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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128d           c6grid_00;
  94     real             *vdwgridparam;
  95     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  96     __m128d           one_half  = _mm_set1_pd(0.5);
  97     __m128d           minus_one = _mm_set1_pd(-1.0);
  98     __m128i          ewitab;
  99     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     __m128d          dummy_mask,cutoff_mask;
 102     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 103     __m128d          one     = _mm_set1_pd(1.0);
 104     __m128d          two     = _mm_set1_pd(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_pd(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121     vdwgridparam     = fr->ljpme_c6grid;
 122     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 123     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 124     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 125
 126     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _mm_setzero_pd();
 157         fiy0             = _mm_setzero_pd();
 158         fiz0             = _mm_setzero_pd();
 159
 160         /* Load parameters for i particles */
 161         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 162         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_pd();
 166         vvdwsum          = _mm_setzero_pd();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_pd(ix0,jx0);
 184             dy00             = _mm_sub_pd(iy0,jy0);
 185             dz00             = _mm_sub_pd(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 189
 190             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 191
 192             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 196             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198
 199             /**************************
 200              * CALCULATE INTERACTIONS *
 201              **************************/
 202
 203             r00              = _mm_mul_pd(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             qq00             = _mm_mul_pd(iq0,jq0);
 207             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 208                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 209             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 210                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 211
 212             /* EWALD ELECTROSTATICS */
 213
 214             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 215             ewrt             = _mm_mul_pd(r00,ewtabscale);
 216             ewitab           = _mm_cvttpd_epi32(ewrt);
 217 #ifdef __XOP__
 218             eweps            = _mm_frcz_pd(ewrt);
 219 #else
 220             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 221 #endif
 222             twoeweps         = _mm_add_pd(eweps,eweps);
 223             ewitab           = _mm_slli_epi32(ewitab,2);
 224             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 225             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 226             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 227             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 228             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 229             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 230             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 231             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 232             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 233             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 234
 235             /* Analytical LJ-PME */
 236             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 237             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 238             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 239             exponent         = gmx_simd_exp_d(ewcljrsq);
 240             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 241             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 242             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 243             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 244             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 245             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 246             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 247             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);
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             velecsum         = _mm_add_pd(velecsum,velec);
 251             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 252
 253             fscal            = _mm_add_pd(felec,fvdw);
 254
 255             /* Update vectorial force */
 256             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 257             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 258             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 259
 260             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 261                                                    _mm_mul_pd(dx00,fscal),
 262                                                    _mm_mul_pd(dy00,fscal),
 263                                                    _mm_mul_pd(dz00,fscal));
 264
 265             /* Inner loop uses 68 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             jnrA             = jjnr[jidx];
 272             j_coord_offsetA  = DIM*jnrA;
 273
 274             /* load j atom coordinates */
 275             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _mm_sub_pd(ix0,jx0);
 280             dy00             = _mm_sub_pd(iy0,jy0);
 281             dz00             = _mm_sub_pd(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 285
 286             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 287
 288             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 289
 290             /* Load parameters for j particles */
 291             jq0              = _mm_load_sd(charge+jnrA+0);
 292             vdwjidx0A        = 2*vdwtype[jnrA+0];
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             r00              = _mm_mul_pd(rsq00,rinv00);
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq00             = _mm_mul_pd(iq0,jq0);
 302             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 303             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 304
 305             /* EWALD ELECTROSTATICS */
 306
 307             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 308             ewrt             = _mm_mul_pd(r00,ewtabscale);
 309             ewitab           = _mm_cvttpd_epi32(ewrt);
 310 #ifdef __XOP__
 311             eweps            = _mm_frcz_pd(ewrt);
 312 #else
 313             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 314 #endif
 315             twoeweps         = _mm_add_pd(eweps,eweps);
 316             ewitab           = _mm_slli_epi32(ewitab,2);
 317             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 318             ewtabD           = _mm_setzero_pd();
 319             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 320             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 321             ewtabFn          = _mm_setzero_pd();
 322             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 323             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 324             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 325             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 326             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 327
 328             /* Analytical LJ-PME */
 329             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 330             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 331             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 332             exponent         = gmx_simd_exp_d(ewcljrsq);
 333             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 334             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 335             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 336             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 337             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 338             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 339             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 340             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);
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 344             velecsum         = _mm_add_pd(velecsum,velec);
 345             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 346             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 347
 348             fscal            = _mm_add_pd(felec,fvdw);
 349
 350             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 351
 352             /* Update vectorial force */
 353             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 354             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 355             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 356
 357             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 358                                                    _mm_mul_pd(dx00,fscal),
 359                                                    _mm_mul_pd(dy00,fscal),
 360                                                    _mm_mul_pd(dz00,fscal));
 361
 362             /* Inner loop uses 68 flops */
 363         }
 364
 365         /* End of innermost loop */
 366
 367         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 368                                               f+i_coord_offset,fshift+i_shift_offset);
 369
 370         ggid                        = gid[iidx];
 371         /* Update potential energies */
 372         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 373         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 374
 375         /* Increment number of inner iterations */
 376         inneriter                  += j_index_end - j_index_start;
 377
 378         /* Outer loop uses 9 flops */
 379     }
 380
 381     /* Increment number of outer iterations */
 382     outeriter        += nri;
 383
 384     /* Update outer/inner flops */
 385
 386     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 387 }
 388 /*
 389  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 390  * Electrostatics interaction: Ewald
 391  * VdW interaction:            LJEwald
 392  * Geometry:                   Particle-Particle
 393  * Calculate force/pot:        Force
 394  */
 395 void
 396 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_128_fma_double
 397                     (t_nblist                    * gmx_restrict       nlist,
 398                      rvec                        * gmx_restrict          xx,
 399                      rvec                        * gmx_restrict          ff,
 400                      t_forcerec                  * gmx_restrict          fr,
 401                      t_mdatoms                   * gmx_restrict     mdatoms,
 402                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 403                      t_nrnb                      * gmx_restrict        nrnb)
 404 {
 405     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 406      * just 0 for non-waters.
 407      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 408      * jnr indices corresponding to data put in the four positions in the SIMD register.
 409      */
 410     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 411     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 412     int              jnrA,jnrB;
 413     int              j_coord_offsetA,j_coord_offsetB;
 414     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 415     real             rcutoff_scalar;
 416     real             *shiftvec,*fshift,*x,*f;
 417     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 418     int              vdwioffset0;
 419     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 420     int              vdwjidx0A,vdwjidx0B;
 421     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 422     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 423     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 424     real             *charge;
 425     int              nvdwtype;
 426     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 427     int              *vdwtype;
 428     real             *vdwparam;
 429     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 430     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 431     __m128d           c6grid_00;
 432     real             *vdwgridparam;
 433     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 434     __m128d           one_half  = _mm_set1_pd(0.5);
 435     __m128d           minus_one = _mm_set1_pd(-1.0);
 436     __m128i          ewitab;
 437     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 438     real             *ewtab;
 439     __m128d          dummy_mask,cutoff_mask;
 440     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 441     __m128d          one     = _mm_set1_pd(1.0);
 442     __m128d          two     = _mm_set1_pd(2.0);
 443     x                = xx[0];
 444     f                = ff[0];
 445
 446     nri              = nlist->nri;
 447     iinr             = nlist->iinr;
 448     jindex           = nlist->jindex;
 449     jjnr             = nlist->jjnr;
 450     shiftidx         = nlist->shift;
 451     gid              = nlist->gid;
 452     shiftvec         = fr->shift_vec[0];
 453     fshift           = fr->fshift[0];
 454     facel            = _mm_set1_pd(fr->epsfac);
 455     charge           = mdatoms->chargeA;
 456     nvdwtype         = fr->ntype;
 457     vdwparam         = fr->nbfp;
 458     vdwtype          = mdatoms->typeA;
 459     vdwgridparam     = fr->ljpme_c6grid;
 460     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 461     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 462     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 463
 464     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 465     ewtab            = fr->ic->tabq_coul_F;
 466     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 467     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 468
 469     /* Avoid stupid compiler warnings */
 470     jnrA = jnrB = 0;
 471     j_coord_offsetA = 0;
 472     j_coord_offsetB = 0;
 473
 474     outeriter        = 0;
 475     inneriter        = 0;
 476
 477     /* Start outer loop over neighborlists */
 478     for(iidx=0; iidx<nri; iidx++)
 479     {
 480         /* Load shift vector for this list */
 481         i_shift_offset   = DIM*shiftidx[iidx];
 482
 483         /* Load limits for loop over neighbors */
 484         j_index_start    = jindex[iidx];
 485         j_index_end      = jindex[iidx+1];
 486
 487         /* Get outer coordinate index */
 488         inr              = iinr[iidx];
 489         i_coord_offset   = DIM*inr;
 490
 491         /* Load i particle coords and add shift vector */
 492         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 493
 494         fix0             = _mm_setzero_pd();
 495         fiy0             = _mm_setzero_pd();
 496         fiz0             = _mm_setzero_pd();
 497
 498         /* Load parameters for i particles */
 499         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 500         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 501
 502         /* Start inner kernel loop */
 503         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 504         {
 505
 506             /* Get j neighbor index, and coordinate index */
 507             jnrA             = jjnr[jidx];
 508             jnrB             = jjnr[jidx+1];
 509             j_coord_offsetA  = DIM*jnrA;
 510             j_coord_offsetB  = DIM*jnrB;
 511
 512             /* load j atom coordinates */
 513             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 514                                               &jx0,&jy0,&jz0);
 515
 516             /* Calculate displacement vector */
 517             dx00             = _mm_sub_pd(ix0,jx0);
 518             dy00             = _mm_sub_pd(iy0,jy0);
 519             dz00             = _mm_sub_pd(iz0,jz0);
 520
 521             /* Calculate squared distance and things based on it */
 522             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 523
 524             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 525
 526             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 527
 528             /* Load parameters for j particles */
 529             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 530             vdwjidx0A        = 2*vdwtype[jnrA+0];
 531             vdwjidx0B        = 2*vdwtype[jnrB+0];
 532
 533             /**************************
 534              * CALCULATE INTERACTIONS *
 535              **************************/
 536
 537             r00              = _mm_mul_pd(rsq00,rinv00);
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq00             = _mm_mul_pd(iq0,jq0);
 541             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 542                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 543             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 544                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 545
 546             /* EWALD ELECTROSTATICS */
 547
 548             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 549             ewrt             = _mm_mul_pd(r00,ewtabscale);
 550             ewitab           = _mm_cvttpd_epi32(ewrt);
 551 #ifdef __XOP__
 552             eweps            = _mm_frcz_pd(ewrt);
 553 #else
 554             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 555 #endif
 556             twoeweps         = _mm_add_pd(eweps,eweps);
 557             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 558                                          &ewtabF,&ewtabFn);
 559             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 560             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 561
 562             /* Analytical LJ-PME */
 563             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 564             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 565             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 566             exponent         = gmx_simd_exp_d(ewcljrsq);
 567             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 568             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 569             /* f6A = 6 * C6grid * (1 - poly) */
 570             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 571             /* f6B = C6grid * exponent * beta^6 */
 572             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 573             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 574             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 575
 576             fscal            = _mm_add_pd(felec,fvdw);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 580             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 581             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 582
 583             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 584                                                    _mm_mul_pd(dx00,fscal),
 585                                                    _mm_mul_pd(dy00,fscal),
 586                                                    _mm_mul_pd(dz00,fscal));
 587
 588             /* Inner loop uses 60 flops */
 589         }
 590
 591         if(jidx<j_index_end)
 592         {
 593
 594             jnrA             = jjnr[jidx];
 595             j_coord_offsetA  = DIM*jnrA;
 596
 597             /* load j atom coordinates */
 598             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 599                                               &jx0,&jy0,&jz0);
 600
 601             /* Calculate displacement vector */
 602             dx00             = _mm_sub_pd(ix0,jx0);
 603             dy00             = _mm_sub_pd(iy0,jy0);
 604             dz00             = _mm_sub_pd(iz0,jz0);
 605
 606             /* Calculate squared distance and things based on it */
 607             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 608
 609             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 610
 611             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 612
 613             /* Load parameters for j particles */
 614             jq0              = _mm_load_sd(charge+jnrA+0);
 615             vdwjidx0A        = 2*vdwtype[jnrA+0];
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             r00              = _mm_mul_pd(rsq00,rinv00);
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq00             = _mm_mul_pd(iq0,jq0);
 625             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 626             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 627
 628             /* EWALD ELECTROSTATICS */
 629
 630             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 631             ewrt             = _mm_mul_pd(r00,ewtabscale);
 632             ewitab           = _mm_cvttpd_epi32(ewrt);
 633 #ifdef __XOP__
 634             eweps            = _mm_frcz_pd(ewrt);
 635 #else
 636             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 637 #endif
 638             twoeweps         = _mm_add_pd(eweps,eweps);
 639             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 640             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 641             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 642
 643             /* Analytical LJ-PME */
 644             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 645             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 646             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 647             exponent         = gmx_simd_exp_d(ewcljrsq);
 648             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 649             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 650             /* f6A = 6 * C6grid * (1 - poly) */
 651             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 652             /* f6B = C6grid * exponent * beta^6 */
 653             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 654             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 655             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 656
 657             fscal            = _mm_add_pd(felec,fvdw);
 658
 659             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 660
 661             /* Update vectorial force */
 662             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 663             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 664             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 665
 666             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 667                                                    _mm_mul_pd(dx00,fscal),
 668                                                    _mm_mul_pd(dy00,fscal),
 669                                                    _mm_mul_pd(dz00,fscal));
 670
 671             /* Inner loop uses 60 flops */
 672         }
 673
 674         /* End of innermost loop */
 675
 676         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 677                                               f+i_coord_offset,fshift+i_shift_offset);
 678
 679         /* Increment number of inner iterations */
 680         inneriter                  += j_index_end - j_index_start;
 681
 682         /* Outer loop uses 7 flops */
 683     }
 684
 685     /* Increment number of outer iterations */
 686     outeriter        += nri;
 687
 688     /* Update outer/inner flops */
 689
 690     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 691 }