589
 590             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 591
 592             /* Update vectorial force */
 593             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 594             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 595             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 596
 597             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 598                                                    _mm_mul_pd(dx00,fscal),
 599                                                    _mm_mul_pd(dy00,fscal),
 600                                                    _mm_mul_pd(dz00,fscal));
 601
 602             }
 603
 604             /* Inner loop uses 42 flops */
 605         }
 606
 607         /* End of innermost loop */
 608
 609         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 610                                               f+i_coord_offset,fshift+i_shift_offset);
 611
 612         /* Increment number of inner iterations */
 613         inneriter                  += j_index_end - j_index_start;
 614
 615         /* Outer loop uses 7 flops */
 616     }
 617
 618     /* Increment number of outer iterations */
 619     outeriter        += nri;
 620
 621     /* Update outer/inner flops */
 622
 623     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*42);
 624 }