src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecNone_VdwLJSw_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_VdwLJSw_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_VdwLJSw_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           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  81     real             rswitch_scalar,d_scalar;
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     nvdwtype         = fr->ntype;
  98     vdwparam         = fr->nbfp;
  99     vdwtype          = mdatoms->typeA;
 100
 101     rcutoff_scalar   = fr->rvdw;
 102     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 103     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 104
 105     rswitch_scalar   = fr->rvdw_switch;
 106     rswitch          = _mm_set1_ps(rswitch_scalar);
 107     /* Setup switch parameters */
 108     d_scalar         = rcutoff_scalar-rswitch_scalar;
 109     d                = _mm_set1_ps(d_scalar);
 110     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 111     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 112     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 113     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 114     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 115     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     for(iidx=0;iidx<4*DIM;iidx++)
 128     {
 129         scratch[iidx] = 0.0;
 130     }
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_ps();
 150         fiy0             = _mm_setzero_ps();
 151         fiz0             = _mm_setzero_ps();
 152
 153         /* Load parameters for i particles */
 154         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 155
 156         /* Reset potential sums */
 157         vvdwsum          = _mm_setzero_ps();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170             j_coord_offsetC  = DIM*jnrC;
 171             j_coord_offsetD  = DIM*jnrD;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               x+j_coord_offsetC,x+j_coord_offsetD,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _mm_sub_ps(ix0,jx0);
 180             dy00             = _mm_sub_ps(iy0,jy0);
 181             dz00             = _mm_sub_ps(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 187
 188             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 189
 190             /* Load parameters for j particles */
 191             vdwjidx0A        = 2*vdwtype[jnrA+0];
 192             vdwjidx0B        = 2*vdwtype[jnrB+0];
 193             vdwjidx0C        = 2*vdwtype[jnrC+0];
 194             vdwjidx0D        = 2*vdwtype[jnrD+0];
 195
 196             /**************************
 197              * CALCULATE INTERACTIONS *
 198              **************************/
 199
 200             if (gmx_mm_any_lt(rsq00,rcutoff2))
 201             {
 202
 203             r00              = _mm_mul_ps(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,
 208                                          vdwparam+vdwioffset0+vdwjidx0C,
 209                                          vdwparam+vdwioffset0+vdwjidx0D,
 210                                          &c6_00,&c12_00);
 211
 212             /* LENNARD-JONES DISPERSION/REPULSION */
 213
 214             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 215             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 216             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 217             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 218             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 219
 220             d                = _mm_sub_ps(r00,rswitch);
 221             d                = _mm_max_ps(d,_mm_setzero_ps());
 222             d2               = _mm_mul_ps(d,d);
 223             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 224
 225             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 226
 227             /* Evaluate switch function */
 228             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 229             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 230             vvdw             = _mm_mul_ps(vvdw,sw);
 231             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 235             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 236
 237             fscal            = fvdw;
 238
 239             fscal            = _mm_and_ps(fscal,cutoff_mask);
 240
 241              /* Update vectorial force */
 242             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 243             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 244             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 245
 246             fjptrA             = f+j_coord_offsetA;
 247             fjptrB             = f+j_coord_offsetB;
 248             fjptrC             = f+j_coord_offsetC;
 249             fjptrD             = f+j_coord_offsetD;
 250             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 251                                                    _mm_mul_ps(dx00,fscal),
 252                                                    _mm_mul_ps(dy00,fscal),
 253                                                    _mm_mul_ps(dz00,fscal));
 254
 255             }
 256
 257             /* Inner loop uses 62 flops */
 258         }
 259
 260         if(jidx<j_index_end)
 261         {
 262
 263             /* Get j neighbor index, and coordinate index */
 264             jnrlistA         = jjnr[jidx];
 265             jnrlistB         = jjnr[jidx+1];
 266             jnrlistC         = jjnr[jidx+2];
 267             jnrlistD         = jjnr[jidx+3];
 268             /* Sign of each element will be negative for non-real atoms.
 269              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 270              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 271              */
 272             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 273             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 274             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 275             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 276             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 277             j_coord_offsetA  = DIM*jnrA;
 278             j_coord_offsetB  = DIM*jnrB;
 279             j_coord_offsetC  = DIM*jnrC;
 280             j_coord_offsetD  = DIM*jnrD;
 281
 282             /* load j atom coordinates */
 283             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 284                                               x+j_coord_offsetC,x+j_coord_offsetD,
 285                                               &jx0,&jy0,&jz0);
 286
 287             /* Calculate displacement vector */
 288             dx00             = _mm_sub_ps(ix0,jx0);
 289             dy00             = _mm_sub_ps(iy0,jy0);
 290             dz00             = _mm_sub_ps(iz0,jz0);
 291
 292             /* Calculate squared distance and things based on it */
 293             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 294
 295             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 296
 297             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 298
 299             /* Load parameters for j particles */
 300             vdwjidx0A        = 2*vdwtype[jnrA+0];
 301             vdwjidx0B        = 2*vdwtype[jnrB+0];
 302             vdwjidx0C        = 2*vdwtype[jnrC+0];
 303             vdwjidx0D        = 2*vdwtype[jnrD+0];
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm_any_lt(rsq00,rcutoff2))
 310             {
 311
 312             r00              = _mm_mul_ps(rsq00,rinv00);
 313             r00              = _mm_andnot_ps(dummy_mask,r00);
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 317                                          vdwparam+vdwioffset0+vdwjidx0B,
 318                                          vdwparam+vdwioffset0+vdwjidx0C,
 319                                          vdwparam+vdwioffset0+vdwjidx0D,
 320                                          &c6_00,&c12_00);
 321
 322             /* LENNARD-JONES DISPERSION/REPULSION */
 323
 324             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 325             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 326             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 327             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 328             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 329
 330             d                = _mm_sub_ps(r00,rswitch);
 331             d                = _mm_max_ps(d,_mm_setzero_ps());
 332             d2               = _mm_mul_ps(d,d);
 333             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 334
 335             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 336
 337             /* Evaluate switch function */
 338             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 339             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 340             vvdw             = _mm_mul_ps(vvdw,sw);
 341             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 345             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 346             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 347
 348             fscal            = fvdw;
 349
 350             fscal            = _mm_and_ps(fscal,cutoff_mask);
 351
 352             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 353
 354              /* Update vectorial force */
 355             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 356             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 357             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 358
 359             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 360             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 361             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 362             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 363             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 364                                                    _mm_mul_ps(dx00,fscal),
 365                                                    _mm_mul_ps(dy00,fscal),
 366                                                    _mm_mul_ps(dz00,fscal));
 367
 368             }
 369
 370             /* Inner loop uses 63 flops */
 371         }
 372
 373         /* End of innermost loop */
 374
 375         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 376                                               f+i_coord_offset,fshift+i_shift_offset);
 377
 378         ggid                        = gid[iidx];
 379         /* Update potential energies */
 380         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 381
 382         /* Increment number of inner iterations */
 383         inneriter                  += j_index_end - j_index_start;
 384
 385         /* Outer loop uses 7 flops */
 386     }
 387
 388     /* Increment number of outer iterations */
 389     outeriter        += nri;
 390
 391     /* Update outer/inner flops */
 392
 393     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*63);
 394 }
 395 /*
 396  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_single
 397  * Electrostatics interaction: None
 398  * VdW interaction:            LennardJones
 399  * Geometry:                   Particle-Particle
 400  * Calculate force/pot:        Force
 401  */
 402 void
 403 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_single
 404                     (t_nblist * gmx_restrict                nlist,
 405                      rvec * gmx_restrict                    xx,
 406                      rvec * gmx_restrict                    ff,
 407                      t_forcerec * gmx_restrict              fr,
 408                      t_mdatoms * gmx_restrict               mdatoms,
 409                      nb_kernel_data_t * gmx_restrict        kernel_data,
 410                      t_nrnb * gmx_restrict                  nrnb)
 411 {
 412     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 413      * just 0 for non-waters.
 414      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 415      * jnr indices corresponding to data put in the four positions in the SIMD register.
 416      */
 417     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 418     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 419     int              jnrA,jnrB,jnrC,jnrD;
 420     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 421     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 422     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 423     real             rcutoff_scalar;
 424     real             *shiftvec,*fshift,*x,*f;
 425     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 426     real             scratch[4*DIM];
 427     __m128           fscal,rcutoff,rcutoff2,jidxall;
 428     int              vdwioffset0;
 429     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 430     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 431     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 432     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 433     int              nvdwtype;
 434     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 435     int              *vdwtype;
 436     real             *vdwparam;
 437     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 438     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 439     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 440     real             rswitch_scalar,d_scalar;
 441     __m128           dummy_mask,cutoff_mask;
 442     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 443     __m128           one     = _mm_set1_ps(1.0);
 444     __m128           two     = _mm_set1_ps(2.0);
 445     x                = xx[0];
 446     f                = ff[0];
 447
 448     nri              = nlist->nri;
 449     iinr             = nlist->iinr;
 450     jindex           = nlist->jindex;
 451     jjnr             = nlist->jjnr;
 452     shiftidx         = nlist->shift;
 453     gid              = nlist->gid;
 454     shiftvec         = fr->shift_vec[0];
 455     fshift           = fr->fshift[0];
 456     nvdwtype         = fr->ntype;
 457     vdwparam         = fr->nbfp;
 458     vdwtype          = mdatoms->typeA;
 459
 460     rcutoff_scalar   = fr->rvdw;
 461     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 462     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 463
 464     rswitch_scalar   = fr->rvdw_switch;
 465     rswitch          = _mm_set1_ps(rswitch_scalar);
 466     /* Setup switch parameters */
 467     d_scalar         = rcutoff_scalar-rswitch_scalar;
 468     d                = _mm_set1_ps(d_scalar);
 469     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 470     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 471     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 472     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 473     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 474     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 475
 476     /* Avoid stupid compiler warnings */
 477     jnrA = jnrB = jnrC = jnrD = 0;
 478     j_coord_offsetA = 0;
 479     j_coord_offsetB = 0;
 480     j_coord_offsetC = 0;
 481     j_coord_offsetD = 0;
 482
 483     outeriter        = 0;
 484     inneriter        = 0;
 485
 486     for(iidx=0;iidx<4*DIM;iidx++)
 487     {
 488         scratch[iidx] = 0.0;
 489     }
 490
 491     /* Start outer loop over neighborlists */
 492     for(iidx=0; iidx<nri; iidx++)
 493     {
 494         /* Load shift vector for this list */
 495         i_shift_offset   = DIM*shiftidx[iidx];
 496
 497         /* Load limits for loop over neighbors */
 498         j_index_start    = jindex[iidx];
 499         j_index_end      = jindex[iidx+1];
 500
 501         /* Get outer coordinate index */
 502         inr              = iinr[iidx];
 503         i_coord_offset   = DIM*inr;
 504
 505         /* Load i particle coords and add shift vector */
 506         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 507
 508         fix0             = _mm_setzero_ps();
 509         fiy0             = _mm_setzero_ps();
 510         fiz0             = _mm_setzero_ps();
 511
 512         /* Load parameters for i particles */
 513         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 514
 515         /* Start inner kernel loop */
 516         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 517         {
 518
 519             /* Get j neighbor index, and coordinate index */
 520             jnrA             = jjnr[jidx];
 521             jnrB             = jjnr[jidx+1];
 522             jnrC             = jjnr[jidx+2];
 523             jnrD             = jjnr[jidx+3];
 524             j_coord_offsetA  = DIM*jnrA;
 525             j_coord_offsetB  = DIM*jnrB;
 526             j_coord_offsetC  = DIM*jnrC;
 527             j_coord_offsetD  = DIM*jnrD;
 528
 529             /* load j atom coordinates */
 530             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 531                                               x+j_coord_offsetC,x+j_coord_offsetD,
 532                                               &jx0,&jy0,&jz0);
 533
 534             /* Calculate displacement vector */
 535             dx00             = _mm_sub_ps(ix0,jx0);
 536             dy00             = _mm_sub_ps(iy0,jy0);
 537             dz00             = _mm_sub_ps(iz0,jz0);
 538
 539             /* Calculate squared distance and things based on it */
 540             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 541
 542             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 543
 544             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 545
 546             /* Load parameters for j particles */
 547             vdwjidx0A        = 2*vdwtype[jnrA+0];
 548             vdwjidx0B        = 2*vdwtype[jnrB+0];
 549             vdwjidx0C        = 2*vdwtype[jnrC+0];
 550             vdwjidx0D        = 2*vdwtype[jnrD+0];
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             if (gmx_mm_any_lt(rsq00,rcutoff2))
 557             {
 558
 559             r00              = _mm_mul_ps(rsq00,rinv00);
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 563                                          vdwparam+vdwioffset0+vdwjidx0B,
 564                                          vdwparam+vdwioffset0+vdwjidx0C,
 565                                          vdwparam+vdwioffset0+vdwjidx0D,
 566                                          &c6_00,&c12_00);
 567
 568             /* LENNARD-JONES DISPERSION/REPULSION */
 569
 570             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 571             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 572             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 573             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 574             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 575
 576             d                = _mm_sub_ps(r00,rswitch);
 577             d                = _mm_max_ps(d,_mm_setzero_ps());
 578             d2               = _mm_mul_ps(d,d);
 579             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 580
 581             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 582
 583             /* Evaluate switch function */
 584             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 585             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 586             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 587
 588             fscal            = fvdw;
 589
 590             fscal            = _mm_and_ps(fscal,cutoff_mask);
 591
 592              /* Update vectorial force */
 593             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 594             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 595             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 596
 597             fjptrA             = f+j_coord_offsetA;
 598             fjptrB             = f+j_coord_offsetB;
 599             fjptrC             = f+j_coord_offsetC;
 600             fjptrD             = f+j_coord_offsetD;
 601             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 602                                                    _mm_mul_ps(dx00,fscal),
 603                                                    _mm_mul_ps(dy00,fscal),
 604                                                    _mm_mul_ps(dz00,fscal));
 605
 606             }
 607
 608             /* Inner loop uses 59 flops */
 609         }
 610
 611         if(jidx<j_index_end)
 612         {
 613
 614             /* Get j neighbor index, and coordinate index */
 615             jnrlistA         = jjnr[jidx];
 616             jnrlistB         = jjnr[jidx+1];
 617             jnrlistC         = jjnr[jidx+2];
 618             jnrlistD         = jjnr[jidx+3];
 619             /* Sign of each element will be negative for non-real atoms.
 620              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 621              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 622              */
 623             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 624             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 625             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 626             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 627             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 628             j_coord_offsetA  = DIM*jnrA;
 629             j_coord_offsetB  = DIM*jnrB;
 630             j_coord_offsetC  = DIM*jnrC;
 631             j_coord_offsetD  = DIM*jnrD;
 632
 633             /* load j atom coordinates */
 634             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 635                                               x+j_coord_offsetC,x+j_coord_offsetD,
 636                                               &jx0,&jy0,&jz0);
 637
 638             /* Calculate displacement vector */
 639             dx00             = _mm_sub_ps(ix0,jx0);
 640             dy00             = _mm_sub_ps(iy0,jy0);
 641             dz00             = _mm_sub_ps(iz0,jz0);
 642
 643             /* Calculate squared distance and things based on it */
 644             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 645
 646             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 647
 648             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 649
 650             /* Load parameters for j particles */
 651             vdwjidx0A        = 2*vdwtype[jnrA+0];
 652             vdwjidx0B        = 2*vdwtype[jnrB+0];
 653             vdwjidx0C        = 2*vdwtype[jnrC+0];
 654             vdwjidx0D        = 2*vdwtype[jnrD+0];
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             if (gmx_mm_any_lt(rsq00,rcutoff2))
 661             {
 662
 663             r00              = _mm_mul_ps(rsq00,rinv00);
 664             r00              = _mm_andnot_ps(dummy_mask,r00);
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 668                                          vdwparam+vdwioffset0+vdwjidx0B,
 669                                          vdwparam+vdwioffset0+vdwjidx0C,
 670                                          vdwparam+vdwioffset0+vdwjidx0D,
 671                                          &c6_00,&c12_00);
 672
 673             /* LENNARD-JONES DISPERSION/REPULSION */
 674
 675             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 676             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 677             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 678             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 679             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 680
 681             d                = _mm_sub_ps(r00,rswitch);
 682             d                = _mm_max_ps(d,_mm_setzero_ps());
 683             d2               = _mm_mul_ps(d,d);
 684             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 685
 686             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 687
 688             /* Evaluate switch function */
 689             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 690             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 691             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 692
 693             fscal            = fvdw;
 694
 695             fscal            = _mm_and_ps(fscal,cutoff_mask);
 696
 697             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 698
 699              /* Update vectorial force */
 700             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 701             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 702             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 703
 704             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 705             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 706             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 707             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 708             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 709                                                    _mm_mul_ps(dx00,fscal),
 710                                                    _mm_mul_ps(dy00,fscal),
 711                                                    _mm_mul_ps(dz00,fscal));
 712
 713             }
 714
 715             /* Inner loop uses 60 flops */
 716         }
 717
 718         /* End of innermost loop */
 719
 720         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 721                                               f+i_coord_offset,fshift+i_shift_offset);
 722
 723         /* Increment number of inner iterations */
 724         inneriter                  += j_index_end - j_index_start;
 725
 726         /* Outer loop uses 6 flops */
 727     }
 728
 729     /* Increment number of outer iterations */
 730     outeriter        += nri;
 731
 732     /* Update outer/inner flops */
 733
 734     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*60);
 735 }