src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecNone_VdwLJ_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: None
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     int              nvdwtype;
  75     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  76     int              *vdwtype;
  77     real             *vdwparam;
  78     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  79     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  80     __m128           dummy_mask,cutoff_mask;
  81     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  82     __m128           one     = _mm_set1_ps(1.0);
  83     __m128           two     = _mm_set1_ps(2.0);
  84     x                = xx[0];
  85     f                = ff[0];
  86
  87     nri              = nlist->nri;
  88     iinr             = nlist->iinr;
  89     jindex           = nlist->jindex;
  90     jjnr             = nlist->jjnr;
  91     shiftidx         = nlist->shift;
  92     gid              = nlist->gid;
  93     shiftvec         = fr->shift_vec[0];
  94     fshift           = fr->fshift[0];
  95     nvdwtype         = fr->ntype;
  96     vdwparam         = fr->nbfp;
  97     vdwtype          = mdatoms->typeA;
  98
  99     /* Avoid stupid compiler warnings */
 100     jnrA = jnrB = jnrC = jnrD = 0;
 101     j_coord_offsetA = 0;
 102     j_coord_offsetB = 0;
 103     j_coord_offsetC = 0;
 104     j_coord_offsetD = 0;
 105
 106     outeriter        = 0;
 107     inneriter        = 0;
 108
 109     for(iidx=0;iidx<4*DIM;iidx++)
 110     {
 111         scratch[iidx] = 0.0;
 112     }
 113
 114     /* Start outer loop over neighborlists */
 115     for(iidx=0; iidx<nri; iidx++)
 116     {
 117         /* Load shift vector for this list */
 118         i_shift_offset   = DIM*shiftidx[iidx];
 119
 120         /* Load limits for loop over neighbors */
 121         j_index_start    = jindex[iidx];
 122         j_index_end      = jindex[iidx+1];
 123
 124         /* Get outer coordinate index */
 125         inr              = iinr[iidx];
 126         i_coord_offset   = DIM*inr;
 127
 128         /* Load i particle coords and add shift vector */
 129         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 130
 131         fix0             = _mm_setzero_ps();
 132         fiy0             = _mm_setzero_ps();
 133         fiz0             = _mm_setzero_ps();
 134
 135         /* Load parameters for i particles */
 136         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138         /* Reset potential sums */
 139         vvdwsum          = _mm_setzero_ps();
 140
 141         /* Start inner kernel loop */
 142         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 143         {
 144
 145             /* Get j neighbor index, and coordinate index */
 146             jnrA             = jjnr[jidx];
 147             jnrB             = jjnr[jidx+1];
 148             jnrC             = jjnr[jidx+2];
 149             jnrD             = jjnr[jidx+3];
 150             j_coord_offsetA  = DIM*jnrA;
 151             j_coord_offsetB  = DIM*jnrB;
 152             j_coord_offsetC  = DIM*jnrC;
 153             j_coord_offsetD  = DIM*jnrD;
 154
 155             /* load j atom coordinates */
 156             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 157                                               x+j_coord_offsetC,x+j_coord_offsetD,
 158                                               &jx0,&jy0,&jz0);
 159
 160             /* Calculate displacement vector */
 161             dx00             = _mm_sub_ps(ix0,jx0);
 162             dy00             = _mm_sub_ps(iy0,jy0);
 163             dz00             = _mm_sub_ps(iz0,jz0);
 164
 165             /* Calculate squared distance and things based on it */
 166             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 167
 168             rinvsq00         = gmx_mm_inv_ps(rsq00);
 169
 170             /* Load parameters for j particles */
 171             vdwjidx0A        = 2*vdwtype[jnrA+0];
 172             vdwjidx0B        = 2*vdwtype[jnrB+0];
 173             vdwjidx0C        = 2*vdwtype[jnrC+0];
 174             vdwjidx0D        = 2*vdwtype[jnrD+0];
 175
 176             /**************************
 177              * CALCULATE INTERACTIONS *
 178              **************************/
 179
 180             /* Compute parameters for interactions between i and j atoms */
 181             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 182                                          vdwparam+vdwioffset0+vdwjidx0B,
 183                                          vdwparam+vdwioffset0+vdwjidx0C,
 184                                          vdwparam+vdwioffset0+vdwjidx0D,
 185                                          &c6_00,&c12_00);
 186
 187             /* LENNARD-JONES DISPERSION/REPULSION */
 188
 189             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 190             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 191             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 192             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 193             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 194
 195             /* Update potential sum for this i atom from the interaction with this j atom. */
 196             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 197
 198             fscal            = fvdw;
 199
 200              /* Update vectorial force */
 201             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 202             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 203             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 204
 205             fjptrA             = f+j_coord_offsetA;
 206             fjptrB             = f+j_coord_offsetB;
 207             fjptrC             = f+j_coord_offsetC;
 208             fjptrD             = f+j_coord_offsetD;
 209             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 210                                                    _mm_mul_ps(dx00,fscal),
 211                                                    _mm_mul_ps(dy00,fscal),
 212                                                    _mm_mul_ps(dz00,fscal));
 213
 214             /* Inner loop uses 35 flops */
 215         }
 216
 217         if(jidx<j_index_end)
 218         {
 219
 220             /* Get j neighbor index, and coordinate index */
 221             jnrlistA         = jjnr[jidx];
 222             jnrlistB         = jjnr[jidx+1];
 223             jnrlistC         = jjnr[jidx+2];
 224             jnrlistD         = jjnr[jidx+3];
 225             /* Sign of each element will be negative for non-real atoms.
 226              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 227              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 228              */
 229             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 230             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 231             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 232             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 233             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 234             j_coord_offsetA  = DIM*jnrA;
 235             j_coord_offsetB  = DIM*jnrB;
 236             j_coord_offsetC  = DIM*jnrC;
 237             j_coord_offsetD  = DIM*jnrD;
 238
 239             /* load j atom coordinates */
 240             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 241                                               x+j_coord_offsetC,x+j_coord_offsetD,
 242                                               &jx0,&jy0,&jz0);
 243
 244             /* Calculate displacement vector */
 245             dx00             = _mm_sub_ps(ix0,jx0);
 246             dy00             = _mm_sub_ps(iy0,jy0);
 247             dz00             = _mm_sub_ps(iz0,jz0);
 248
 249             /* Calculate squared distance and things based on it */
 250             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 251
 252             rinvsq00         = gmx_mm_inv_ps(rsq00);
 253
 254             /* Load parameters for j particles */
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256             vdwjidx0B        = 2*vdwtype[jnrB+0];
 257             vdwjidx0C        = 2*vdwtype[jnrC+0];
 258             vdwjidx0D        = 2*vdwtype[jnrD+0];
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 266                                          vdwparam+vdwioffset0+vdwjidx0B,
 267                                          vdwparam+vdwioffset0+vdwjidx0C,
 268                                          vdwparam+vdwioffset0+vdwjidx0D,
 269                                          &c6_00,&c12_00);
 270
 271             /* LENNARD-JONES DISPERSION/REPULSION */
 272
 273             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 274             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 275             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 276             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 277             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 281             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 282
 283             fscal            = fvdw;
 284
 285             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 286
 287              /* Update vectorial force */
 288             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 289             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 290             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 291
 292             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 293             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 294             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 295             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 296             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 297                                                    _mm_mul_ps(dx00,fscal),
 298                                                    _mm_mul_ps(dy00,fscal),
 299                                                    _mm_mul_ps(dz00,fscal));
 300
 301             /* Inner loop uses 35 flops */
 302         }
 303
 304         /* End of innermost loop */
 305
 306         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 307                                               f+i_coord_offset,fshift+i_shift_offset);
 308
 309         ggid                        = gid[iidx];
 310         /* Update potential energies */
 311         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 312
 313         /* Increment number of inner iterations */
 314         inneriter                  += j_index_end - j_index_start;
 315
 316         /* Outer loop uses 7 flops */
 317     }
 318
 319     /* Increment number of outer iterations */
 320     outeriter        += nri;
 321
 322     /* Update outer/inner flops */
 323
 324     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*35);
 325 }
 326 /*
 327  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_single
 328  * Electrostatics interaction: None
 329  * VdW interaction:            LennardJones
 330  * Geometry:                   Particle-Particle
 331  * Calculate force/pot:        Force
 332  */
 333 void
 334 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_single
 335                     (t_nblist * gmx_restrict                nlist,
 336                      rvec * gmx_restrict                    xx,
 337                      rvec * gmx_restrict                    ff,
 338                      t_forcerec * gmx_restrict              fr,
 339                      t_mdatoms * gmx_restrict               mdatoms,
 340                      nb_kernel_data_t * gmx_restrict        kernel_data,
 341                      t_nrnb * gmx_restrict                  nrnb)
 342 {
 343     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 344      * just 0 for non-waters.
 345      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 346      * jnr indices corresponding to data put in the four positions in the SIMD register.
 347      */
 348     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 349     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 350     int              jnrA,jnrB,jnrC,jnrD;
 351     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 352     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 353     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 354     real             rcutoff_scalar;
 355     real             *shiftvec,*fshift,*x,*f;
 356     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 357     real             scratch[4*DIM];
 358     __m128           fscal,rcutoff,rcutoff2,jidxall;
 359     int              vdwioffset0;
 360     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 361     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 362     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 363     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 364     int              nvdwtype;
 365     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 366     int              *vdwtype;
 367     real             *vdwparam;
 368     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 369     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 370     __m128           dummy_mask,cutoff_mask;
 371     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 372     __m128           one     = _mm_set1_ps(1.0);
 373     __m128           two     = _mm_set1_ps(2.0);
 374     x                = xx[0];
 375     f                = ff[0];
 376
 377     nri              = nlist->nri;
 378     iinr             = nlist->iinr;
 379     jindex           = nlist->jindex;
 380     jjnr             = nlist->jjnr;
 381     shiftidx         = nlist->shift;
 382     gid              = nlist->gid;
 383     shiftvec         = fr->shift_vec[0];
 384     fshift           = fr->fshift[0];
 385     nvdwtype         = fr->ntype;
 386     vdwparam         = fr->nbfp;
 387     vdwtype          = mdatoms->typeA;
 388
 389     /* Avoid stupid compiler warnings */
 390     jnrA = jnrB = jnrC = jnrD = 0;
 391     j_coord_offsetA = 0;
 392     j_coord_offsetB = 0;
 393     j_coord_offsetC = 0;
 394     j_coord_offsetD = 0;
 395
 396     outeriter        = 0;
 397     inneriter        = 0;
 398
 399     for(iidx=0;iidx<4*DIM;iidx++)
 400     {
 401         scratch[iidx] = 0.0;
 402     }
 403
 404     /* Start outer loop over neighborlists */
 405     for(iidx=0; iidx<nri; iidx++)
 406     {
 407         /* Load shift vector for this list */
 408         i_shift_offset   = DIM*shiftidx[iidx];
 409
 410         /* Load limits for loop over neighbors */
 411         j_index_start    = jindex[iidx];
 412         j_index_end      = jindex[iidx+1];
 413
 414         /* Get outer coordinate index */
 415         inr              = iinr[iidx];
 416         i_coord_offset   = DIM*inr;
 417
 418         /* Load i particle coords and add shift vector */
 419         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 420
 421         fix0             = _mm_setzero_ps();
 422         fiy0             = _mm_setzero_ps();
 423         fiz0             = _mm_setzero_ps();
 424
 425         /* Load parameters for i particles */
 426         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 427
 428         /* Start inner kernel loop */
 429         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 430         {
 431
 432             /* Get j neighbor index, and coordinate index */
 433             jnrA             = jjnr[jidx];
 434             jnrB             = jjnr[jidx+1];
 435             jnrC             = jjnr[jidx+2];
 436             jnrD             = jjnr[jidx+3];
 437             j_coord_offsetA  = DIM*jnrA;
 438             j_coord_offsetB  = DIM*jnrB;
 439             j_coord_offsetC  = DIM*jnrC;
 440             j_coord_offsetD  = DIM*jnrD;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                               x+j_coord_offsetC,x+j_coord_offsetD,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_ps(ix0,jx0);
 449             dy00             = _mm_sub_ps(iy0,jy0);
 450             dz00             = _mm_sub_ps(iz0,jz0);
 451
 452             /* Calculate squared distance and things based on it */
 453             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 454
 455             rinvsq00         = gmx_mm_inv_ps(rsq00);
 456
 457             /* Load parameters for j particles */
 458             vdwjidx0A        = 2*vdwtype[jnrA+0];
 459             vdwjidx0B        = 2*vdwtype[jnrB+0];
 460             vdwjidx0C        = 2*vdwtype[jnrC+0];
 461             vdwjidx0D        = 2*vdwtype[jnrD+0];
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 469                                          vdwparam+vdwioffset0+vdwjidx0B,
 470                                          vdwparam+vdwioffset0+vdwjidx0C,
 471                                          vdwparam+vdwioffset0+vdwjidx0D,
 472                                          &c6_00,&c12_00);
 473
 474             /* LENNARD-JONES DISPERSION/REPULSION */
 475
 476             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 477             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 478
 479             fscal            = fvdw;
 480
 481              /* Update vectorial force */
 482             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 483             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 484             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 485
 486             fjptrA             = f+j_coord_offsetA;
 487             fjptrB             = f+j_coord_offsetB;
 488             fjptrC             = f+j_coord_offsetC;
 489             fjptrD             = f+j_coord_offsetD;
 490             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 491                                                    _mm_mul_ps(dx00,fscal),
 492                                                    _mm_mul_ps(dy00,fscal),
 493                                                    _mm_mul_ps(dz00,fscal));
 494
 495             /* Inner loop uses 30 flops */
 496         }
 497
 498         if(jidx<j_index_end)
 499         {
 500
 501             /* Get j neighbor index, and coordinate index */
 502             jnrlistA         = jjnr[jidx];
 503             jnrlistB         = jjnr[jidx+1];
 504             jnrlistC         = jjnr[jidx+2];
 505             jnrlistD         = jjnr[jidx+3];
 506             /* Sign of each element will be negative for non-real atoms.
 507              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 508              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 509              */
 510             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 511             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 512             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 513             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 514             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 515             j_coord_offsetA  = DIM*jnrA;
 516             j_coord_offsetB  = DIM*jnrB;
 517             j_coord_offsetC  = DIM*jnrC;
 518             j_coord_offsetD  = DIM*jnrD;
 519
 520             /* load j atom coordinates */
 521             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 522                                               x+j_coord_offsetC,x+j_coord_offsetD,
 523                                               &jx0,&jy0,&jz0);
 524
 525             /* Calculate displacement vector */
 526             dx00             = _mm_sub_ps(ix0,jx0);
 527             dy00             = _mm_sub_ps(iy0,jy0);
 528             dz00             = _mm_sub_ps(iz0,jz0);
 529
 530             /* Calculate squared distance and things based on it */
 531             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 532
 533             rinvsq00         = gmx_mm_inv_ps(rsq00);
 534
 535             /* Load parameters for j particles */
 536             vdwjidx0A        = 2*vdwtype[jnrA+0];
 537             vdwjidx0B        = 2*vdwtype[jnrB+0];
 538             vdwjidx0C        = 2*vdwtype[jnrC+0];
 539             vdwjidx0D        = 2*vdwtype[jnrD+0];
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 547                                          vdwparam+vdwioffset0+vdwjidx0B,
 548                                          vdwparam+vdwioffset0+vdwjidx0C,
 549                                          vdwparam+vdwioffset0+vdwjidx0D,
 550                                          &c6_00,&c12_00);
 551
 552             /* LENNARD-JONES DISPERSION/REPULSION */
 553
 554             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 555             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 556
 557             fscal            = fvdw;
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561              /* Update vectorial force */
 562             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 563             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 564             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 565
 566             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 567             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 568             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 569             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 570             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 571                                                    _mm_mul_ps(dx00,fscal),
 572                                                    _mm_mul_ps(dy00,fscal),
 573                                                    _mm_mul_ps(dz00,fscal));
 574
 575             /* Inner loop uses 30 flops */
 576         }
 577
 578         /* End of innermost loop */
 579
 580         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 581                                               f+i_coord_offset,fshift+i_shift_offset);
 582
 583         /* Increment number of inner iterations */
 584         inneriter                  += j_index_end - j_index_start;
 585
 586         /* Outer loop uses 6 flops */
 587     }
 588
 589     /* Increment number of outer iterations */
 590     outeriter        += nri;
 591
 592     /* Update outer/inner flops */
 593
 594     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 595 }