src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_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 "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_ElecCoul_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_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     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117
 118     vftab            = kernel_data->table_vdw->data;
 119     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149
 150         /* Load parameters for i particles */
 151         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 152         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 153
 154         /* Reset potential sums */
 155         velecsum         = _mm_setzero_pd();
 156         vvdwsum          = _mm_setzero_pd();
 157
 158         /* Start inner kernel loop */
 159         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 160         {
 161
 162             /* Get j neighbor index, and coordinate index */
 163             jnrA             = jjnr[jidx];
 164             jnrB             = jjnr[jidx+1];
 165             j_coord_offsetA  = DIM*jnrA;
 166             j_coord_offsetB  = DIM*jnrB;
 167
 168             /* load j atom coordinates */
 169             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                               &jx0,&jy0,&jz0);
 171
 172             /* Calculate displacement vector */
 173             dx00             = _mm_sub_pd(ix0,jx0);
 174             dy00             = _mm_sub_pd(iy0,jy0);
 175             dz00             = _mm_sub_pd(iz0,jz0);
 176
 177             /* Calculate squared distance and things based on it */
 178             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 179
 180             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 181
 182             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 183
 184             /* Load parameters for j particles */
 185             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 186             vdwjidx0A        = 2*vdwtype[jnrA+0];
 187             vdwjidx0B        = 2*vdwtype[jnrB+0];
 188
 189             /**************************
 190              * CALCULATE INTERACTIONS *
 191              **************************/
 192
 193             r00              = _mm_mul_pd(rsq00,rinv00);
 194
 195             /* Compute parameters for interactions between i and j atoms */
 196             qq00             = _mm_mul_pd(iq0,jq0);
 197             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 198                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 199
 200             /* Calculate table index by multiplying r with table scale and truncate to integer */
 201             rt               = _mm_mul_pd(r00,vftabscale);
 202             vfitab           = _mm_cvttpd_epi32(rt);
 203 #ifdef __XOP__
 204             vfeps            = _mm_frcz_pd(rt);
 205 #else
 206             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 207 #endif
 208             twovfeps         = _mm_add_pd(vfeps,vfeps);
 209             vfitab           = _mm_slli_epi32(vfitab,3);
 210
 211             /* COULOMB ELECTROSTATICS */
 212             velec            = _mm_mul_pd(qq00,rinv00);
 213             felec            = _mm_mul_pd(velec,rinvsq00);
 214
 215             /* CUBIC SPLINE TABLE DISPERSION */
 216             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 217             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 218             GMX_MM_TRANSPOSE2_PD(Y,F);
 219             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 220             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 221             GMX_MM_TRANSPOSE2_PD(G,H);
 222             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 223             VV               = _mm_macc_pd(vfeps,Fp,Y);
 224             vvdw6            = _mm_mul_pd(c6_00,VV);
 225             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 226             fvdw6            = _mm_mul_pd(c6_00,FF);
 227
 228             /* CUBIC SPLINE TABLE REPULSION */
 229             vfitab           = _mm_add_epi32(vfitab,ifour);
 230             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 231             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 232             GMX_MM_TRANSPOSE2_PD(Y,F);
 233             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 234             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 235             GMX_MM_TRANSPOSE2_PD(G,H);
 236             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 237             VV               = _mm_macc_pd(vfeps,Fp,Y);
 238             vvdw12           = _mm_mul_pd(c12_00,VV);
 239             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 240             fvdw12           = _mm_mul_pd(c12_00,FF);
 241             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 242             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velecsum         = _mm_add_pd(velecsum,velec);
 246             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 247
 248             fscal            = _mm_add_pd(felec,fvdw);
 249
 250             /* Update vectorial force */
 251             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 252             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 253             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 254
 255             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 256                                                    _mm_mul_pd(dx00,fscal),
 257                                                    _mm_mul_pd(dy00,fscal),
 258                                                    _mm_mul_pd(dz00,fscal));
 259
 260             /* Inner loop uses 66 flops */
 261         }
 262
 263         if(jidx<j_index_end)
 264         {
 265
 266             jnrA             = jjnr[jidx];
 267             j_coord_offsetA  = DIM*jnrA;
 268
 269             /* load j atom coordinates */
 270             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 271                                               &jx0,&jy0,&jz0);
 272
 273             /* Calculate displacement vector */
 274             dx00             = _mm_sub_pd(ix0,jx0);
 275             dy00             = _mm_sub_pd(iy0,jy0);
 276             dz00             = _mm_sub_pd(iz0,jz0);
 277
 278             /* Calculate squared distance and things based on it */
 279             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 280
 281             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 282
 283             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 284
 285             /* Load parameters for j particles */
 286             jq0              = _mm_load_sd(charge+jnrA+0);
 287             vdwjidx0A        = 2*vdwtype[jnrA+0];
 288
 289             /**************************
 290              * CALCULATE INTERACTIONS *
 291              **************************/
 292
 293             r00              = _mm_mul_pd(rsq00,rinv00);
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             qq00             = _mm_mul_pd(iq0,jq0);
 297             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 298
 299             /* Calculate table index by multiplying r with table scale and truncate to integer */
 300             rt               = _mm_mul_pd(r00,vftabscale);
 301             vfitab           = _mm_cvttpd_epi32(rt);
 302 #ifdef __XOP__
 303             vfeps            = _mm_frcz_pd(rt);
 304 #else
 305             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 306 #endif
 307             twovfeps         = _mm_add_pd(vfeps,vfeps);
 308             vfitab           = _mm_slli_epi32(vfitab,3);
 309
 310             /* COULOMB ELECTROSTATICS */
 311             velec            = _mm_mul_pd(qq00,rinv00);
 312             felec            = _mm_mul_pd(velec,rinvsq00);
 313
 314             /* CUBIC SPLINE TABLE DISPERSION */
 315             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 316             F                = _mm_setzero_pd();
 317             GMX_MM_TRANSPOSE2_PD(Y,F);
 318             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 319             H                = _mm_setzero_pd();
 320             GMX_MM_TRANSPOSE2_PD(G,H);
 321             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 322             VV               = _mm_macc_pd(vfeps,Fp,Y);
 323             vvdw6            = _mm_mul_pd(c6_00,VV);
 324             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 325             fvdw6            = _mm_mul_pd(c6_00,FF);
 326
 327             /* CUBIC SPLINE TABLE REPULSION */
 328             vfitab           = _mm_add_epi32(vfitab,ifour);
 329             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 330             F                = _mm_setzero_pd();
 331             GMX_MM_TRANSPOSE2_PD(Y,F);
 332             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 333             H                = _mm_setzero_pd();
 334             GMX_MM_TRANSPOSE2_PD(G,H);
 335             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 336             VV               = _mm_macc_pd(vfeps,Fp,Y);
 337             vvdw12           = _mm_mul_pd(c12_00,VV);
 338             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 339             fvdw12           = _mm_mul_pd(c12_00,FF);
 340             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 341             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 345             velecsum         = _mm_add_pd(velecsum,velec);
 346             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 347             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 348
 349             fscal            = _mm_add_pd(felec,fvdw);
 350
 351             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 352
 353             /* Update vectorial force */
 354             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 355             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 356             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 357
 358             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 359                                                    _mm_mul_pd(dx00,fscal),
 360                                                    _mm_mul_pd(dy00,fscal),
 361                                                    _mm_mul_pd(dz00,fscal));
 362
 363             /* Inner loop uses 66 flops */
 364         }
 365
 366         /* End of innermost loop */
 367
 368         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 369                                               f+i_coord_offset,fshift+i_shift_offset);
 370
 371         ggid                        = gid[iidx];
 372         /* Update potential energies */
 373         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 374         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 375
 376         /* Increment number of inner iterations */
 377         inneriter                  += j_index_end - j_index_start;
 378
 379         /* Outer loop uses 9 flops */
 380     }
 381
 382     /* Increment number of outer iterations */
 383     outeriter        += nri;
 384
 385     /* Update outer/inner flops */
 386
 387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*66);
 388 }
 389 /*
 390  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 391  * Electrostatics interaction: Coulomb
 392  * VdW interaction:            CubicSplineTable
 393  * Geometry:                   Particle-Particle
 394  * Calculate force/pot:        Force
 395  */
 396 void
 397 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 398                     (t_nblist                    * gmx_restrict       nlist,
 399                      rvec                        * gmx_restrict          xx,
 400                      rvec                        * gmx_restrict          ff,
 401                      t_forcerec                  * gmx_restrict          fr,
 402                      t_mdatoms                   * gmx_restrict     mdatoms,
 403                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 404                      t_nrnb                      * gmx_restrict        nrnb)
 405 {
 406     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 407      * just 0 for non-waters.
 408      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 409      * jnr indices corresponding to data put in the four positions in the SIMD register.
 410      */
 411     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 412     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 413     int              jnrA,jnrB;
 414     int              j_coord_offsetA,j_coord_offsetB;
 415     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 416     real             rcutoff_scalar;
 417     real             *shiftvec,*fshift,*x,*f;
 418     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 419     int              vdwioffset0;
 420     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 421     int              vdwjidx0A,vdwjidx0B;
 422     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 423     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 424     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 425     real             *charge;
 426     int              nvdwtype;
 427     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 428     int              *vdwtype;
 429     real             *vdwparam;
 430     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 431     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 432     __m128i          vfitab;
 433     __m128i          ifour       = _mm_set1_epi32(4);
 434     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 435     real             *vftab;
 436     __m128d          dummy_mask,cutoff_mask;
 437     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 438     __m128d          one     = _mm_set1_pd(1.0);
 439     __m128d          two     = _mm_set1_pd(2.0);
 440     x                = xx[0];
 441     f                = ff[0];
 442
 443     nri              = nlist->nri;
 444     iinr             = nlist->iinr;
 445     jindex           = nlist->jindex;
 446     jjnr             = nlist->jjnr;
 447     shiftidx         = nlist->shift;
 448     gid              = nlist->gid;
 449     shiftvec         = fr->shift_vec[0];
 450     fshift           = fr->fshift[0];
 451     facel            = _mm_set1_pd(fr->epsfac);
 452     charge           = mdatoms->chargeA;
 453     nvdwtype         = fr->ntype;
 454     vdwparam         = fr->nbfp;
 455     vdwtype          = mdatoms->typeA;
 456
 457     vftab            = kernel_data->table_vdw->data;
 458     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 459
 460     /* Avoid stupid compiler warnings */
 461     jnrA = jnrB = 0;
 462     j_coord_offsetA = 0;
 463     j_coord_offsetB = 0;
 464
 465     outeriter        = 0;
 466     inneriter        = 0;
 467
 468     /* Start outer loop over neighborlists */
 469     for(iidx=0; iidx<nri; iidx++)
 470     {
 471         /* Load shift vector for this list */
 472         i_shift_offset   = DIM*shiftidx[iidx];
 473
 474         /* Load limits for loop over neighbors */
 475         j_index_start    = jindex[iidx];
 476         j_index_end      = jindex[iidx+1];
 477
 478         /* Get outer coordinate index */
 479         inr              = iinr[iidx];
 480         i_coord_offset   = DIM*inr;
 481
 482         /* Load i particle coords and add shift vector */
 483         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 484
 485         fix0             = _mm_setzero_pd();
 486         fiy0             = _mm_setzero_pd();
 487         fiz0             = _mm_setzero_pd();
 488
 489         /* Load parameters for i particles */
 490         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 491         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 492
 493         /* Start inner kernel loop */
 494         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrA             = jjnr[jidx];
 499             jnrB             = jjnr[jidx+1];
 500             j_coord_offsetA  = DIM*jnrA;
 501             j_coord_offsetB  = DIM*jnrB;
 502
 503             /* load j atom coordinates */
 504             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 505                                               &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _mm_sub_pd(ix0,jx0);
 509             dy00             = _mm_sub_pd(iy0,jy0);
 510             dz00             = _mm_sub_pd(iz0,jz0);
 511
 512             /* Calculate squared distance and things based on it */
 513             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 514
 515             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 516
 517             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522             vdwjidx0B        = 2*vdwtype[jnrB+0];
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r00              = _mm_mul_pd(rsq00,rinv00);
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq00             = _mm_mul_pd(iq0,jq0);
 532             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 533                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 534
 535             /* Calculate table index by multiplying r with table scale and truncate to integer */
 536             rt               = _mm_mul_pd(r00,vftabscale);
 537             vfitab           = _mm_cvttpd_epi32(rt);
 538 #ifdef __XOP__
 539             vfeps            = _mm_frcz_pd(rt);
 540 #else
 541             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 542 #endif
 543             twovfeps         = _mm_add_pd(vfeps,vfeps);
 544             vfitab           = _mm_slli_epi32(vfitab,3);
 545
 546             /* COULOMB ELECTROSTATICS */
 547             velec            = _mm_mul_pd(qq00,rinv00);
 548             felec            = _mm_mul_pd(velec,rinvsq00);
 549
 550             /* CUBIC SPLINE TABLE DISPERSION */
 551             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 552             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 553             GMX_MM_TRANSPOSE2_PD(Y,F);
 554             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 555             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 556             GMX_MM_TRANSPOSE2_PD(G,H);
 557             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 558             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 559             fvdw6            = _mm_mul_pd(c6_00,FF);
 560
 561             /* CUBIC SPLINE TABLE REPULSION */
 562             vfitab           = _mm_add_epi32(vfitab,ifour);
 563             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 564             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 565             GMX_MM_TRANSPOSE2_PD(Y,F);
 566             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 567             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 568             GMX_MM_TRANSPOSE2_PD(G,H);
 569             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 570             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 571             fvdw12           = _mm_mul_pd(c12_00,FF);
 572             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 573
 574             fscal            = _mm_add_pd(felec,fvdw);
 575
 576             /* Update vectorial force */
 577             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 578             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 579             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 580
 581             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 582                                                    _mm_mul_pd(dx00,fscal),
 583                                                    _mm_mul_pd(dy00,fscal),
 584                                                    _mm_mul_pd(dz00,fscal));
 585
 586             /* Inner loop uses 57 flops */
 587         }
 588
 589         if(jidx<j_index_end)
 590         {
 591
 592             jnrA             = jjnr[jidx];
 593             j_coord_offsetA  = DIM*jnrA;
 594
 595             /* load j atom coordinates */
 596             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 597                                               &jx0,&jy0,&jz0);
 598
 599             /* Calculate displacement vector */
 600             dx00             = _mm_sub_pd(ix0,jx0);
 601             dy00             = _mm_sub_pd(iy0,jy0);
 602             dz00             = _mm_sub_pd(iz0,jz0);
 603
 604             /* Calculate squared distance and things based on it */
 605             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 606
 607             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 608
 609             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 610
 611             /* Load parameters for j particles */
 612             jq0              = _mm_load_sd(charge+jnrA+0);
 613             vdwjidx0A        = 2*vdwtype[jnrA+0];
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             r00              = _mm_mul_pd(rsq00,rinv00);
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq00             = _mm_mul_pd(iq0,jq0);
 623             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 624
 625             /* Calculate table index by multiplying r with table scale and truncate to integer */
 626             rt               = _mm_mul_pd(r00,vftabscale);
 627             vfitab           = _mm_cvttpd_epi32(rt);
 628 #ifdef __XOP__
 629             vfeps            = _mm_frcz_pd(rt);
 630 #else
 631             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 632 #endif
 633             twovfeps         = _mm_add_pd(vfeps,vfeps);
 634             vfitab           = _mm_slli_epi32(vfitab,3);
 635
 636             /* COULOMB ELECTROSTATICS */
 637             velec            = _mm_mul_pd(qq00,rinv00);
 638             felec            = _mm_mul_pd(velec,rinvsq00);
 639
 640             /* CUBIC SPLINE TABLE DISPERSION */
 641             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 642             F                = _mm_setzero_pd();
 643             GMX_MM_TRANSPOSE2_PD(Y,F);
 644             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 645             H                = _mm_setzero_pd();
 646             GMX_MM_TRANSPOSE2_PD(G,H);
 647             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 648             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 649             fvdw6            = _mm_mul_pd(c6_00,FF);
 650
 651             /* CUBIC SPLINE TABLE REPULSION */
 652             vfitab           = _mm_add_epi32(vfitab,ifour);
 653             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 654             F                = _mm_setzero_pd();
 655             GMX_MM_TRANSPOSE2_PD(Y,F);
 656             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 657             H                = _mm_setzero_pd();
 658             GMX_MM_TRANSPOSE2_PD(G,H);
 659             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 660             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 661             fvdw12           = _mm_mul_pd(c12_00,FF);
 662             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 663
 664             fscal            = _mm_add_pd(felec,fvdw);
 665
 666             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 667
 668             /* Update vectorial force */
 669             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 670             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 671             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 672
 673             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 674                                                    _mm_mul_pd(dx00,fscal),
 675                                                    _mm_mul_pd(dy00,fscal),
 676                                                    _mm_mul_pd(dz00,fscal));
 677
 678             /* Inner loop uses 57 flops */
 679         }
 680
 681         /* End of innermost loop */
 682
 683         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 684                                               f+i_coord_offset,fshift+i_shift_offset);
 685
 686         /* Increment number of inner iterations */
 687         inneriter                  += j_index_end - j_index_start;
 688
 689         /* Outer loop uses 7 flops */
 690     }
 691
 692     /* Increment number of outer iterations */
 693     outeriter        += nri;
 694
 695     /* Update outer/inner flops */
 696
 697     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 698 }