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, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: None
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     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     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 143
 144         fix0             = _mm_setzero_pd();
 145         fiy0             = _mm_setzero_pd();
 146         fiz0             = _mm_setzero_pd();
 147
 148         /* Load parameters for i particles */
 149         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151         /* Reset potential sums */
 152         vvdwsum          = _mm_setzero_pd();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163
 164             /* load j atom coordinates */
 165             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 166                                               &jx0,&jy0,&jz0);
 167
 168             /* Calculate displacement vector */
 169             dx00             = _mm_sub_pd(ix0,jx0);
 170             dy00             = _mm_sub_pd(iy0,jy0);
 171             dz00             = _mm_sub_pd(iz0,jz0);
 172
 173             /* Calculate squared distance and things based on it */
 174             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 175
 176             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 177
 178             /* Load parameters for j particles */
 179             vdwjidx0A        = 2*vdwtype[jnrA+0];
 180             vdwjidx0B        = 2*vdwtype[jnrB+0];
 181
 182             /**************************
 183              * CALCULATE INTERACTIONS *
 184              **************************/
 185
 186             r00              = _mm_mul_pd(rsq00,rinv00);
 187
 188             /* Compute parameters for interactions between i and j atoms */
 189             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 190                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 191
 192             /* Calculate table index by multiplying r with table scale and truncate to integer */
 193             rt               = _mm_mul_pd(r00,vftabscale);
 194             vfitab           = _mm_cvttpd_epi32(rt);
 195 #ifdef __XOP__
 196             vfeps            = _mm_frcz_pd(rt);
 197 #else
 198             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 199 #endif
 200             twovfeps         = _mm_add_pd(vfeps,vfeps);
 201             vfitab           = _mm_slli_epi32(vfitab,3);
 202
 203             /* CUBIC SPLINE TABLE DISPERSION */
 204             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 205             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 206             GMX_MM_TRANSPOSE2_PD(Y,F);
 207             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 208             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 209             GMX_MM_TRANSPOSE2_PD(G,H);
 210             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 211             VV               = _mm_macc_pd(vfeps,Fp,Y);
 212             vvdw6            = _mm_mul_pd(c6_00,VV);
 213             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 214             fvdw6            = _mm_mul_pd(c6_00,FF);
 215
 216             /* CUBIC SPLINE TABLE REPULSION */
 217             vfitab           = _mm_add_epi32(vfitab,ifour);
 218             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 219             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 220             GMX_MM_TRANSPOSE2_PD(Y,F);
 221             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 222             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 223             GMX_MM_TRANSPOSE2_PD(G,H);
 224             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 225             VV               = _mm_macc_pd(vfeps,Fp,Y);
 226             vvdw12           = _mm_mul_pd(c12_00,VV);
 227             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 228             fvdw12           = _mm_mul_pd(c12_00,FF);
 229             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 230             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 234
 235             fscal            = fvdw;
 236
 237             /* Update vectorial force */
 238             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 239             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 240             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 241
 242             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 243                                                    _mm_mul_pd(dx00,fscal),
 244                                                    _mm_mul_pd(dy00,fscal),
 245                                                    _mm_mul_pd(dz00,fscal));
 246
 247             /* Inner loop uses 59 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             jnrA             = jjnr[jidx];
 254             j_coord_offsetA  = DIM*jnrA;
 255
 256             /* load j atom coordinates */
 257             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_pd(ix0,jx0);
 262             dy00             = _mm_sub_pd(iy0,jy0);
 263             dz00             = _mm_sub_pd(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 267
 268             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 269
 270             /* Load parameters for j particles */
 271             vdwjidx0A        = 2*vdwtype[jnrA+0];
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             r00              = _mm_mul_pd(rsq00,rinv00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 281
 282             /* Calculate table index by multiplying r with table scale and truncate to integer */
 283             rt               = _mm_mul_pd(r00,vftabscale);
 284             vfitab           = _mm_cvttpd_epi32(rt);
 285 #ifdef __XOP__
 286             vfeps            = _mm_frcz_pd(rt);
 287 #else
 288             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 289 #endif
 290             twovfeps         = _mm_add_pd(vfeps,vfeps);
 291             vfitab           = _mm_slli_epi32(vfitab,3);
 292
 293             /* CUBIC SPLINE TABLE DISPERSION */
 294             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 295             F                = _mm_setzero_pd();
 296             GMX_MM_TRANSPOSE2_PD(Y,F);
 297             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 298             H                = _mm_setzero_pd();
 299             GMX_MM_TRANSPOSE2_PD(G,H);
 300             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 301             VV               = _mm_macc_pd(vfeps,Fp,Y);
 302             vvdw6            = _mm_mul_pd(c6_00,VV);
 303             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 304             fvdw6            = _mm_mul_pd(c6_00,FF);
 305
 306             /* CUBIC SPLINE TABLE REPULSION */
 307             vfitab           = _mm_add_epi32(vfitab,ifour);
 308             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 309             F                = _mm_setzero_pd();
 310             GMX_MM_TRANSPOSE2_PD(Y,F);
 311             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 312             H                = _mm_setzero_pd();
 313             GMX_MM_TRANSPOSE2_PD(G,H);
 314             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 315             VV               = _mm_macc_pd(vfeps,Fp,Y);
 316             vvdw12           = _mm_mul_pd(c12_00,VV);
 317             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 318             fvdw12           = _mm_mul_pd(c12_00,FF);
 319             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 320             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 324             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 325
 326             fscal            = fvdw;
 327
 328             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 329
 330             /* Update vectorial force */
 331             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 332             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 333             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 334
 335             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 336                                                    _mm_mul_pd(dx00,fscal),
 337                                                    _mm_mul_pd(dy00,fscal),
 338                                                    _mm_mul_pd(dz00,fscal));
 339
 340             /* Inner loop uses 59 flops */
 341         }
 342
 343         /* End of innermost loop */
 344
 345         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 346                                               f+i_coord_offset,fshift+i_shift_offset);
 347
 348         ggid                        = gid[iidx];
 349         /* Update potential energies */
 350         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 351
 352         /* Increment number of inner iterations */
 353         inneriter                  += j_index_end - j_index_start;
 354
 355         /* Outer loop uses 7 flops */
 356     }
 357
 358     /* Increment number of outer iterations */
 359     outeriter        += nri;
 360
 361     /* Update outer/inner flops */
 362
 363     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 364 }
 365 /*
 366  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 367  * Electrostatics interaction: None
 368  * VdW interaction:            CubicSplineTable
 369  * Geometry:                   Particle-Particle
 370  * Calculate force/pot:        Force
 371  */
 372 void
 373 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 374                     (t_nblist                    * gmx_restrict       nlist,
 375                      rvec                        * gmx_restrict          xx,
 376                      rvec                        * gmx_restrict          ff,
 377                      t_forcerec                  * gmx_restrict          fr,
 378                      t_mdatoms                   * gmx_restrict     mdatoms,
 379                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 380                      t_nrnb                      * gmx_restrict        nrnb)
 381 {
 382     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 383      * just 0 for non-waters.
 384      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 385      * jnr indices corresponding to data put in the four positions in the SIMD register.
 386      */
 387     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 388     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 389     int              jnrA,jnrB;
 390     int              j_coord_offsetA,j_coord_offsetB;
 391     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 392     real             rcutoff_scalar;
 393     real             *shiftvec,*fshift,*x,*f;
 394     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 395     int              vdwioffset0;
 396     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 397     int              vdwjidx0A,vdwjidx0B;
 398     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 399     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 400     int              nvdwtype;
 401     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 402     int              *vdwtype;
 403     real             *vdwparam;
 404     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 405     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 406     __m128i          vfitab;
 407     __m128i          ifour       = _mm_set1_epi32(4);
 408     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 409     real             *vftab;
 410     __m128d          dummy_mask,cutoff_mask;
 411     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 412     __m128d          one     = _mm_set1_pd(1.0);
 413     __m128d          two     = _mm_set1_pd(2.0);
 414     x                = xx[0];
 415     f                = ff[0];
 416
 417     nri              = nlist->nri;
 418     iinr             = nlist->iinr;
 419     jindex           = nlist->jindex;
 420     jjnr             = nlist->jjnr;
 421     shiftidx         = nlist->shift;
 422     gid              = nlist->gid;
 423     shiftvec         = fr->shift_vec[0];
 424     fshift           = fr->fshift[0];
 425     nvdwtype         = fr->ntype;
 426     vdwparam         = fr->nbfp;
 427     vdwtype          = mdatoms->typeA;
 428
 429     vftab            = kernel_data->table_vdw->data;
 430     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 431
 432     /* Avoid stupid compiler warnings */
 433     jnrA = jnrB = 0;
 434     j_coord_offsetA = 0;
 435     j_coord_offsetB = 0;
 436
 437     outeriter        = 0;
 438     inneriter        = 0;
 439
 440     /* Start outer loop over neighborlists */
 441     for(iidx=0; iidx<nri; iidx++)
 442     {
 443         /* Load shift vector for this list */
 444         i_shift_offset   = DIM*shiftidx[iidx];
 445
 446         /* Load limits for loop over neighbors */
 447         j_index_start    = jindex[iidx];
 448         j_index_end      = jindex[iidx+1];
 449
 450         /* Get outer coordinate index */
 451         inr              = iinr[iidx];
 452         i_coord_offset   = DIM*inr;
 453
 454         /* Load i particle coords and add shift vector */
 455         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 456
 457         fix0             = _mm_setzero_pd();
 458         fiy0             = _mm_setzero_pd();
 459         fiz0             = _mm_setzero_pd();
 460
 461         /* Load parameters for i particles */
 462         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 463
 464         /* Start inner kernel loop */
 465         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 466         {
 467
 468             /* Get j neighbor index, and coordinate index */
 469             jnrA             = jjnr[jidx];
 470             jnrB             = jjnr[jidx+1];
 471             j_coord_offsetA  = DIM*jnrA;
 472             j_coord_offsetB  = DIM*jnrB;
 473
 474             /* load j atom coordinates */
 475             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 476                                               &jx0,&jy0,&jz0);
 477
 478             /* Calculate displacement vector */
 479             dx00             = _mm_sub_pd(ix0,jx0);
 480             dy00             = _mm_sub_pd(iy0,jy0);
 481             dz00             = _mm_sub_pd(iz0,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 485
 486             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 487
 488             /* Load parameters for j particles */
 489             vdwjidx0A        = 2*vdwtype[jnrA+0];
 490             vdwjidx0B        = 2*vdwtype[jnrB+0];
 491
 492             /**************************
 493              * CALCULATE INTERACTIONS *
 494              **************************/
 495
 496             r00              = _mm_mul_pd(rsq00,rinv00);
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 500                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 501
 502             /* Calculate table index by multiplying r with table scale and truncate to integer */
 503             rt               = _mm_mul_pd(r00,vftabscale);
 504             vfitab           = _mm_cvttpd_epi32(rt);
 505 #ifdef __XOP__
 506             vfeps            = _mm_frcz_pd(rt);
 507 #else
 508             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 509 #endif
 510             twovfeps         = _mm_add_pd(vfeps,vfeps);
 511             vfitab           = _mm_slli_epi32(vfitab,3);
 512
 513             /* CUBIC SPLINE TABLE DISPERSION */
 514             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 515             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 516             GMX_MM_TRANSPOSE2_PD(Y,F);
 517             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 518             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 519             GMX_MM_TRANSPOSE2_PD(G,H);
 520             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 521             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 522             fvdw6            = _mm_mul_pd(c6_00,FF);
 523
 524             /* CUBIC SPLINE TABLE REPULSION */
 525             vfitab           = _mm_add_epi32(vfitab,ifour);
 526             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 527             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 528             GMX_MM_TRANSPOSE2_PD(Y,F);
 529             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 530             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 531             GMX_MM_TRANSPOSE2_PD(G,H);
 532             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 533             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 534             fvdw12           = _mm_mul_pd(c12_00,FF);
 535             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 536
 537             fscal            = fvdw;
 538
 539             /* Update vectorial force */
 540             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 541             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 542             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 543
 544             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 545                                                    _mm_mul_pd(dx00,fscal),
 546                                                    _mm_mul_pd(dy00,fscal),
 547                                                    _mm_mul_pd(dz00,fscal));
 548
 549             /* Inner loop uses 51 flops */
 550         }
 551
 552         if(jidx<j_index_end)
 553         {
 554
 555             jnrA             = jjnr[jidx];
 556             j_coord_offsetA  = DIM*jnrA;
 557
 558             /* load j atom coordinates */
 559             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 560                                               &jx0,&jy0,&jz0);
 561
 562             /* Calculate displacement vector */
 563             dx00             = _mm_sub_pd(ix0,jx0);
 564             dy00             = _mm_sub_pd(iy0,jy0);
 565             dz00             = _mm_sub_pd(iz0,jz0);
 566
 567             /* Calculate squared distance and things based on it */
 568             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 569
 570             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 571
 572             /* Load parameters for j particles */
 573             vdwjidx0A        = 2*vdwtype[jnrA+0];
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r00              = _mm_mul_pd(rsq00,rinv00);
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 583
 584             /* Calculate table index by multiplying r with table scale and truncate to integer */
 585             rt               = _mm_mul_pd(r00,vftabscale);
 586             vfitab           = _mm_cvttpd_epi32(rt);
 587 #ifdef __XOP__
 588             vfeps            = _mm_frcz_pd(rt);
 589 #else
 590             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 591 #endif
 592             twovfeps         = _mm_add_pd(vfeps,vfeps);
 593             vfitab           = _mm_slli_epi32(vfitab,3);
 594
 595             /* CUBIC SPLINE TABLE DISPERSION */
 596             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 597             F                = _mm_setzero_pd();
 598             GMX_MM_TRANSPOSE2_PD(Y,F);
 599             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 600             H                = _mm_setzero_pd();
 601             GMX_MM_TRANSPOSE2_PD(G,H);
 602             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 603             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 604             fvdw6            = _mm_mul_pd(c6_00,FF);
 605
 606             /* CUBIC SPLINE TABLE REPULSION */
 607             vfitab           = _mm_add_epi32(vfitab,ifour);
 608             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 609             F                = _mm_setzero_pd();
 610             GMX_MM_TRANSPOSE2_PD(Y,F);
 611             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 612             H                = _mm_setzero_pd();
 613             GMX_MM_TRANSPOSE2_PD(G,H);
 614             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 615             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 616             fvdw12           = _mm_mul_pd(c12_00,FF);
 617             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 618
 619             fscal            = fvdw;
 620
 621             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 622
 623             /* Update vectorial force */
 624             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 625             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 626             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 627
 628             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 629                                                    _mm_mul_pd(dx00,fscal),
 630                                                    _mm_mul_pd(dy00,fscal),
 631                                                    _mm_mul_pd(dz00,fscal));
 632
 633             /* Inner loop uses 51 flops */
 634         }
 635
 636         /* End of innermost loop */
 637
 638         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 639                                               f+i_coord_offset,fshift+i_shift_offset);
 640
 641         /* Increment number of inner iterations */
 642         inneriter                  += j_index_end - j_index_start;
 643
 644         /* Outer loop uses 6 flops */
 645     }
 646
 647     /* Increment number of outer iterations */
 648     outeriter        += nri;
 649
 650     /* Update outer/inner flops */
 651
 652     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*51);
 653 }