src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  98     real             *vftab;
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     krf              = _mm_set1_pd(fr->ic->k_rf);
 117     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 118     crf              = _mm_set1_pd(fr->ic->c_rf);
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 150
 151         fix0             = _mm_setzero_pd();
 152         fiy0             = _mm_setzero_pd();
 153         fiz0             = _mm_setzero_pd();
 154
 155         /* Load parameters for i particles */
 156         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 157         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_pd();
 161         vvdwsum          = _mm_setzero_pd();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             j_coord_offsetA  = DIM*jnrA;
 171             j_coord_offsetB  = DIM*jnrB;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               &jx0,&jy0,&jz0);
 176
 177             /* Calculate displacement vector */
 178             dx00             = _mm_sub_pd(ix0,jx0);
 179             dy00             = _mm_sub_pd(iy0,jy0);
 180             dz00             = _mm_sub_pd(iz0,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 184
 185             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 186
 187             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 188
 189             /* Load parameters for j particles */
 190             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 191             vdwjidx0A        = 2*vdwtype[jnrA+0];
 192             vdwjidx0B        = 2*vdwtype[jnrB+0];
 193
 194             /**************************
 195              * CALCULATE INTERACTIONS *
 196              **************************/
 197
 198             r00              = _mm_mul_pd(rsq00,rinv00);
 199
 200             /* Compute parameters for interactions between i and j atoms */
 201             qq00             = _mm_mul_pd(iq0,jq0);
 202             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 203                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 204
 205             /* Calculate table index by multiplying r with table scale and truncate to integer */
 206             rt               = _mm_mul_pd(r00,vftabscale);
 207             vfitab           = _mm_cvttpd_epi32(rt);
 208 #ifdef __XOP__
 209             vfeps            = _mm_frcz_pd(rt);
 210 #else
 211             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 212 #endif
 213             twovfeps         = _mm_add_pd(vfeps,vfeps);
 214             vfitab           = _mm_slli_epi32(vfitab,3);
 215
 216             /* REACTION-FIELD ELECTROSTATICS */
 217             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 218             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 219
 220             /* CUBIC SPLINE TABLE DISPERSION */
 221             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 222             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 223             GMX_MM_TRANSPOSE2_PD(Y,F);
 224             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 225             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 226             GMX_MM_TRANSPOSE2_PD(G,H);
 227             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 228             VV               = _mm_macc_pd(vfeps,Fp,Y);
 229             vvdw6            = _mm_mul_pd(c6_00,VV);
 230             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 231             fvdw6            = _mm_mul_pd(c6_00,FF);
 232
 233             /* CUBIC SPLINE TABLE REPULSION */
 234             vfitab           = _mm_add_epi32(vfitab,ifour);
 235             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 236             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 237             GMX_MM_TRANSPOSE2_PD(Y,F);
 238             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 239             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 240             GMX_MM_TRANSPOSE2_PD(G,H);
 241             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 242             VV               = _mm_macc_pd(vfeps,Fp,Y);
 243             vvdw12           = _mm_mul_pd(c12_00,VV);
 244             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 245             fvdw12           = _mm_mul_pd(c12_00,FF);
 246             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 247             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 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 70 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
 304             /* Calculate table index by multiplying r with table scale and truncate to integer */
 305             rt               = _mm_mul_pd(r00,vftabscale);
 306             vfitab           = _mm_cvttpd_epi32(rt);
 307 #ifdef __XOP__
 308             vfeps            = _mm_frcz_pd(rt);
 309 #else
 310             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 311 #endif
 312             twovfeps         = _mm_add_pd(vfeps,vfeps);
 313             vfitab           = _mm_slli_epi32(vfitab,3);
 314
 315             /* REACTION-FIELD ELECTROSTATICS */
 316             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 317             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 318
 319             /* CUBIC SPLINE TABLE DISPERSION */
 320             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 321             F                = _mm_setzero_pd();
 322             GMX_MM_TRANSPOSE2_PD(Y,F);
 323             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 324             H                = _mm_setzero_pd();
 325             GMX_MM_TRANSPOSE2_PD(G,H);
 326             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 327             VV               = _mm_macc_pd(vfeps,Fp,Y);
 328             vvdw6            = _mm_mul_pd(c6_00,VV);
 329             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 330             fvdw6            = _mm_mul_pd(c6_00,FF);
 331
 332             /* CUBIC SPLINE TABLE REPULSION */
 333             vfitab           = _mm_add_epi32(vfitab,ifour);
 334             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 335             F                = _mm_setzero_pd();
 336             GMX_MM_TRANSPOSE2_PD(Y,F);
 337             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 338             H                = _mm_setzero_pd();
 339             GMX_MM_TRANSPOSE2_PD(G,H);
 340             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 341             VV               = _mm_macc_pd(vfeps,Fp,Y);
 342             vvdw12           = _mm_mul_pd(c12_00,VV);
 343             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 344             fvdw12           = _mm_mul_pd(c12_00,FF);
 345             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 346             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 350             velecsum         = _mm_add_pd(velecsum,velec);
 351             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 352             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 353
 354             fscal            = _mm_add_pd(felec,fvdw);
 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             /* Inner loop uses 70 flops */
 369         }
 370
 371         /* End of innermost loop */
 372
 373         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 374                                               f+i_coord_offset,fshift+i_shift_offset);
 375
 376         ggid                        = gid[iidx];
 377         /* Update potential energies */
 378         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 379         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 380
 381         /* Increment number of inner iterations */
 382         inneriter                  += j_index_end - j_index_start;
 383
 384         /* Outer loop uses 9 flops */
 385     }
 386
 387     /* Increment number of outer iterations */
 388     outeriter        += nri;
 389
 390     /* Update outer/inner flops */
 391
 392     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 393 }
 394 /*
 395  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 396  * Electrostatics interaction: ReactionField
 397  * VdW interaction:            CubicSplineTable
 398  * Geometry:                   Particle-Particle
 399  * Calculate force/pot:        Force
 400  */
 401 void
 402 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 403                     (t_nblist                    * gmx_restrict       nlist,
 404                      rvec                        * gmx_restrict          xx,
 405                      rvec                        * gmx_restrict          ff,
 406                      t_forcerec                  * gmx_restrict          fr,
 407                      t_mdatoms                   * gmx_restrict     mdatoms,
 408                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 409                      t_nrnb                      * gmx_restrict        nrnb)
 410 {
 411     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 412      * just 0 for non-waters.
 413      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 414      * jnr indices corresponding to data put in the four positions in the SIMD register.
 415      */
 416     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 417     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 418     int              jnrA,jnrB;
 419     int              j_coord_offsetA,j_coord_offsetB;
 420     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 421     real             rcutoff_scalar;
 422     real             *shiftvec,*fshift,*x,*f;
 423     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 424     int              vdwioffset0;
 425     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 426     int              vdwjidx0A,vdwjidx0B;
 427     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 428     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 429     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 430     real             *charge;
 431     int              nvdwtype;
 432     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 433     int              *vdwtype;
 434     real             *vdwparam;
 435     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 436     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 437     __m128i          vfitab;
 438     __m128i          ifour       = _mm_set1_epi32(4);
 439     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 440     real             *vftab;
 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     krf              = _mm_set1_pd(fr->ic->k_rf);
 459     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 460     crf              = _mm_set1_pd(fr->ic->c_rf);
 461     nvdwtype         = fr->ntype;
 462     vdwparam         = fr->nbfp;
 463     vdwtype          = mdatoms->typeA;
 464
 465     vftab            = kernel_data->table_vdw->data;
 466     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 467
 468     /* Avoid stupid compiler warnings */
 469     jnrA = jnrB = 0;
 470     j_coord_offsetA = 0;
 471     j_coord_offsetB = 0;
 472
 473     outeriter        = 0;
 474     inneriter        = 0;
 475
 476     /* Start outer loop over neighborlists */
 477     for(iidx=0; iidx<nri; iidx++)
 478     {
 479         /* Load shift vector for this list */
 480         i_shift_offset   = DIM*shiftidx[iidx];
 481
 482         /* Load limits for loop over neighbors */
 483         j_index_start    = jindex[iidx];
 484         j_index_end      = jindex[iidx+1];
 485
 486         /* Get outer coordinate index */
 487         inr              = iinr[iidx];
 488         i_coord_offset   = DIM*inr;
 489
 490         /* Load i particle coords and add shift vector */
 491         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 492
 493         fix0             = _mm_setzero_pd();
 494         fiy0             = _mm_setzero_pd();
 495         fiz0             = _mm_setzero_pd();
 496
 497         /* Load parameters for i particles */
 498         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 499         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 500
 501         /* Start inner kernel loop */
 502         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 503         {
 504
 505             /* Get j neighbor index, and coordinate index */
 506             jnrA             = jjnr[jidx];
 507             jnrB             = jjnr[jidx+1];
 508             j_coord_offsetA  = DIM*jnrA;
 509             j_coord_offsetB  = DIM*jnrB;
 510
 511             /* load j atom coordinates */
 512             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 513                                               &jx0,&jy0,&jz0);
 514
 515             /* Calculate displacement vector */
 516             dx00             = _mm_sub_pd(ix0,jx0);
 517             dy00             = _mm_sub_pd(iy0,jy0);
 518             dz00             = _mm_sub_pd(iz0,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 522
 523             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 524
 525             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 526
 527             /* Load parameters for j particles */
 528             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 529             vdwjidx0A        = 2*vdwtype[jnrA+0];
 530             vdwjidx0B        = 2*vdwtype[jnrB+0];
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             r00              = _mm_mul_pd(rsq00,rinv00);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq00             = _mm_mul_pd(iq0,jq0);
 540             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 541                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 542
 543             /* Calculate table index by multiplying r with table scale and truncate to integer */
 544             rt               = _mm_mul_pd(r00,vftabscale);
 545             vfitab           = _mm_cvttpd_epi32(rt);
 546 #ifdef __XOP__
 547             vfeps            = _mm_frcz_pd(rt);
 548 #else
 549             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 550 #endif
 551             twovfeps         = _mm_add_pd(vfeps,vfeps);
 552             vfitab           = _mm_slli_epi32(vfitab,3);
 553
 554             /* REACTION-FIELD ELECTROSTATICS */
 555             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 556
 557             /* CUBIC SPLINE TABLE DISPERSION */
 558             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 559             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 560             GMX_MM_TRANSPOSE2_PD(Y,F);
 561             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 562             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 563             GMX_MM_TRANSPOSE2_PD(G,H);
 564             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 565             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 566             fvdw6            = _mm_mul_pd(c6_00,FF);
 567
 568             /* CUBIC SPLINE TABLE REPULSION */
 569             vfitab           = _mm_add_epi32(vfitab,ifour);
 570             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 571             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 572             GMX_MM_TRANSPOSE2_PD(Y,F);
 573             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 574             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 575             GMX_MM_TRANSPOSE2_PD(G,H);
 576             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 577             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 578             fvdw12           = _mm_mul_pd(c12_00,FF);
 579             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 580
 581             fscal            = _mm_add_pd(felec,fvdw);
 582
 583             /* Update vectorial force */
 584             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 585             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 586             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 587
 588             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 589                                                    _mm_mul_pd(dx00,fscal),
 590                                                    _mm_mul_pd(dy00,fscal),
 591                                                    _mm_mul_pd(dz00,fscal));
 592
 593             /* Inner loop uses 57 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             r00              = _mm_mul_pd(rsq00,rinv00);
 627
 628             /* Compute parameters for interactions between i and j atoms */
 629             qq00             = _mm_mul_pd(iq0,jq0);
 630             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 631
 632             /* Calculate table index by multiplying r with table scale and truncate to integer */
 633             rt               = _mm_mul_pd(r00,vftabscale);
 634             vfitab           = _mm_cvttpd_epi32(rt);
 635 #ifdef __XOP__
 636             vfeps            = _mm_frcz_pd(rt);
 637 #else
 638             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 639 #endif
 640             twovfeps         = _mm_add_pd(vfeps,vfeps);
 641             vfitab           = _mm_slli_epi32(vfitab,3);
 642
 643             /* REACTION-FIELD ELECTROSTATICS */
 644             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 645
 646             /* CUBIC SPLINE TABLE DISPERSION */
 647             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 648             F                = _mm_setzero_pd();
 649             GMX_MM_TRANSPOSE2_PD(Y,F);
 650             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 651             H                = _mm_setzero_pd();
 652             GMX_MM_TRANSPOSE2_PD(G,H);
 653             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 654             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 655             fvdw6            = _mm_mul_pd(c6_00,FF);
 656
 657             /* CUBIC SPLINE TABLE REPULSION */
 658             vfitab           = _mm_add_epi32(vfitab,ifour);
 659             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 660             F                = _mm_setzero_pd();
 661             GMX_MM_TRANSPOSE2_PD(Y,F);
 662             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 663             H                = _mm_setzero_pd();
 664             GMX_MM_TRANSPOSE2_PD(G,H);
 665             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 666             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 667             fvdw12           = _mm_mul_pd(c12_00,FF);
 668             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 669
 670             fscal            = _mm_add_pd(felec,fvdw);
 671
 672             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 673
 674             /* Update vectorial force */
 675             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 676             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 677             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 678
 679             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 680                                                    _mm_mul_pd(dx00,fscal),
 681                                                    _mm_mul_pd(dy00,fscal),
 682                                                    _mm_mul_pd(dz00,fscal));
 683
 684             /* Inner loop uses 57 flops */
 685         }
 686
 687         /* End of innermost loop */
 688
 689         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 690                                               f+i_coord_offset,fshift+i_shift_offset);
 691
 692         /* Increment number of inner iterations */
 693         inneriter                  += j_index_end - j_index_start;
 694
 695         /* Outer loop uses 7 flops */
 696     }
 697
 698     /* Increment number of outer iterations */
 699     outeriter        += nri;
 700
 701     /* Update outer/inner flops */
 702
 703     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 704 }