src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_VdwCSTab_GeomP1P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: None
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     int              nvdwtype;
  85     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  86     int              *vdwtype;
  87     real             *vdwparam;
  88     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  89     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  90     __m128i          vfitab;
  91     __m128i          ifour       = _mm_set1_epi32(4);
  92     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  93     real             *vftab;
  94     __m128d          dummy_mask,cutoff_mask;
  95     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  96     __m128d          one     = _mm_set1_pd(1.0);
  97     __m128d          two     = _mm_set1_pd(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 140
 141         fix0             = _mm_setzero_pd();
 142         fiy0             = _mm_setzero_pd();
 143         fiz0             = _mm_setzero_pd();
 144
 145         /* Load parameters for i particles */
 146         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148         /* Reset potential sums */
 149         vvdwsum          = _mm_setzero_pd();
 150
 151         /* Start inner kernel loop */
 152         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 153         {
 154
 155             /* Get j neighbor index, and coordinate index */
 156             jnrA             = jjnr[jidx];
 157             jnrB             = jjnr[jidx+1];
 158             j_coord_offsetA  = DIM*jnrA;
 159             j_coord_offsetB  = DIM*jnrB;
 160
 161             /* load j atom coordinates */
 162             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 163                                               &jx0,&jy0,&jz0);
 164
 165             /* Calculate displacement vector */
 166             dx00             = _mm_sub_pd(ix0,jx0);
 167             dy00             = _mm_sub_pd(iy0,jy0);
 168             dz00             = _mm_sub_pd(iz0,jz0);
 169
 170             /* Calculate squared distance and things based on it */
 171             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 172
 173             rinv00           = avx128fma_invsqrt_d(rsq00);
 174
 175             /* Load parameters for j particles */
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             r00              = _mm_mul_pd(rsq00,rinv00);
 184
 185             /* Compute parameters for interactions between i and j atoms */
 186             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 187                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 188
 189             /* Calculate table index by multiplying r with table scale and truncate to integer */
 190             rt               = _mm_mul_pd(r00,vftabscale);
 191             vfitab           = _mm_cvttpd_epi32(rt);
 192 #ifdef __XOP__
 193             vfeps            = _mm_frcz_pd(rt);
 194 #else
 195             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 196 #endif
 197             twovfeps         = _mm_add_pd(vfeps,vfeps);
 198             vfitab           = _mm_slli_epi32(vfitab,3);
 199
 200             /* CUBIC SPLINE TABLE DISPERSION */
 201             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 202             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 203             GMX_MM_TRANSPOSE2_PD(Y,F);
 204             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 205             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 206             GMX_MM_TRANSPOSE2_PD(G,H);
 207             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 208             VV               = _mm_macc_pd(vfeps,Fp,Y);
 209             vvdw6            = _mm_mul_pd(c6_00,VV);
 210             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 211             fvdw6            = _mm_mul_pd(c6_00,FF);
 212
 213             /* CUBIC SPLINE TABLE REPULSION */
 214             vfitab           = _mm_add_epi32(vfitab,ifour);
 215             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 216             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 217             GMX_MM_TRANSPOSE2_PD(Y,F);
 218             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 219             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 220             GMX_MM_TRANSPOSE2_PD(G,H);
 221             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 222             VV               = _mm_macc_pd(vfeps,Fp,Y);
 223             vvdw12           = _mm_mul_pd(c12_00,VV);
 224             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 225             fvdw12           = _mm_mul_pd(c12_00,FF);
 226             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 227             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 231
 232             fscal            = fvdw;
 233
 234             /* Update vectorial force */
 235             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 236             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 237             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 238
 239             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 240                                                    _mm_mul_pd(dx00,fscal),
 241                                                    _mm_mul_pd(dy00,fscal),
 242                                                    _mm_mul_pd(dz00,fscal));
 243
 244             /* Inner loop uses 59 flops */
 245         }
 246
 247         if(jidx<j_index_end)
 248         {
 249
 250             jnrA             = jjnr[jidx];
 251             j_coord_offsetA  = DIM*jnrA;
 252
 253             /* load j atom coordinates */
 254             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 255                                               &jx0,&jy0,&jz0);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _mm_sub_pd(ix0,jx0);
 259             dy00             = _mm_sub_pd(iy0,jy0);
 260             dz00             = _mm_sub_pd(iz0,jz0);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 264
 265             rinv00           = avx128fma_invsqrt_d(rsq00);
 266
 267             /* Load parameters for j particles */
 268             vdwjidx0A        = 2*vdwtype[jnrA+0];
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r00              = _mm_mul_pd(rsq00,rinv00);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 278
 279             /* Calculate table index by multiplying r with table scale and truncate to integer */
 280             rt               = _mm_mul_pd(r00,vftabscale);
 281             vfitab           = _mm_cvttpd_epi32(rt);
 282 #ifdef __XOP__
 283             vfeps            = _mm_frcz_pd(rt);
 284 #else
 285             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 286 #endif
 287             twovfeps         = _mm_add_pd(vfeps,vfeps);
 288             vfitab           = _mm_slli_epi32(vfitab,3);
 289
 290             /* CUBIC SPLINE TABLE DISPERSION */
 291             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 292             F                = _mm_setzero_pd();
 293             GMX_MM_TRANSPOSE2_PD(Y,F);
 294             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 295             H                = _mm_setzero_pd();
 296             GMX_MM_TRANSPOSE2_PD(G,H);
 297             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 298             VV               = _mm_macc_pd(vfeps,Fp,Y);
 299             vvdw6            = _mm_mul_pd(c6_00,VV);
 300             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 301             fvdw6            = _mm_mul_pd(c6_00,FF);
 302
 303             /* CUBIC SPLINE TABLE REPULSION */
 304             vfitab           = _mm_add_epi32(vfitab,ifour);
 305             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 306             F                = _mm_setzero_pd();
 307             GMX_MM_TRANSPOSE2_PD(Y,F);
 308             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 309             H                = _mm_setzero_pd();
 310             GMX_MM_TRANSPOSE2_PD(G,H);
 311             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 312             VV               = _mm_macc_pd(vfeps,Fp,Y);
 313             vvdw12           = _mm_mul_pd(c12_00,VV);
 314             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 315             fvdw12           = _mm_mul_pd(c12_00,FF);
 316             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 317             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 321             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 322
 323             fscal            = fvdw;
 324
 325             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 326
 327             /* Update vectorial force */
 328             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 329             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 330             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 331
 332             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 333                                                    _mm_mul_pd(dx00,fscal),
 334                                                    _mm_mul_pd(dy00,fscal),
 335                                                    _mm_mul_pd(dz00,fscal));
 336
 337             /* Inner loop uses 59 flops */
 338         }
 339
 340         /* End of innermost loop */
 341
 342         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 343                                               f+i_coord_offset,fshift+i_shift_offset);
 344
 345         ggid                        = gid[iidx];
 346         /* Update potential energies */
 347         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 348
 349         /* Increment number of inner iterations */
 350         inneriter                  += j_index_end - j_index_start;
 351
 352         /* Outer loop uses 7 flops */
 353     }
 354
 355     /* Increment number of outer iterations */
 356     outeriter        += nri;
 357
 358     /* Update outer/inner flops */
 359
 360     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 361 }
 362 /*
 363  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 364  * Electrostatics interaction: None
 365  * VdW interaction:            CubicSplineTable
 366  * Geometry:                   Particle-Particle
 367  * Calculate force/pot:        Force
 368  */
 369 void
 370 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 371                     (t_nblist                    * gmx_restrict       nlist,
 372                      rvec                        * gmx_restrict          xx,
 373                      rvec                        * gmx_restrict          ff,
 374                      struct t_forcerec           * gmx_restrict          fr,
 375                      t_mdatoms                   * gmx_restrict     mdatoms,
 376                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 377                      t_nrnb                      * gmx_restrict        nrnb)
 378 {
 379     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 380      * just 0 for non-waters.
 381      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 382      * jnr indices corresponding to data put in the four positions in the SIMD register.
 383      */
 384     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 385     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 386     int              jnrA,jnrB;
 387     int              j_coord_offsetA,j_coord_offsetB;
 388     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 389     real             rcutoff_scalar;
 390     real             *shiftvec,*fshift,*x,*f;
 391     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 392     int              vdwioffset0;
 393     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 394     int              vdwjidx0A,vdwjidx0B;
 395     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 396     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 397     int              nvdwtype;
 398     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 399     int              *vdwtype;
 400     real             *vdwparam;
 401     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 402     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 403     __m128i          vfitab;
 404     __m128i          ifour       = _mm_set1_epi32(4);
 405     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 406     real             *vftab;
 407     __m128d          dummy_mask,cutoff_mask;
 408     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 409     __m128d          one     = _mm_set1_pd(1.0);
 410     __m128d          two     = _mm_set1_pd(2.0);
 411     x                = xx[0];
 412     f                = ff[0];
 413
 414     nri              = nlist->nri;
 415     iinr             = nlist->iinr;
 416     jindex           = nlist->jindex;
 417     jjnr             = nlist->jjnr;
 418     shiftidx         = nlist->shift;
 419     gid              = nlist->gid;
 420     shiftvec         = fr->shift_vec[0];
 421     fshift           = fr->fshift[0];
 422     nvdwtype         = fr->ntype;
 423     vdwparam         = fr->nbfp;
 424     vdwtype          = mdatoms->typeA;
 425
 426     vftab            = kernel_data->table_vdw->data;
 427     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 428
 429     /* Avoid stupid compiler warnings */
 430     jnrA = jnrB = 0;
 431     j_coord_offsetA = 0;
 432     j_coord_offsetB = 0;
 433
 434     outeriter        = 0;
 435     inneriter        = 0;
 436
 437     /* Start outer loop over neighborlists */
 438     for(iidx=0; iidx<nri; iidx++)
 439     {
 440         /* Load shift vector for this list */
 441         i_shift_offset   = DIM*shiftidx[iidx];
 442
 443         /* Load limits for loop over neighbors */
 444         j_index_start    = jindex[iidx];
 445         j_index_end      = jindex[iidx+1];
 446
 447         /* Get outer coordinate index */
 448         inr              = iinr[iidx];
 449         i_coord_offset   = DIM*inr;
 450
 451         /* Load i particle coords and add shift vector */
 452         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 453
 454         fix0             = _mm_setzero_pd();
 455         fiy0             = _mm_setzero_pd();
 456         fiz0             = _mm_setzero_pd();
 457
 458         /* Load parameters for i particles */
 459         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 460
 461         /* Start inner kernel loop */
 462         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 463         {
 464
 465             /* Get j neighbor index, and coordinate index */
 466             jnrA             = jjnr[jidx];
 467             jnrB             = jjnr[jidx+1];
 468             j_coord_offsetA  = DIM*jnrA;
 469             j_coord_offsetB  = DIM*jnrB;
 470
 471             /* load j atom coordinates */
 472             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 473                                               &jx0,&jy0,&jz0);
 474
 475             /* Calculate displacement vector */
 476             dx00             = _mm_sub_pd(ix0,jx0);
 477             dy00             = _mm_sub_pd(iy0,jy0);
 478             dz00             = _mm_sub_pd(iz0,jz0);
 479
 480             /* Calculate squared distance and things based on it */
 481             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 482
 483             rinv00           = avx128fma_invsqrt_d(rsq00);
 484
 485             /* Load parameters for j particles */
 486             vdwjidx0A        = 2*vdwtype[jnrA+0];
 487             vdwjidx0B        = 2*vdwtype[jnrB+0];
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             r00              = _mm_mul_pd(rsq00,rinv00);
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 497                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 498
 499             /* Calculate table index by multiplying r with table scale and truncate to integer */
 500             rt               = _mm_mul_pd(r00,vftabscale);
 501             vfitab           = _mm_cvttpd_epi32(rt);
 502 #ifdef __XOP__
 503             vfeps            = _mm_frcz_pd(rt);
 504 #else
 505             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 506 #endif
 507             twovfeps         = _mm_add_pd(vfeps,vfeps);
 508             vfitab           = _mm_slli_epi32(vfitab,3);
 509
 510             /* CUBIC SPLINE TABLE DISPERSION */
 511             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 512             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 513             GMX_MM_TRANSPOSE2_PD(Y,F);
 514             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 515             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 516             GMX_MM_TRANSPOSE2_PD(G,H);
 517             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 518             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 519             fvdw6            = _mm_mul_pd(c6_00,FF);
 520
 521             /* CUBIC SPLINE TABLE REPULSION */
 522             vfitab           = _mm_add_epi32(vfitab,ifour);
 523             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 524             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 525             GMX_MM_TRANSPOSE2_PD(Y,F);
 526             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 527             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 528             GMX_MM_TRANSPOSE2_PD(G,H);
 529             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 530             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 531             fvdw12           = _mm_mul_pd(c12_00,FF);
 532             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 533
 534             fscal            = fvdw;
 535
 536             /* Update vectorial force */
 537             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 538             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 539             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 540
 541             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 542                                                    _mm_mul_pd(dx00,fscal),
 543                                                    _mm_mul_pd(dy00,fscal),
 544                                                    _mm_mul_pd(dz00,fscal));
 545
 546             /* Inner loop uses 51 flops */
 547         }
 548
 549         if(jidx<j_index_end)
 550         {
 551
 552             jnrA             = jjnr[jidx];
 553             j_coord_offsetA  = DIM*jnrA;
 554
 555             /* load j atom coordinates */
 556             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 557                                               &jx0,&jy0,&jz0);
 558
 559             /* Calculate displacement vector */
 560             dx00             = _mm_sub_pd(ix0,jx0);
 561             dy00             = _mm_sub_pd(iy0,jy0);
 562             dz00             = _mm_sub_pd(iz0,jz0);
 563
 564             /* Calculate squared distance and things based on it */
 565             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 566
 567             rinv00           = avx128fma_invsqrt_d(rsq00);
 568
 569             /* Load parameters for j particles */
 570             vdwjidx0A        = 2*vdwtype[jnrA+0];
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             r00              = _mm_mul_pd(rsq00,rinv00);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 580
 581             /* Calculate table index by multiplying r with table scale and truncate to integer */
 582             rt               = _mm_mul_pd(r00,vftabscale);
 583             vfitab           = _mm_cvttpd_epi32(rt);
 584 #ifdef __XOP__
 585             vfeps            = _mm_frcz_pd(rt);
 586 #else
 587             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 588 #endif
 589             twovfeps         = _mm_add_pd(vfeps,vfeps);
 590             vfitab           = _mm_slli_epi32(vfitab,3);
 591
 592             /* CUBIC SPLINE TABLE DISPERSION */
 593             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 594             F                = _mm_setzero_pd();
 595             GMX_MM_TRANSPOSE2_PD(Y,F);
 596             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 597             H                = _mm_setzero_pd();
 598             GMX_MM_TRANSPOSE2_PD(G,H);
 599             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 600             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 601             fvdw6            = _mm_mul_pd(c6_00,FF);
 602
 603             /* CUBIC SPLINE TABLE REPULSION */
 604             vfitab           = _mm_add_epi32(vfitab,ifour);
 605             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 606             F                = _mm_setzero_pd();
 607             GMX_MM_TRANSPOSE2_PD(Y,F);
 608             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 609             H                = _mm_setzero_pd();
 610             GMX_MM_TRANSPOSE2_PD(G,H);
 611             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 612             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 613             fvdw12           = _mm_mul_pd(c12_00,FF);
 614             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 615
 616             fscal            = fvdw;
 617
 618             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 619
 620             /* Update vectorial force */
 621             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 622             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 623             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 624
 625             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 626                                                    _mm_mul_pd(dx00,fscal),
 627                                                    _mm_mul_pd(dy00,fscal),
 628                                                    _mm_mul_pd(dz00,fscal));
 629
 630             /* Inner loop uses 51 flops */
 631         }
 632
 633         /* End of innermost loop */
 634
 635         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 636                                               f+i_coord_offset,fshift+i_shift_offset);
 637
 638         /* Increment number of inner iterations */
 639         inneriter                  += j_index_end - j_index_start;
 640
 641         /* Outer loop uses 6 flops */
 642     }
 643
 644     /* Increment number of outer iterations */
 645     outeriter        += nri;
 646
 647     /* Update outer/inner flops */
 648
 649     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*51);
 650 }