src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJEw_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 #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_VdwLJEw_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_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     __m128d           c6grid_00;
  96     real             *vdwgridparam;
  97     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  98     __m128d           one_half  = _mm_set1_pd(0.5);
  99     __m128d           minus_one = _mm_set1_pd(-1.0);
 100     __m128i          ewitab;
 101     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     real             *ewtab;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123     vdwgridparam     = fr->ljpme_c6grid;
 124     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 125     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 126     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 127
 128     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 157
 158         fix0             = _mm_setzero_pd();
 159         fiy0             = _mm_setzero_pd();
 160         fiz0             = _mm_setzero_pd();
 161
 162         /* Load parameters for i particles */
 163         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 164         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_pd();
 168         vvdwsum          = _mm_setzero_pd();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179
 180             /* load j atom coordinates */
 181             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm_sub_pd(ix0,jx0);
 186             dy00             = _mm_sub_pd(iy0,jy0);
 187             dz00             = _mm_sub_pd(iz0,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 191
 192             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 193
 194             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 195
 196             /* Load parameters for j particles */
 197             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 198             vdwjidx0A        = 2*vdwtype[jnrA+0];
 199             vdwjidx0B        = 2*vdwtype[jnrB+0];
 200
 201             /**************************
 202              * CALCULATE INTERACTIONS *
 203              **************************/
 204
 205             r00              = _mm_mul_pd(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _mm_mul_pd(iq0,jq0);
 209             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 210                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 211             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 212                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 213
 214             /* EWALD ELECTROSTATICS */
 215
 216             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 217             ewrt             = _mm_mul_pd(r00,ewtabscale);
 218             ewitab           = _mm_cvttpd_epi32(ewrt);
 219 #ifdef __XOP__
 220             eweps            = _mm_frcz_pd(ewrt);
 221 #else
 222             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 223 #endif
 224             twoeweps         = _mm_add_pd(eweps,eweps);
 225             ewitab           = _mm_slli_epi32(ewitab,2);
 226             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 227             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 228             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 229             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 230             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 231             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 232             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 233             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 234             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 235             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 236
 237             /* Analytical LJ-PME */
 238             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 239             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 240             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 241             exponent         = gmx_simd_exp_d(ewcljrsq);
 242             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 243             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 244             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 245             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 246             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 247             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 248             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 249             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 250
 251             /* Update potential sum for this i atom from the interaction with this j atom. */
 252             velecsum         = _mm_add_pd(velecsum,velec);
 253             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 254
 255             fscal            = _mm_add_pd(felec,fvdw);
 256
 257             /* Update vectorial force */
 258             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 259             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 260             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 261
 262             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 263                                                    _mm_mul_pd(dx00,fscal),
 264                                                    _mm_mul_pd(dy00,fscal),
 265                                                    _mm_mul_pd(dz00,fscal));
 266
 267             /* Inner loop uses 68 flops */
 268         }
 269
 270         if(jidx<j_index_end)
 271         {
 272
 273             jnrA             = jjnr[jidx];
 274             j_coord_offsetA  = DIM*jnrA;
 275
 276             /* load j atom coordinates */
 277             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 278                                               &jx0,&jy0,&jz0);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm_sub_pd(ix0,jx0);
 282             dy00             = _mm_sub_pd(iy0,jy0);
 283             dz00             = _mm_sub_pd(iz0,jz0);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 287
 288             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 289
 290             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 291
 292             /* Load parameters for j particles */
 293             jq0              = _mm_load_sd(charge+jnrA+0);
 294             vdwjidx0A        = 2*vdwtype[jnrA+0];
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             r00              = _mm_mul_pd(rsq00,rinv00);
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq00             = _mm_mul_pd(iq0,jq0);
 304             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 305             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 310             ewrt             = _mm_mul_pd(r00,ewtabscale);
 311             ewitab           = _mm_cvttpd_epi32(ewrt);
 312 #ifdef __XOP__
 313             eweps            = _mm_frcz_pd(ewrt);
 314 #else
 315             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 316 #endif
 317             twoeweps         = _mm_add_pd(eweps,eweps);
 318             ewitab           = _mm_slli_epi32(ewitab,2);
 319             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 320             ewtabD           = _mm_setzero_pd();
 321             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 322             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 323             ewtabFn          = _mm_setzero_pd();
 324             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 325             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 326             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 327             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 328             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 329
 330             /* Analytical LJ-PME */
 331             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 332             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 333             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 334             exponent         = gmx_simd_exp_d(ewcljrsq);
 335             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 336             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 337             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 338             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 339             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 340             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 341             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 342             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 346             velecsum         = _mm_add_pd(velecsum,velec);
 347             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 348             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 349
 350             fscal            = _mm_add_pd(felec,fvdw);
 351
 352             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 353
 354             /* Update vectorial force */
 355             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 356             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 357             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 358
 359             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 360                                                    _mm_mul_pd(dx00,fscal),
 361                                                    _mm_mul_pd(dy00,fscal),
 362                                                    _mm_mul_pd(dz00,fscal));
 363
 364             /* Inner loop uses 68 flops */
 365         }
 366
 367         /* End of innermost loop */
 368
 369         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 370                                               f+i_coord_offset,fshift+i_shift_offset);
 371
 372         ggid                        = gid[iidx];
 373         /* Update potential energies */
 374         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 375         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 376
 377         /* Increment number of inner iterations */
 378         inneriter                  += j_index_end - j_index_start;
 379
 380         /* Outer loop uses 9 flops */
 381     }
 382
 383     /* Increment number of outer iterations */
 384     outeriter        += nri;
 385
 386     /* Update outer/inner flops */
 387
 388     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 389 }
 390 /*
 391  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 392  * Electrostatics interaction: Ewald
 393  * VdW interaction:            LJEwald
 394  * Geometry:                   Particle-Particle
 395  * Calculate force/pot:        Force
 396  */
 397 void
 398 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 399                     (t_nblist                    * gmx_restrict       nlist,
 400                      rvec                        * gmx_restrict          xx,
 401                      rvec                        * gmx_restrict          ff,
 402                      t_forcerec                  * gmx_restrict          fr,
 403                      t_mdatoms                   * gmx_restrict     mdatoms,
 404                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 405                      t_nrnb                      * gmx_restrict        nrnb)
 406 {
 407     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 408      * just 0 for non-waters.
 409      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 410      * jnr indices corresponding to data put in the four positions in the SIMD register.
 411      */
 412     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 413     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 414     int              jnrA,jnrB;
 415     int              j_coord_offsetA,j_coord_offsetB;
 416     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 417     real             rcutoff_scalar;
 418     real             *shiftvec,*fshift,*x,*f;
 419     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 420     int              vdwioffset0;
 421     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 422     int              vdwjidx0A,vdwjidx0B;
 423     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 424     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 425     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 426     real             *charge;
 427     int              nvdwtype;
 428     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 429     int              *vdwtype;
 430     real             *vdwparam;
 431     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 432     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 433     __m128d           c6grid_00;
 434     real             *vdwgridparam;
 435     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 436     __m128d           one_half  = _mm_set1_pd(0.5);
 437     __m128d           minus_one = _mm_set1_pd(-1.0);
 438     __m128i          ewitab;
 439     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 440     real             *ewtab;
 441     __m128d          dummy_mask,cutoff_mask;
 442     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 443     __m128d          one     = _mm_set1_pd(1.0);
 444     __m128d          two     = _mm_set1_pd(2.0);
 445     x                = xx[0];
 446     f                = ff[0];
 447
 448     nri              = nlist->nri;
 449     iinr             = nlist->iinr;
 450     jindex           = nlist->jindex;
 451     jjnr             = nlist->jjnr;
 452     shiftidx         = nlist->shift;
 453     gid              = nlist->gid;
 454     shiftvec         = fr->shift_vec[0];
 455     fshift           = fr->fshift[0];
 456     facel            = _mm_set1_pd(fr->epsfac);
 457     charge           = mdatoms->chargeA;
 458     nvdwtype         = fr->ntype;
 459     vdwparam         = fr->nbfp;
 460     vdwtype          = mdatoms->typeA;
 461     vdwgridparam     = fr->ljpme_c6grid;
 462     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 463     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 464     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 465
 466     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 467     ewtab            = fr->ic->tabq_coul_F;
 468     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 469     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 470
 471     /* Avoid stupid compiler warnings */
 472     jnrA = jnrB = 0;
 473     j_coord_offsetA = 0;
 474     j_coord_offsetB = 0;
 475
 476     outeriter        = 0;
 477     inneriter        = 0;
 478
 479     /* Start outer loop over neighborlists */
 480     for(iidx=0; iidx<nri; iidx++)
 481     {
 482         /* Load shift vector for this list */
 483         i_shift_offset   = DIM*shiftidx[iidx];
 484
 485         /* Load limits for loop over neighbors */
 486         j_index_start    = jindex[iidx];
 487         j_index_end      = jindex[iidx+1];
 488
 489         /* Get outer coordinate index */
 490         inr              = iinr[iidx];
 491         i_coord_offset   = DIM*inr;
 492
 493         /* Load i particle coords and add shift vector */
 494         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 495
 496         fix0             = _mm_setzero_pd();
 497         fiy0             = _mm_setzero_pd();
 498         fiz0             = _mm_setzero_pd();
 499
 500         /* Load parameters for i particles */
 501         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 502         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 503
 504         /* Start inner kernel loop */
 505         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 506         {
 507
 508             /* Get j neighbor index, and coordinate index */
 509             jnrA             = jjnr[jidx];
 510             jnrB             = jjnr[jidx+1];
 511             j_coord_offsetA  = DIM*jnrA;
 512             j_coord_offsetB  = DIM*jnrB;
 513
 514             /* load j atom coordinates */
 515             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 516                                               &jx0,&jy0,&jz0);
 517
 518             /* Calculate displacement vector */
 519             dx00             = _mm_sub_pd(ix0,jx0);
 520             dy00             = _mm_sub_pd(iy0,jy0);
 521             dz00             = _mm_sub_pd(iz0,jz0);
 522
 523             /* Calculate squared distance and things based on it */
 524             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 525
 526             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 527
 528             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 529
 530             /* Load parameters for j particles */
 531             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 532             vdwjidx0A        = 2*vdwtype[jnrA+0];
 533             vdwjidx0B        = 2*vdwtype[jnrB+0];
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             r00              = _mm_mul_pd(rsq00,rinv00);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq00             = _mm_mul_pd(iq0,jq0);
 543             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 544                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 545             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 546                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 547
 548             /* EWALD ELECTROSTATICS */
 549
 550             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 551             ewrt             = _mm_mul_pd(r00,ewtabscale);
 552             ewitab           = _mm_cvttpd_epi32(ewrt);
 553 #ifdef __XOP__
 554             eweps            = _mm_frcz_pd(ewrt);
 555 #else
 556             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 557 #endif
 558             twoeweps         = _mm_add_pd(eweps,eweps);
 559             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 560                                          &ewtabF,&ewtabFn);
 561             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 562             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 563
 564             /* Analytical LJ-PME */
 565             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 566             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 567             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 568             exponent         = gmx_simd_exp_d(ewcljrsq);
 569             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 570             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 571             /* f6A = 6 * C6grid * (1 - poly) */
 572             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 573             /* f6B = C6grid * exponent * beta^6 */
 574             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 575             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 576             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 577
 578             fscal            = _mm_add_pd(felec,fvdw);
 579
 580             /* Update vectorial force */
 581             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 582             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 583             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 584
 585             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 586                                                    _mm_mul_pd(dx00,fscal),
 587                                                    _mm_mul_pd(dy00,fscal),
 588                                                    _mm_mul_pd(dz00,fscal));
 589
 590             /* Inner loop uses 60 flops */
 591         }
 592
 593         if(jidx<j_index_end)
 594         {
 595
 596             jnrA             = jjnr[jidx];
 597             j_coord_offsetA  = DIM*jnrA;
 598
 599             /* load j atom coordinates */
 600             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 601                                               &jx0,&jy0,&jz0);
 602
 603             /* Calculate displacement vector */
 604             dx00             = _mm_sub_pd(ix0,jx0);
 605             dy00             = _mm_sub_pd(iy0,jy0);
 606             dz00             = _mm_sub_pd(iz0,jz0);
 607
 608             /* Calculate squared distance and things based on it */
 609             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 610
 611             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 612
 613             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 614
 615             /* Load parameters for j particles */
 616             jq0              = _mm_load_sd(charge+jnrA+0);
 617             vdwjidx0A        = 2*vdwtype[jnrA+0];
 618
 619             /**************************
 620              * CALCULATE INTERACTIONS *
 621              **************************/
 622
 623             r00              = _mm_mul_pd(rsq00,rinv00);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq00             = _mm_mul_pd(iq0,jq0);
 627             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 628             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 629
 630             /* EWALD ELECTROSTATICS */
 631
 632             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 633             ewrt             = _mm_mul_pd(r00,ewtabscale);
 634             ewitab           = _mm_cvttpd_epi32(ewrt);
 635 #ifdef __XOP__
 636             eweps            = _mm_frcz_pd(ewrt);
 637 #else
 638             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 639 #endif
 640             twoeweps         = _mm_add_pd(eweps,eweps);
 641             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 642             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 643             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 644
 645             /* Analytical LJ-PME */
 646             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 647             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 648             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 649             exponent         = gmx_simd_exp_d(ewcljrsq);
 650             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 651             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 652             /* f6A = 6 * C6grid * (1 - poly) */
 653             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 654             /* f6B = C6grid * exponent * beta^6 */
 655             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 656             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 657             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 658
 659             fscal            = _mm_add_pd(felec,fvdw);
 660
 661             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 662
 663             /* Update vectorial force */
 664             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 665             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 666             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 667
 668             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 669                                                    _mm_mul_pd(dx00,fscal),
 670                                                    _mm_mul_pd(dy00,fscal),
 671                                                    _mm_mul_pd(dz00,fscal));
 672
 673             /* Inner loop uses 60 flops */
 674         }
 675
 676         /* End of innermost loop */
 677
 678         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 679                                               f+i_coord_offset,fshift+i_shift_offset);
 680
 681         /* Increment number of inner iterations */
 682         inneriter                  += j_index_end - j_index_start;
 683
 684         /* Outer loop uses 7 flops */
 685     }
 686
 687     /* Increment number of outer iterations */
 688     outeriter        += nri;
 689
 690     /* Update outer/inner flops */
 691
 692     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 693 }