src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_ElecEw_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_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     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128i          ewitab;
  96     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_pd();
 150         fiy0             = _mm_setzero_pd();
 151         fiz0             = _mm_setzero_pd();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 155         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_pd();
 159         vvdwsum          = _mm_setzero_pd();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm_sub_pd(ix0,jx0);
 177             dy00             = _mm_sub_pd(iy0,jy0);
 178             dz00             = _mm_sub_pd(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 182
 183             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 184
 185             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 186
 187             /* Load parameters for j particles */
 188             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 189             vdwjidx0A        = 2*vdwtype[jnrA+0];
 190             vdwjidx0B        = 2*vdwtype[jnrB+0];
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             r00              = _mm_mul_pd(rsq00,rinv00);
 197
 198             /* Compute parameters for interactions between i and j atoms */
 199             qq00             = _mm_mul_pd(iq0,jq0);
 200             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 201                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 202
 203             /* EWALD ELECTROSTATICS */
 204
 205             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 206             ewrt             = _mm_mul_pd(r00,ewtabscale);
 207             ewitab           = _mm_cvttpd_epi32(ewrt);
 208 #ifdef __XOP__
 209             eweps            = _mm_frcz_pd(ewrt);
 210 #else
 211             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 212 #endif
 213             twoeweps         = _mm_add_pd(eweps,eweps);
 214             ewitab           = _mm_slli_epi32(ewitab,2);
 215             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 216             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 217             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 218             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 219             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 220             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 221             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 222             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 223             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 224             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 225
 226             /* LENNARD-JONES DISPERSION/REPULSION */
 227
 228             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 229             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 230             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 231             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 232             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 233
 234             /* Update potential sum for this i atom from the interaction with this j atom. */
 235             velecsum         = _mm_add_pd(velecsum,velec);
 236             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 237
 238             fscal            = _mm_add_pd(felec,fvdw);
 239
 240             /* Update vectorial force */
 241             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 242             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 243             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 244
 245             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 246                                                    _mm_mul_pd(dx00,fscal),
 247                                                    _mm_mul_pd(dy00,fscal),
 248                                                    _mm_mul_pd(dz00,fscal));
 249
 250             /* Inner loop uses 56 flops */
 251         }
 252
 253         if(jidx<j_index_end)
 254         {
 255
 256             jnrA             = jjnr[jidx];
 257             j_coord_offsetA  = DIM*jnrA;
 258
 259             /* load j atom coordinates */
 260             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 261                                               &jx0,&jy0,&jz0);
 262
 263             /* Calculate displacement vector */
 264             dx00             = _mm_sub_pd(ix0,jx0);
 265             dy00             = _mm_sub_pd(iy0,jy0);
 266             dz00             = _mm_sub_pd(iz0,jz0);
 267
 268             /* Calculate squared distance and things based on it */
 269             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 270
 271             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 272
 273             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 274
 275             /* Load parameters for j particles */
 276             jq0              = _mm_load_sd(charge+jnrA+0);
 277             vdwjidx0A        = 2*vdwtype[jnrA+0];
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r00              = _mm_mul_pd(rsq00,rinv00);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq00             = _mm_mul_pd(iq0,jq0);
 287             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 288
 289             /* EWALD ELECTROSTATICS */
 290
 291             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 292             ewrt             = _mm_mul_pd(r00,ewtabscale);
 293             ewitab           = _mm_cvttpd_epi32(ewrt);
 294 #ifdef __XOP__
 295             eweps            = _mm_frcz_pd(ewrt);
 296 #else
 297             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 298 #endif
 299             twoeweps         = _mm_add_pd(eweps,eweps);
 300             ewitab           = _mm_slli_epi32(ewitab,2);
 301             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 302             ewtabD           = _mm_setzero_pd();
 303             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 304             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 305             ewtabFn          = _mm_setzero_pd();
 306             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 307             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 308             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 309             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 310             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 311
 312             /* LENNARD-JONES DISPERSION/REPULSION */
 313
 314             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 315             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 316             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 317             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 318             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 322             velecsum         = _mm_add_pd(velecsum,velec);
 323             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 324             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 325
 326             fscal            = _mm_add_pd(felec,fvdw);
 327
 328             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 329
 330             /* Update vectorial force */
 331             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 332             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 333             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 334
 335             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 336                                                    _mm_mul_pd(dx00,fscal),
 337                                                    _mm_mul_pd(dy00,fscal),
 338                                                    _mm_mul_pd(dz00,fscal));
 339
 340             /* Inner loop uses 56 flops */
 341         }
 342
 343         /* End of innermost loop */
 344
 345         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 346                                               f+i_coord_offset,fshift+i_shift_offset);
 347
 348         ggid                        = gid[iidx];
 349         /* Update potential energies */
 350         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 351         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 352
 353         /* Increment number of inner iterations */
 354         inneriter                  += j_index_end - j_index_start;
 355
 356         /* Outer loop uses 9 flops */
 357     }
 358
 359     /* Increment number of outer iterations */
 360     outeriter        += nri;
 361
 362     /* Update outer/inner flops */
 363
 364     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*56);
 365 }
 366 /*
 367  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_double
 368  * Electrostatics interaction: Ewald
 369  * VdW interaction:            LennardJones
 370  * Geometry:                   Particle-Particle
 371  * Calculate force/pot:        Force
 372  */
 373 void
 374 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_double
 375                     (t_nblist                    * gmx_restrict       nlist,
 376                      rvec                        * gmx_restrict          xx,
 377                      rvec                        * gmx_restrict          ff,
 378                      t_forcerec                  * gmx_restrict          fr,
 379                      t_mdatoms                   * gmx_restrict     mdatoms,
 380                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 381                      t_nrnb                      * gmx_restrict        nrnb)
 382 {
 383     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 384      * just 0 for non-waters.
 385      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 386      * jnr indices corresponding to data put in the four positions in the SIMD register.
 387      */
 388     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 389     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 390     int              jnrA,jnrB;
 391     int              j_coord_offsetA,j_coord_offsetB;
 392     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 393     real             rcutoff_scalar;
 394     real             *shiftvec,*fshift,*x,*f;
 395     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 396     int              vdwioffset0;
 397     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 398     int              vdwjidx0A,vdwjidx0B;
 399     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 400     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 401     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 402     real             *charge;
 403     int              nvdwtype;
 404     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 405     int              *vdwtype;
 406     real             *vdwparam;
 407     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 408     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 409     __m128i          ewitab;
 410     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 411     real             *ewtab;
 412     __m128d          dummy_mask,cutoff_mask;
 413     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 414     __m128d          one     = _mm_set1_pd(1.0);
 415     __m128d          two     = _mm_set1_pd(2.0);
 416     x                = xx[0];
 417     f                = ff[0];
 418
 419     nri              = nlist->nri;
 420     iinr             = nlist->iinr;
 421     jindex           = nlist->jindex;
 422     jjnr             = nlist->jjnr;
 423     shiftidx         = nlist->shift;
 424     gid              = nlist->gid;
 425     shiftvec         = fr->shift_vec[0];
 426     fshift           = fr->fshift[0];
 427     facel            = _mm_set1_pd(fr->epsfac);
 428     charge           = mdatoms->chargeA;
 429     nvdwtype         = fr->ntype;
 430     vdwparam         = fr->nbfp;
 431     vdwtype          = mdatoms->typeA;
 432
 433     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 434     ewtab            = fr->ic->tabq_coul_F;
 435     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 436     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 437
 438     /* Avoid stupid compiler warnings */
 439     jnrA = jnrB = 0;
 440     j_coord_offsetA = 0;
 441     j_coord_offsetB = 0;
 442
 443     outeriter        = 0;
 444     inneriter        = 0;
 445
 446     /* Start outer loop over neighborlists */
 447     for(iidx=0; iidx<nri; iidx++)
 448     {
 449         /* Load shift vector for this list */
 450         i_shift_offset   = DIM*shiftidx[iidx];
 451
 452         /* Load limits for loop over neighbors */
 453         j_index_start    = jindex[iidx];
 454         j_index_end      = jindex[iidx+1];
 455
 456         /* Get outer coordinate index */
 457         inr              = iinr[iidx];
 458         i_coord_offset   = DIM*inr;
 459
 460         /* Load i particle coords and add shift vector */
 461         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 462
 463         fix0             = _mm_setzero_pd();
 464         fiy0             = _mm_setzero_pd();
 465         fiz0             = _mm_setzero_pd();
 466
 467         /* Load parameters for i particles */
 468         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 469         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 470
 471         /* Start inner kernel loop */
 472         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 473         {
 474
 475             /* Get j neighbor index, and coordinate index */
 476             jnrA             = jjnr[jidx];
 477             jnrB             = jjnr[jidx+1];
 478             j_coord_offsetA  = DIM*jnrA;
 479             j_coord_offsetB  = DIM*jnrB;
 480
 481             /* load j atom coordinates */
 482             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                               &jx0,&jy0,&jz0);
 484
 485             /* Calculate displacement vector */
 486             dx00             = _mm_sub_pd(ix0,jx0);
 487             dy00             = _mm_sub_pd(iy0,jy0);
 488             dz00             = _mm_sub_pd(iz0,jz0);
 489
 490             /* Calculate squared distance and things based on it */
 491             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 492
 493             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 494
 495             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 496
 497             /* Load parameters for j particles */
 498             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 499             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500             vdwjidx0B        = 2*vdwtype[jnrB+0];
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r00              = _mm_mul_pd(rsq00,rinv00);
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             qq00             = _mm_mul_pd(iq0,jq0);
 510             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 511                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 512
 513             /* EWALD ELECTROSTATICS */
 514
 515             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 516             ewrt             = _mm_mul_pd(r00,ewtabscale);
 517             ewitab           = _mm_cvttpd_epi32(ewrt);
 518 #ifdef __XOP__
 519             eweps            = _mm_frcz_pd(ewrt);
 520 #else
 521             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 522 #endif
 523             twoeweps         = _mm_add_pd(eweps,eweps);
 524             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 525                                          &ewtabF,&ewtabFn);
 526             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 527             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 528
 529             /* LENNARD-JONES DISPERSION/REPULSION */
 530
 531             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 532             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 533
 534             fscal            = _mm_add_pd(felec,fvdw);
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 538             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 539             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 540
 541             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 542                                                    _mm_mul_pd(dx00,fscal),
 543                                                    _mm_mul_pd(dy00,fscal),
 544                                                    _mm_mul_pd(dz00,fscal));
 545
 546             /* Inner loop uses 46 flops */
 547         }
 548
 549         if(jidx<j_index_end)
 550         {
 551
 552             jnrA             = jjnr[jidx];
 553             j_coord_offsetA  = DIM*jnrA;
 554
 555             /* load j atom coordinates */
 556             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 557                                               &jx0,&jy0,&jz0);
 558
 559             /* Calculate displacement vector */
 560             dx00             = _mm_sub_pd(ix0,jx0);
 561             dy00             = _mm_sub_pd(iy0,jy0);
 562             dz00             = _mm_sub_pd(iz0,jz0);
 563
 564             /* Calculate squared distance and things based on it */
 565             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 566
 567             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 568
 569             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 570
 571             /* Load parameters for j particles */
 572             jq0              = _mm_load_sd(charge+jnrA+0);
 573             vdwjidx0A        = 2*vdwtype[jnrA+0];
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r00              = _mm_mul_pd(rsq00,rinv00);
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             qq00             = _mm_mul_pd(iq0,jq0);
 583             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 584
 585             /* EWALD ELECTROSTATICS */
 586
 587             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 588             ewrt             = _mm_mul_pd(r00,ewtabscale);
 589             ewitab           = _mm_cvttpd_epi32(ewrt);
 590 #ifdef __XOP__
 591             eweps            = _mm_frcz_pd(ewrt);
 592 #else
 593             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 594 #endif
 595             twoeweps         = _mm_add_pd(eweps,eweps);
 596             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 597             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 598             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 599
 600             /* LENNARD-JONES DISPERSION/REPULSION */
 601
 602             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 603             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 604
 605             fscal            = _mm_add_pd(felec,fvdw);
 606
 607             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 608
 609             /* Update vectorial force */
 610             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 611             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 612             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 613
 614             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 615                                                    _mm_mul_pd(dx00,fscal),
 616                                                    _mm_mul_pd(dy00,fscal),
 617                                                    _mm_mul_pd(dz00,fscal));
 618
 619             /* Inner loop uses 46 flops */
 620         }
 621
 622         /* End of innermost loop */
 623
 624         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 625                                               f+i_coord_offset,fshift+i_shift_offset);
 626
 627         /* Increment number of inner iterations */
 628         inneriter                  += j_index_end - j_index_start;
 629
 630         /* Outer loop uses 7 flops */
 631     }
 632
 633     /* Increment number of outer iterations */
 634     outeriter        += nri;
 635
 636     /* Update outer/inner flops */
 637
 638     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*46);
 639 }