src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJSh_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_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_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     __m128i          ewitab;
  94     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     real             *ewtab;
  96     __m128d          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_pd(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 118     ewtab            = fr->ic->tabq_coul_FDV0;
 119     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 120     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 126
 127     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 128     rvdw             = _mm_set1_pd(fr->rvdw);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _mm_setzero_pd();
 156         fiy0             = _mm_setzero_pd();
 157         fiz0             = _mm_setzero_pd();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165         vvdwsum          = _mm_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_pd(ix0,jx0);
 183             dy00             = _mm_sub_pd(iy0,jy0);
 184             dz00             = _mm_sub_pd(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 188
 189             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 190
 191             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 195             vdwjidx0A        = 2*vdwtype[jnrA+0];
 196             vdwjidx0B        = 2*vdwtype[jnrB+0];
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_mm_any_lt(rsq00,rcutoff2))
 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
 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(_mm_sub_pd(rinv00,sh_ewald),velec));
 233             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 234
 235             /* LENNARD-JONES DISPERSION/REPULSION */
 236
 237             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 238             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 239             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 240             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 241                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 242             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 243
 244             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velec            = _mm_and_pd(velec,cutoff_mask);
 248             velecsum         = _mm_add_pd(velecsum,velec);
 249             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 250             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 251
 252             fscal            = _mm_add_pd(felec,fvdw);
 253
 254             fscal            = _mm_and_pd(fscal,cutoff_mask);
 255
 256             /* Update vectorial force */
 257             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 258             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 259             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 260
 261             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 262                                                    _mm_mul_pd(dx00,fscal),
 263                                                    _mm_mul_pd(dy00,fscal),
 264                                                    _mm_mul_pd(dz00,fscal));
 265
 266             }
 267
 268             /* Inner loop uses 67 flops */
 269         }
 270
 271         if(jidx<j_index_end)
 272         {
 273
 274             jnrA             = jjnr[jidx];
 275             j_coord_offsetA  = DIM*jnrA;
 276
 277             /* load j atom coordinates */
 278             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 279                                               &jx0,&jy0,&jz0);
 280
 281             /* Calculate displacement vector */
 282             dx00             = _mm_sub_pd(ix0,jx0);
 283             dy00             = _mm_sub_pd(iy0,jy0);
 284             dz00             = _mm_sub_pd(iz0,jz0);
 285
 286             /* Calculate squared distance and things based on it */
 287             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 288
 289             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 290
 291             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 292
 293             /* Load parameters for j particles */
 294             jq0              = _mm_load_sd(charge+jnrA+0);
 295             vdwjidx0A        = 2*vdwtype[jnrA+0];
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             if (gmx_mm_any_lt(rsq00,rcutoff2))
 302             {
 303
 304             r00              = _mm_mul_pd(rsq00,rinv00);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq00             = _mm_mul_pd(iq0,jq0);
 308             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 309
 310             /* EWALD ELECTROSTATICS */
 311
 312             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 313             ewrt             = _mm_mul_pd(r00,ewtabscale);
 314             ewitab           = _mm_cvttpd_epi32(ewrt);
 315 #ifdef __XOP__
 316             eweps            = _mm_frcz_pd(ewrt);
 317 #else
 318             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 319 #endif
 320             twoeweps         = _mm_add_pd(eweps,eweps);
 321             ewitab           = _mm_slli_epi32(ewitab,2);
 322             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 323             ewtabD           = _mm_setzero_pd();
 324             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 325             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 326             ewtabFn          = _mm_setzero_pd();
 327             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 328             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 329             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 330             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 331             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 332
 333             /* LENNARD-JONES DISPERSION/REPULSION */
 334
 335             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 336             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 337             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 338             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 339                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 340             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 341
 342             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velec            = _mm_and_pd(velec,cutoff_mask);
 346             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 347             velecsum         = _mm_add_pd(velecsum,velec);
 348             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 349             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 350             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 351
 352             fscal            = _mm_add_pd(felec,fvdw);
 353
 354             fscal            = _mm_and_pd(fscal,cutoff_mask);
 355
 356             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 357
 358             /* Update vectorial force */
 359             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 360             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 361             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 362
 363             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 364                                                    _mm_mul_pd(dx00,fscal),
 365                                                    _mm_mul_pd(dy00,fscal),
 366                                                    _mm_mul_pd(dz00,fscal));
 367
 368             }
 369
 370             /* Inner loop uses 67 flops */
 371         }
 372
 373         /* End of innermost loop */
 374
 375         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 376                                               f+i_coord_offset,fshift+i_shift_offset);
 377
 378         ggid                        = gid[iidx];
 379         /* Update potential energies */
 380         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 381         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 382
 383         /* Increment number of inner iterations */
 384         inneriter                  += j_index_end - j_index_start;
 385
 386         /* Outer loop uses 9 flops */
 387     }
 388
 389     /* Increment number of outer iterations */
 390     outeriter        += nri;
 391
 392     /* Update outer/inner flops */
 393
 394     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*67);
 395 }
 396 /*
 397  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 398  * Electrostatics interaction: Ewald
 399  * VdW interaction:            LennardJones
 400  * Geometry:                   Particle-Particle
 401  * Calculate force/pot:        Force
 402  */
 403 void
 404 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 405                     (t_nblist                    * gmx_restrict       nlist,
 406                      rvec                        * gmx_restrict          xx,
 407                      rvec                        * gmx_restrict          ff,
 408                      t_forcerec                  * gmx_restrict          fr,
 409                      t_mdatoms                   * gmx_restrict     mdatoms,
 410                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 411                      t_nrnb                      * gmx_restrict        nrnb)
 412 {
 413     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 414      * just 0 for non-waters.
 415      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 416      * jnr indices corresponding to data put in the four positions in the SIMD register.
 417      */
 418     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 419     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 420     int              jnrA,jnrB;
 421     int              j_coord_offsetA,j_coord_offsetB;
 422     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 423     real             rcutoff_scalar;
 424     real             *shiftvec,*fshift,*x,*f;
 425     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 426     int              vdwioffset0;
 427     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 428     int              vdwjidx0A,vdwjidx0B;
 429     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 430     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 431     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 432     real             *charge;
 433     int              nvdwtype;
 434     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 435     int              *vdwtype;
 436     real             *vdwparam;
 437     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 438     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 439     __m128i          ewitab;
 440     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 441     real             *ewtab;
 442     __m128d          dummy_mask,cutoff_mask;
 443     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 444     __m128d          one     = _mm_set1_pd(1.0);
 445     __m128d          two     = _mm_set1_pd(2.0);
 446     x                = xx[0];
 447     f                = ff[0];
 448
 449     nri              = nlist->nri;
 450     iinr             = nlist->iinr;
 451     jindex           = nlist->jindex;
 452     jjnr             = nlist->jjnr;
 453     shiftidx         = nlist->shift;
 454     gid              = nlist->gid;
 455     shiftvec         = fr->shift_vec[0];
 456     fshift           = fr->fshift[0];
 457     facel            = _mm_set1_pd(fr->epsfac);
 458     charge           = mdatoms->chargeA;
 459     nvdwtype         = fr->ntype;
 460     vdwparam         = fr->nbfp;
 461     vdwtype          = mdatoms->typeA;
 462
 463     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 464     ewtab            = fr->ic->tabq_coul_F;
 465     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 466     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 467
 468     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 469     rcutoff_scalar   = fr->rcoulomb;
 470     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 471     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 472
 473     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 474     rvdw             = _mm_set1_pd(fr->rvdw);
 475
 476     /* Avoid stupid compiler warnings */
 477     jnrA = jnrB = 0;
 478     j_coord_offsetA = 0;
 479     j_coord_offsetB = 0;
 480
 481     outeriter        = 0;
 482     inneriter        = 0;
 483
 484     /* Start outer loop over neighborlists */
 485     for(iidx=0; iidx<nri; iidx++)
 486     {
 487         /* Load shift vector for this list */
 488         i_shift_offset   = DIM*shiftidx[iidx];
 489
 490         /* Load limits for loop over neighbors */
 491         j_index_start    = jindex[iidx];
 492         j_index_end      = jindex[iidx+1];
 493
 494         /* Get outer coordinate index */
 495         inr              = iinr[iidx];
 496         i_coord_offset   = DIM*inr;
 497
 498         /* Load i particle coords and add shift vector */
 499         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 500
 501         fix0             = _mm_setzero_pd();
 502         fiy0             = _mm_setzero_pd();
 503         fiz0             = _mm_setzero_pd();
 504
 505         /* Load parameters for i particles */
 506         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 507         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 508
 509         /* Start inner kernel loop */
 510         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 511         {
 512
 513             /* Get j neighbor index, and coordinate index */
 514             jnrA             = jjnr[jidx];
 515             jnrB             = jjnr[jidx+1];
 516             j_coord_offsetA  = DIM*jnrA;
 517             j_coord_offsetB  = DIM*jnrB;
 518
 519             /* load j atom coordinates */
 520             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 521                                               &jx0,&jy0,&jz0);
 522
 523             /* Calculate displacement vector */
 524             dx00             = _mm_sub_pd(ix0,jx0);
 525             dy00             = _mm_sub_pd(iy0,jy0);
 526             dz00             = _mm_sub_pd(iz0,jz0);
 527
 528             /* Calculate squared distance and things based on it */
 529             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 530
 531             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 532
 533             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 534
 535             /* Load parameters for j particles */
 536             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 537             vdwjidx0A        = 2*vdwtype[jnrA+0];
 538             vdwjidx0B        = 2*vdwtype[jnrB+0];
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             if (gmx_mm_any_lt(rsq00,rcutoff2))
 545             {
 546
 547             r00              = _mm_mul_pd(rsq00,rinv00);
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             qq00             = _mm_mul_pd(iq0,jq0);
 551             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 552                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 553
 554             /* EWALD ELECTROSTATICS */
 555
 556             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 557             ewrt             = _mm_mul_pd(r00,ewtabscale);
 558             ewitab           = _mm_cvttpd_epi32(ewrt);
 559 #ifdef __XOP__
 560             eweps            = _mm_frcz_pd(ewrt);
 561 #else
 562             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 563 #endif
 564             twoeweps         = _mm_add_pd(eweps,eweps);
 565             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 566                                          &ewtabF,&ewtabFn);
 567             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 568             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 569
 570             /* LENNARD-JONES DISPERSION/REPULSION */
 571
 572             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 573             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 574
 575             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 576
 577             fscal            = _mm_add_pd(felec,fvdw);
 578
 579             fscal            = _mm_and_pd(fscal,cutoff_mask);
 580
 581             /* Update vectorial force */
 582             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 583             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 584             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 585
 586             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 587                                                    _mm_mul_pd(dx00,fscal),
 588                                                    _mm_mul_pd(dy00,fscal),
 589                                                    _mm_mul_pd(dz00,fscal));
 590
 591             }
 592
 593             /* Inner loop uses 49 flops */
 594         }
 595
 596         if(jidx<j_index_end)
 597         {
 598
 599             jnrA             = jjnr[jidx];
 600             j_coord_offsetA  = DIM*jnrA;
 601
 602             /* load j atom coordinates */
 603             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 604                                               &jx0,&jy0,&jz0);
 605
 606             /* Calculate displacement vector */
 607             dx00             = _mm_sub_pd(ix0,jx0);
 608             dy00             = _mm_sub_pd(iy0,jy0);
 609             dz00             = _mm_sub_pd(iz0,jz0);
 610
 611             /* Calculate squared distance and things based on it */
 612             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 613
 614             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 615
 616             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 617
 618             /* Load parameters for j particles */
 619             jq0              = _mm_load_sd(charge+jnrA+0);
 620             vdwjidx0A        = 2*vdwtype[jnrA+0];
 621
 622             /**************************
 623              * CALCULATE INTERACTIONS *
 624              **************************/
 625
 626             if (gmx_mm_any_lt(rsq00,rcutoff2))
 627             {
 628
 629             r00              = _mm_mul_pd(rsq00,rinv00);
 630
 631             /* Compute parameters for interactions between i and j atoms */
 632             qq00             = _mm_mul_pd(iq0,jq0);
 633             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 634
 635             /* EWALD ELECTROSTATICS */
 636
 637             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 638             ewrt             = _mm_mul_pd(r00,ewtabscale);
 639             ewitab           = _mm_cvttpd_epi32(ewrt);
 640 #ifdef __XOP__
 641             eweps            = _mm_frcz_pd(ewrt);
 642 #else
 643             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 644 #endif
 645             twoeweps         = _mm_add_pd(eweps,eweps);
 646             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 647             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 648             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 649
 650             /* LENNARD-JONES DISPERSION/REPULSION */
 651
 652             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 653             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 654
 655             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 656
 657             fscal            = _mm_add_pd(felec,fvdw);
 658
 659             fscal            = _mm_and_pd(fscal,cutoff_mask);
 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             }
 674
 675             /* Inner loop uses 49 flops */
 676         }
 677
 678         /* End of innermost loop */
 679
 680         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 681                                               f+i_coord_offset,fshift+i_shift_offset);
 682
 683         /* Increment number of inner iterations */
 684         inneriter                  += j_index_end - j_index_start;
 685
 686         /* Outer loop uses 7 flops */
 687     }
 688
 689     /* Increment number of outer iterations */
 690     outeriter        += nri;
 691
 692     /* Update outer/inner flops */
 693
 694     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*49);
 695 }