src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_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_ElecNone_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: None
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecNone_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     int              nvdwtype;
  86     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  90     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  94     real             *vftab;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     vftab            = kernel_data->table_vdw->data;
 115     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _mm_setzero_pd();
 143         fiy0             = _mm_setzero_pd();
 144         fiz0             = _mm_setzero_pd();
 145
 146         /* Load parameters for i particles */
 147         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149         /* Reset potential sums */
 150         vvdwsum          = _mm_setzero_pd();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161
 162             /* load j atom coordinates */
 163             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 164                                               &jx0,&jy0,&jz0);
 165
 166             /* Calculate displacement vector */
 167             dx00             = _mm_sub_pd(ix0,jx0);
 168             dy00             = _mm_sub_pd(iy0,jy0);
 169             dz00             = _mm_sub_pd(iz0,jz0);
 170
 171             /* Calculate squared distance and things based on it */
 172             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 173
 174             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 175
 176             /* Load parameters for j particles */
 177             vdwjidx0A        = 2*vdwtype[jnrA+0];
 178             vdwjidx0B        = 2*vdwtype[jnrB+0];
 179
 180             /**************************
 181              * CALCULATE INTERACTIONS *
 182              **************************/
 183
 184             r00              = _mm_mul_pd(rsq00,rinv00);
 185
 186             /* Compute parameters for interactions between i and j atoms */
 187             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 188                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 189
 190             /* Calculate table index by multiplying r with table scale and truncate to integer */
 191             rt               = _mm_mul_pd(r00,vftabscale);
 192             vfitab           = _mm_cvttpd_epi32(rt);
 193 #ifdef __XOP__
 194             vfeps            = _mm_frcz_pd(rt);
 195 #else
 196             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 197 #endif
 198             twovfeps         = _mm_add_pd(vfeps,vfeps);
 199             vfitab           = _mm_slli_epi32(vfitab,3);
 200
 201             /* CUBIC SPLINE TABLE DISPERSION */
 202             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 203             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 204             GMX_MM_TRANSPOSE2_PD(Y,F);
 205             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 206             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 207             GMX_MM_TRANSPOSE2_PD(G,H);
 208             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 209             VV               = _mm_macc_pd(vfeps,Fp,Y);
 210             vvdw6            = _mm_mul_pd(c6_00,VV);
 211             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 212             fvdw6            = _mm_mul_pd(c6_00,FF);
 213
 214             /* CUBIC SPLINE TABLE REPULSION */
 215             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm_mul_pd(c12_00,VV);
 225             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 226             fvdw12           = _mm_mul_pd(c12_00,FF);
 227             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 228             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 229
 230             /* Update potential sum for this i atom from the interaction with this j atom. */
 231             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 232
 233             fscal            = fvdw;
 234
 235             /* Update vectorial force */
 236             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 237             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 238             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 239
 240             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 241                                                    _mm_mul_pd(dx00,fscal),
 242                                                    _mm_mul_pd(dy00,fscal),
 243                                                    _mm_mul_pd(dz00,fscal));
 244
 245             /* Inner loop uses 59 flops */
 246         }
 247
 248         if(jidx<j_index_end)
 249         {
 250
 251             jnrA             = jjnr[jidx];
 252             j_coord_offsetA  = DIM*jnrA;
 253
 254             /* load j atom coordinates */
 255             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 256                                               &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm_sub_pd(ix0,jx0);
 260             dy00             = _mm_sub_pd(iy0,jy0);
 261             dz00             = _mm_sub_pd(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 265
 266             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 267
 268             /* Load parameters for j particles */
 269             vdwjidx0A        = 2*vdwtype[jnrA+0];
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             r00              = _mm_mul_pd(rsq00,rinv00);
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 279
 280             /* Calculate table index by multiplying r with table scale and truncate to integer */
 281             rt               = _mm_mul_pd(r00,vftabscale);
 282             vfitab           = _mm_cvttpd_epi32(rt);
 283 #ifdef __XOP__
 284             vfeps            = _mm_frcz_pd(rt);
 285 #else
 286             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 287 #endif
 288             twovfeps         = _mm_add_pd(vfeps,vfeps);
 289             vfitab           = _mm_slli_epi32(vfitab,3);
 290
 291             /* CUBIC SPLINE TABLE DISPERSION */
 292             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 293             F                = _mm_setzero_pd();
 294             GMX_MM_TRANSPOSE2_PD(Y,F);
 295             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 296             H                = _mm_setzero_pd();
 297             GMX_MM_TRANSPOSE2_PD(G,H);
 298             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 299             VV               = _mm_macc_pd(vfeps,Fp,Y);
 300             vvdw6            = _mm_mul_pd(c6_00,VV);
 301             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 302             fvdw6            = _mm_mul_pd(c6_00,FF);
 303
 304             /* CUBIC SPLINE TABLE REPULSION */
 305             vfitab           = _mm_add_epi32(vfitab,ifour);
 306             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 307             F                = _mm_setzero_pd();
 308             GMX_MM_TRANSPOSE2_PD(Y,F);
 309             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 310             H                = _mm_setzero_pd();
 311             GMX_MM_TRANSPOSE2_PD(G,H);
 312             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 313             VV               = _mm_macc_pd(vfeps,Fp,Y);
 314             vvdw12           = _mm_mul_pd(c12_00,VV);
 315             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 316             fvdw12           = _mm_mul_pd(c12_00,FF);
 317             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 318             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 322             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 323
 324             fscal            = fvdw;
 325
 326             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 327
 328             /* Update vectorial force */
 329             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 330             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 331             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 332
 333             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 334                                                    _mm_mul_pd(dx00,fscal),
 335                                                    _mm_mul_pd(dy00,fscal),
 336                                                    _mm_mul_pd(dz00,fscal));
 337
 338             /* Inner loop uses 59 flops */
 339         }
 340
 341         /* End of innermost loop */
 342
 343         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 344                                               f+i_coord_offset,fshift+i_shift_offset);
 345
 346         ggid                        = gid[iidx];
 347         /* Update potential energies */
 348         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 349
 350         /* Increment number of inner iterations */
 351         inneriter                  += j_index_end - j_index_start;
 352
 353         /* Outer loop uses 7 flops */
 354     }
 355
 356     /* Increment number of outer iterations */
 357     outeriter        += nri;
 358
 359     /* Update outer/inner flops */
 360
 361     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 362 }
 363 /*
 364  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 365  * Electrostatics interaction: None
 366  * VdW interaction:            CubicSplineTable
 367  * Geometry:                   Particle-Particle
 368  * Calculate force/pot:        Force
 369  */
 370 void
 371 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 372                     (t_nblist                    * gmx_restrict       nlist,
 373                      rvec                        * gmx_restrict          xx,
 374                      rvec                        * gmx_restrict          ff,
 375                      t_forcerec                  * gmx_restrict          fr,
 376                      t_mdatoms                   * gmx_restrict     mdatoms,
 377                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 378                      t_nrnb                      * gmx_restrict        nrnb)
 379 {
 380     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 381      * just 0 for non-waters.
 382      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 383      * jnr indices corresponding to data put in the four positions in the SIMD register.
 384      */
 385     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 386     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 387     int              jnrA,jnrB;
 388     int              j_coord_offsetA,j_coord_offsetB;
 389     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 390     real             rcutoff_scalar;
 391     real             *shiftvec,*fshift,*x,*f;
 392     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 393     int              vdwioffset0;
 394     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 395     int              vdwjidx0A,vdwjidx0B;
 396     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 397     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 398     int              nvdwtype;
 399     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 400     int              *vdwtype;
 401     real             *vdwparam;
 402     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 403     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 404     __m128i          vfitab;
 405     __m128i          ifour       = _mm_set1_epi32(4);
 406     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 407     real             *vftab;
 408     __m128d          dummy_mask,cutoff_mask;
 409     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 410     __m128d          one     = _mm_set1_pd(1.0);
 411     __m128d          two     = _mm_set1_pd(2.0);
 412     x                = xx[0];
 413     f                = ff[0];
 414
 415     nri              = nlist->nri;
 416     iinr             = nlist->iinr;
 417     jindex           = nlist->jindex;
 418     jjnr             = nlist->jjnr;
 419     shiftidx         = nlist->shift;
 420     gid              = nlist->gid;
 421     shiftvec         = fr->shift_vec[0];
 422     fshift           = fr->fshift[0];
 423     nvdwtype         = fr->ntype;
 424     vdwparam         = fr->nbfp;
 425     vdwtype          = mdatoms->typeA;
 426
 427     vftab            = kernel_data->table_vdw->data;
 428     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 429
 430     /* Avoid stupid compiler warnings */
 431     jnrA = jnrB = 0;
 432     j_coord_offsetA = 0;
 433     j_coord_offsetB = 0;
 434
 435     outeriter        = 0;
 436     inneriter        = 0;
 437
 438     /* Start outer loop over neighborlists */
 439     for(iidx=0; iidx<nri; iidx++)
 440     {
 441         /* Load shift vector for this list */
 442         i_shift_offset   = DIM*shiftidx[iidx];
 443
 444         /* Load limits for loop over neighbors */
 445         j_index_start    = jindex[iidx];
 446         j_index_end      = jindex[iidx+1];
 447
 448         /* Get outer coordinate index */
 449         inr              = iinr[iidx];
 450         i_coord_offset   = DIM*inr;
 451
 452         /* Load i particle coords and add shift vector */
 453         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 454
 455         fix0             = _mm_setzero_pd();
 456         fiy0             = _mm_setzero_pd();
 457         fiz0             = _mm_setzero_pd();
 458
 459         /* Load parameters for i particles */
 460         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 461
 462         /* Start inner kernel loop */
 463         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 464         {
 465
 466             /* Get j neighbor index, and coordinate index */
 467             jnrA             = jjnr[jidx];
 468             jnrB             = jjnr[jidx+1];
 469             j_coord_offsetA  = DIM*jnrA;
 470             j_coord_offsetB  = DIM*jnrB;
 471
 472             /* load j atom coordinates */
 473             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 474                                               &jx0,&jy0,&jz0);
 475
 476             /* Calculate displacement vector */
 477             dx00             = _mm_sub_pd(ix0,jx0);
 478             dy00             = _mm_sub_pd(iy0,jy0);
 479             dz00             = _mm_sub_pd(iz0,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 483
 484             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 485
 486             /* Load parameters for j particles */
 487             vdwjidx0A        = 2*vdwtype[jnrA+0];
 488             vdwjidx0B        = 2*vdwtype[jnrB+0];
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             r00              = _mm_mul_pd(rsq00,rinv00);
 495
 496             /* Compute parameters for interactions between i and j atoms */
 497             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 498                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 499
 500             /* Calculate table index by multiplying r with table scale and truncate to integer */
 501             rt               = _mm_mul_pd(r00,vftabscale);
 502             vfitab           = _mm_cvttpd_epi32(rt);
 503 #ifdef __XOP__
 504             vfeps            = _mm_frcz_pd(rt);
 505 #else
 506             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 507 #endif
 508             twovfeps         = _mm_add_pd(vfeps,vfeps);
 509             vfitab           = _mm_slli_epi32(vfitab,3);
 510
 511             /* CUBIC SPLINE TABLE DISPERSION */
 512             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 513             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 514             GMX_MM_TRANSPOSE2_PD(Y,F);
 515             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 516             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 517             GMX_MM_TRANSPOSE2_PD(G,H);
 518             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 519             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 520             fvdw6            = _mm_mul_pd(c6_00,FF);
 521
 522             /* CUBIC SPLINE TABLE REPULSION */
 523             vfitab           = _mm_add_epi32(vfitab,ifour);
 524             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 525             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 526             GMX_MM_TRANSPOSE2_PD(Y,F);
 527             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 528             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 529             GMX_MM_TRANSPOSE2_PD(G,H);
 530             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 531             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 532             fvdw12           = _mm_mul_pd(c12_00,FF);
 533             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 534
 535             fscal            = fvdw;
 536
 537             /* Update vectorial force */
 538             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 539             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 540             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 541
 542             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 543                                                    _mm_mul_pd(dx00,fscal),
 544                                                    _mm_mul_pd(dy00,fscal),
 545                                                    _mm_mul_pd(dz00,fscal));
 546
 547             /* Inner loop uses 51 flops */
 548         }
 549
 550         if(jidx<j_index_end)
 551         {
 552
 553             jnrA             = jjnr[jidx];
 554             j_coord_offsetA  = DIM*jnrA;
 555
 556             /* load j atom coordinates */
 557             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 558                                               &jx0,&jy0,&jz0);
 559
 560             /* Calculate displacement vector */
 561             dx00             = _mm_sub_pd(ix0,jx0);
 562             dy00             = _mm_sub_pd(iy0,jy0);
 563             dz00             = _mm_sub_pd(iz0,jz0);
 564
 565             /* Calculate squared distance and things based on it */
 566             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 567
 568             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 569
 570             /* Load parameters for j particles */
 571             vdwjidx0A        = 2*vdwtype[jnrA+0];
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r00              = _mm_mul_pd(rsq00,rinv00);
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 581
 582             /* Calculate table index by multiplying r with table scale and truncate to integer */
 583             rt               = _mm_mul_pd(r00,vftabscale);
 584             vfitab           = _mm_cvttpd_epi32(rt);
 585 #ifdef __XOP__
 586             vfeps            = _mm_frcz_pd(rt);
 587 #else
 588             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 589 #endif
 590             twovfeps         = _mm_add_pd(vfeps,vfeps);
 591             vfitab           = _mm_slli_epi32(vfitab,3);
 592
 593             /* CUBIC SPLINE TABLE DISPERSION */
 594             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 595             F                = _mm_setzero_pd();
 596             GMX_MM_TRANSPOSE2_PD(Y,F);
 597             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 598             H                = _mm_setzero_pd();
 599             GMX_MM_TRANSPOSE2_PD(G,H);
 600             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 601             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 602             fvdw6            = _mm_mul_pd(c6_00,FF);
 603
 604             /* CUBIC SPLINE TABLE REPULSION */
 605             vfitab           = _mm_add_epi32(vfitab,ifour);
 606             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 607             F                = _mm_setzero_pd();
 608             GMX_MM_TRANSPOSE2_PD(Y,F);
 609             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 610             H                = _mm_setzero_pd();
 611             GMX_MM_TRANSPOSE2_PD(G,H);
 612             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 613             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 614             fvdw12           = _mm_mul_pd(c12_00,FF);
 615             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 616
 617             fscal            = fvdw;
 618
 619             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 620
 621             /* Update vectorial force */
 622             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 623             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 624             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 625
 626             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 627                                                    _mm_mul_pd(dx00,fscal),
 628                                                    _mm_mul_pd(dy00,fscal),
 629                                                    _mm_mul_pd(dz00,fscal));
 630
 631             /* Inner loop uses 51 flops */
 632         }
 633
 634         /* End of innermost loop */
 635
 636         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 637                                               f+i_coord_offset,fshift+i_shift_offset);
 638
 639         /* Increment number of inner iterations */
 640         inneriter                  += j_index_end - j_index_start;
 641
 642         /* Outer loop uses 6 flops */
 643     }
 644
 645     /* Increment number of outer iterations */
 646     outeriter        += nri;
 647
 648     /* Update outer/inner flops */
 649
 650     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*51);
 651 }