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) 2012,2013,2014, 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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_ElecRF_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_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          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  96     real             *vftab;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114     krf              = _mm_set1_pd(fr->ic->k_rf);
 115     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 116     crf              = _mm_set1_pd(fr->ic->c_rf);
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     vftab            = kernel_data->table_vdw->data;
 122     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_pd();
 150         fiy0             = _mm_setzero_pd();
 151         fiz0             = _mm_setzero_pd();
 152
 153         /* Load parameters for i particles */
 154         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 155         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_pd();
 159         vvdwsum          = _mm_setzero_pd();
 160
 161         /* Start inner kernel loop */
 162         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 163         {
 164
 165             /* Get j neighbor index, and coordinate index */
 166             jnrA             = jjnr[jidx];
 167             jnrB             = jjnr[jidx+1];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170
 171             /* load j atom coordinates */
 172             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm_sub_pd(ix0,jx0);
 177             dy00             = _mm_sub_pd(iy0,jy0);
 178             dz00             = _mm_sub_pd(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 182
 183             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 184
 185             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 186
 187             /* Load parameters for j particles */
 188             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 189             vdwjidx0A        = 2*vdwtype[jnrA+0];
 190             vdwjidx0B        = 2*vdwtype[jnrB+0];
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             r00              = _mm_mul_pd(rsq00,rinv00);
 197
 198             /* Compute parameters for interactions between i and j atoms */
 199             qq00             = _mm_mul_pd(iq0,jq0);
 200             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 201                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 202
 203             /* Calculate table index by multiplying r with table scale and truncate to integer */
 204             rt               = _mm_mul_pd(r00,vftabscale);
 205             vfitab           = _mm_cvttpd_epi32(rt);
 206 #ifdef __XOP__
 207             vfeps            = _mm_frcz_pd(rt);
 208 #else
 209             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 210 #endif
 211             twovfeps         = _mm_add_pd(vfeps,vfeps);
 212             vfitab           = _mm_slli_epi32(vfitab,3);
 213
 214             /* REACTION-FIELD ELECTROSTATICS */
 215             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 216             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 217
 218             /* CUBIC SPLINE TABLE DISPERSION */
 219             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 220             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 221             GMX_MM_TRANSPOSE2_PD(Y,F);
 222             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 223             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 224             GMX_MM_TRANSPOSE2_PD(G,H);
 225             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 226             VV               = _mm_macc_pd(vfeps,Fp,Y);
 227             vvdw6            = _mm_mul_pd(c6_00,VV);
 228             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 229             fvdw6            = _mm_mul_pd(c6_00,FF);
 230
 231             /* CUBIC SPLINE TABLE REPULSION */
 232             vfitab           = _mm_add_epi32(vfitab,ifour);
 233             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 234             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 235             GMX_MM_TRANSPOSE2_PD(Y,F);
 236             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 237             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 238             GMX_MM_TRANSPOSE2_PD(G,H);
 239             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 240             VV               = _mm_macc_pd(vfeps,Fp,Y);
 241             vvdw12           = _mm_mul_pd(c12_00,VV);
 242             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 243             fvdw12           = _mm_mul_pd(c12_00,FF);
 244             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 245             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velecsum         = _mm_add_pd(velecsum,velec);
 249             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 250
 251             fscal            = _mm_add_pd(felec,fvdw);
 252
 253             /* Update vectorial force */
 254             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 255             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 256             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 257
 258             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 259                                                    _mm_mul_pd(dx00,fscal),
 260                                                    _mm_mul_pd(dy00,fscal),
 261                                                    _mm_mul_pd(dz00,fscal));
 262
 263             /* Inner loop uses 70 flops */
 264         }
 265
 266         if(jidx<j_index_end)
 267         {
 268
 269             jnrA             = jjnr[jidx];
 270             j_coord_offsetA  = DIM*jnrA;
 271
 272             /* load j atom coordinates */
 273             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 274                                               &jx0,&jy0,&jz0);
 275
 276             /* Calculate displacement vector */
 277             dx00             = _mm_sub_pd(ix0,jx0);
 278             dy00             = _mm_sub_pd(iy0,jy0);
 279             dz00             = _mm_sub_pd(iz0,jz0);
 280
 281             /* Calculate squared distance and things based on it */
 282             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 283
 284             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 285
 286             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 287
 288             /* Load parameters for j particles */
 289             jq0              = _mm_load_sd(charge+jnrA+0);
 290             vdwjidx0A        = 2*vdwtype[jnrA+0];
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             r00              = _mm_mul_pd(rsq00,rinv00);
 297
 298             /* Compute parameters for interactions between i and j atoms */
 299             qq00             = _mm_mul_pd(iq0,jq0);
 300             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 301
 302             /* Calculate table index by multiplying r with table scale and truncate to integer */
 303             rt               = _mm_mul_pd(r00,vftabscale);
 304             vfitab           = _mm_cvttpd_epi32(rt);
 305 #ifdef __XOP__
 306             vfeps            = _mm_frcz_pd(rt);
 307 #else
 308             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 309 #endif
 310             twovfeps         = _mm_add_pd(vfeps,vfeps);
 311             vfitab           = _mm_slli_epi32(vfitab,3);
 312
 313             /* REACTION-FIELD ELECTROSTATICS */
 314             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 315             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 316
 317             /* CUBIC SPLINE TABLE DISPERSION */
 318             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 319             F                = _mm_setzero_pd();
 320             GMX_MM_TRANSPOSE2_PD(Y,F);
 321             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 322             H                = _mm_setzero_pd();
 323             GMX_MM_TRANSPOSE2_PD(G,H);
 324             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 325             VV               = _mm_macc_pd(vfeps,Fp,Y);
 326             vvdw6            = _mm_mul_pd(c6_00,VV);
 327             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 328             fvdw6            = _mm_mul_pd(c6_00,FF);
 329
 330             /* CUBIC SPLINE TABLE REPULSION */
 331             vfitab           = _mm_add_epi32(vfitab,ifour);
 332             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 333             F                = _mm_setzero_pd();
 334             GMX_MM_TRANSPOSE2_PD(Y,F);
 335             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 336             H                = _mm_setzero_pd();
 337             GMX_MM_TRANSPOSE2_PD(G,H);
 338             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 339             VV               = _mm_macc_pd(vfeps,Fp,Y);
 340             vvdw12           = _mm_mul_pd(c12_00,VV);
 341             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 342             fvdw12           = _mm_mul_pd(c12_00,FF);
 343             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 344             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 348             velecsum         = _mm_add_pd(velecsum,velec);
 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_unpacklo_pd(fscal,_mm_setzero_pd());
 355
 356             /* Update vectorial force */
 357             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 358             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 359             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 360
 361             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 362                                                    _mm_mul_pd(dx00,fscal),
 363                                                    _mm_mul_pd(dy00,fscal),
 364                                                    _mm_mul_pd(dz00,fscal));
 365
 366             /* Inner loop uses 70 flops */
 367         }
 368
 369         /* End of innermost loop */
 370
 371         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 372                                               f+i_coord_offset,fshift+i_shift_offset);
 373
 374         ggid                        = gid[iidx];
 375         /* Update potential energies */
 376         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 377         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 378
 379         /* Increment number of inner iterations */
 380         inneriter                  += j_index_end - j_index_start;
 381
 382         /* Outer loop uses 9 flops */
 383     }
 384
 385     /* Increment number of outer iterations */
 386     outeriter        += nri;
 387
 388     /* Update outer/inner flops */
 389
 390     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 391 }
 392 /*
 393  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 394  * Electrostatics interaction: ReactionField
 395  * VdW interaction:            CubicSplineTable
 396  * Geometry:                   Particle-Particle
 397  * Calculate force/pot:        Force
 398  */
 399 void
 400 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 401                     (t_nblist                    * gmx_restrict       nlist,
 402                      rvec                        * gmx_restrict          xx,
 403                      rvec                        * gmx_restrict          ff,
 404                      t_forcerec                  * gmx_restrict          fr,
 405                      t_mdatoms                   * gmx_restrict     mdatoms,
 406                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 407                      t_nrnb                      * gmx_restrict        nrnb)
 408 {
 409     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 410      * just 0 for non-waters.
 411      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 412      * jnr indices corresponding to data put in the four positions in the SIMD register.
 413      */
 414     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 415     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 416     int              jnrA,jnrB;
 417     int              j_coord_offsetA,j_coord_offsetB;
 418     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 419     real             rcutoff_scalar;
 420     real             *shiftvec,*fshift,*x,*f;
 421     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 422     int              vdwioffset0;
 423     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 424     int              vdwjidx0A,vdwjidx0B;
 425     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 426     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 427     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 428     real             *charge;
 429     int              nvdwtype;
 430     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 431     int              *vdwtype;
 432     real             *vdwparam;
 433     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 434     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 435     __m128i          vfitab;
 436     __m128i          ifour       = _mm_set1_epi32(4);
 437     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 438     real             *vftab;
 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     krf              = _mm_set1_pd(fr->ic->k_rf);
 457     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 458     crf              = _mm_set1_pd(fr->ic->c_rf);
 459     nvdwtype         = fr->ntype;
 460     vdwparam         = fr->nbfp;
 461     vdwtype          = mdatoms->typeA;
 462
 463     vftab            = kernel_data->table_vdw->data;
 464     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 465
 466     /* Avoid stupid compiler warnings */
 467     jnrA = jnrB = 0;
 468     j_coord_offsetA = 0;
 469     j_coord_offsetB = 0;
 470
 471     outeriter        = 0;
 472     inneriter        = 0;
 473
 474     /* Start outer loop over neighborlists */
 475     for(iidx=0; iidx<nri; iidx++)
 476     {
 477         /* Load shift vector for this list */
 478         i_shift_offset   = DIM*shiftidx[iidx];
 479
 480         /* Load limits for loop over neighbors */
 481         j_index_start    = jindex[iidx];
 482         j_index_end      = jindex[iidx+1];
 483
 484         /* Get outer coordinate index */
 485         inr              = iinr[iidx];
 486         i_coord_offset   = DIM*inr;
 487
 488         /* Load i particle coords and add shift vector */
 489         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 490
 491         fix0             = _mm_setzero_pd();
 492         fiy0             = _mm_setzero_pd();
 493         fiz0             = _mm_setzero_pd();
 494
 495         /* Load parameters for i particles */
 496         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 497         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 498
 499         /* Start inner kernel loop */
 500         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 501         {
 502
 503             /* Get j neighbor index, and coordinate index */
 504             jnrA             = jjnr[jidx];
 505             jnrB             = jjnr[jidx+1];
 506             j_coord_offsetA  = DIM*jnrA;
 507             j_coord_offsetB  = DIM*jnrB;
 508
 509             /* load j atom coordinates */
 510             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 511                                               &jx0,&jy0,&jz0);
 512
 513             /* Calculate displacement vector */
 514             dx00             = _mm_sub_pd(ix0,jx0);
 515             dy00             = _mm_sub_pd(iy0,jy0);
 516             dz00             = _mm_sub_pd(iz0,jz0);
 517
 518             /* Calculate squared distance and things based on it */
 519             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 520
 521             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 522
 523             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 524
 525             /* Load parameters for j particles */
 526             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528             vdwjidx0B        = 2*vdwtype[jnrB+0];
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r00              = _mm_mul_pd(rsq00,rinv00);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq00             = _mm_mul_pd(iq0,jq0);
 538             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 539                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 540
 541             /* Calculate table index by multiplying r with table scale and truncate to integer */
 542             rt               = _mm_mul_pd(r00,vftabscale);
 543             vfitab           = _mm_cvttpd_epi32(rt);
 544 #ifdef __XOP__
 545             vfeps            = _mm_frcz_pd(rt);
 546 #else
 547             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 548 #endif
 549             twovfeps         = _mm_add_pd(vfeps,vfeps);
 550             vfitab           = _mm_slli_epi32(vfitab,3);
 551
 552             /* REACTION-FIELD ELECTROSTATICS */
 553             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 554
 555             /* CUBIC SPLINE TABLE DISPERSION */
 556             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 557             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 558             GMX_MM_TRANSPOSE2_PD(Y,F);
 559             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 560             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 561             GMX_MM_TRANSPOSE2_PD(G,H);
 562             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 563             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 564             fvdw6            = _mm_mul_pd(c6_00,FF);
 565
 566             /* CUBIC SPLINE TABLE REPULSION */
 567             vfitab           = _mm_add_epi32(vfitab,ifour);
 568             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 569             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 570             GMX_MM_TRANSPOSE2_PD(Y,F);
 571             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 572             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 573             GMX_MM_TRANSPOSE2_PD(G,H);
 574             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 575             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 576             fvdw12           = _mm_mul_pd(c12_00,FF);
 577             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 578
 579             fscal            = _mm_add_pd(felec,fvdw);
 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             /* Inner loop uses 57 flops */
 592         }
 593
 594         if(jidx<j_index_end)
 595         {
 596
 597             jnrA             = jjnr[jidx];
 598             j_coord_offsetA  = DIM*jnrA;
 599
 600             /* load j atom coordinates */
 601             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 602                                               &jx0,&jy0,&jz0);
 603
 604             /* Calculate displacement vector */
 605             dx00             = _mm_sub_pd(ix0,jx0);
 606             dy00             = _mm_sub_pd(iy0,jy0);
 607             dz00             = _mm_sub_pd(iz0,jz0);
 608
 609             /* Calculate squared distance and things based on it */
 610             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 611
 612             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 613
 614             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 615
 616             /* Load parameters for j particles */
 617             jq0              = _mm_load_sd(charge+jnrA+0);
 618             vdwjidx0A        = 2*vdwtype[jnrA+0];
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             r00              = _mm_mul_pd(rsq00,rinv00);
 625
 626             /* Compute parameters for interactions between i and j atoms */
 627             qq00             = _mm_mul_pd(iq0,jq0);
 628             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 629
 630             /* Calculate table index by multiplying r with table scale and truncate to integer */
 631             rt               = _mm_mul_pd(r00,vftabscale);
 632             vfitab           = _mm_cvttpd_epi32(rt);
 633 #ifdef __XOP__
 634             vfeps            = _mm_frcz_pd(rt);
 635 #else
 636             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 637 #endif
 638             twovfeps         = _mm_add_pd(vfeps,vfeps);
 639             vfitab           = _mm_slli_epi32(vfitab,3);
 640
 641             /* REACTION-FIELD ELECTROSTATICS */
 642             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 643
 644             /* CUBIC SPLINE TABLE DISPERSION */
 645             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 646             F                = _mm_setzero_pd();
 647             GMX_MM_TRANSPOSE2_PD(Y,F);
 648             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 649             H                = _mm_setzero_pd();
 650             GMX_MM_TRANSPOSE2_PD(G,H);
 651             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 652             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 653             fvdw6            = _mm_mul_pd(c6_00,FF);
 654
 655             /* CUBIC SPLINE TABLE REPULSION */
 656             vfitab           = _mm_add_epi32(vfitab,ifour);
 657             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 658             F                = _mm_setzero_pd();
 659             GMX_MM_TRANSPOSE2_PD(Y,F);
 660             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 661             H                = _mm_setzero_pd();
 662             GMX_MM_TRANSPOSE2_PD(G,H);
 663             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 664             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 665             fvdw12           = _mm_mul_pd(c12_00,FF);
 666             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 667
 668             fscal            = _mm_add_pd(felec,fvdw);
 669
 670             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 671
 672             /* Update vectorial force */
 673             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 674             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 675             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 676
 677             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 678                                                    _mm_mul_pd(dx00,fscal),
 679                                                    _mm_mul_pd(dy00,fscal),
 680                                                    _mm_mul_pd(dz00,fscal));
 681
 682             /* Inner loop uses 57 flops */
 683         }
 684
 685         /* End of innermost loop */
 686
 687         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 688                                               f+i_coord_offset,fshift+i_shift_offset);
 689
 690         /* Increment number of inner iterations */
 691         inneriter                  += j_index_end - j_index_start;
 692
 693         /* Outer loop uses 7 flops */
 694     }
 695
 696     /* Increment number of outer iterations */
 697     outeriter        += nri;
 698
 699     /* Update outer/inner flops */
 700
 701     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 702 }