src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_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 SSE, 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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  69     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  78     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  79     __m128i          vfitab;
  80     __m128i          ifour       = _mm_set1_epi32(4);
  81     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  82     real             *vftab;
  83     __m128           dummy_mask,cutoff_mask;
  84     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  85     __m128           one     = _mm_set1_ps(1.0);
  86     __m128           two     = _mm_set1_ps(2.0);
  87     x                = xx[0];
  88     f                = ff[0];
  89
  90     nri              = nlist->nri;
  91     iinr             = nlist->iinr;
  92     jindex           = nlist->jindex;
  93     jjnr             = nlist->jjnr;
  94     shiftidx         = nlist->shift;
  95     gid              = nlist->gid;
  96     shiftvec         = fr->shift_vec[0];
  97     fshift           = fr->fshift[0];
  98     facel            = _mm_set1_ps(fr->epsfac);
  99     charge           = mdatoms->chargeA;
 100     nvdwtype         = fr->ntype;
 101     vdwparam         = fr->nbfp;
 102     vdwtype          = mdatoms->typeA;
 103
 104     vftab            = kernel_data->table_elec->data;
 105     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 106
 107     /* Avoid stupid compiler warnings */
 108     jnrA = jnrB = jnrC = jnrD = 0;
 109     j_coord_offsetA = 0;
 110     j_coord_offsetB = 0;
 111     j_coord_offsetC = 0;
 112     j_coord_offsetD = 0;
 113
 114     outeriter        = 0;
 115     inneriter        = 0;
 116
 117     /* Start outer loop over neighborlists */
 118     for(iidx=0; iidx<nri; iidx++)
 119     {
 120         /* Load shift vector for this list */
 121         i_shift_offset   = DIM*shiftidx[iidx];
 122         shX              = shiftvec[i_shift_offset+XX];
 123         shY              = shiftvec[i_shift_offset+YY];
 124         shZ              = shiftvec[i_shift_offset+ZZ];
 125
 126         /* Load limits for loop over neighbors */
 127         j_index_start    = jindex[iidx];
 128         j_index_end      = jindex[iidx+1];
 129
 130         /* Get outer coordinate index */
 131         inr              = iinr[iidx];
 132         i_coord_offset   = DIM*inr;
 133
 134         /* Load i particle coords and add shift vector */
 135         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 136         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 137         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 138
 139         fix0             = _mm_setzero_ps();
 140         fiy0             = _mm_setzero_ps();
 141         fiz0             = _mm_setzero_ps();
 142
 143         /* Load parameters for i particles */
 144         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 145         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 146
 147         /* Reset potential sums */
 148         velecsum         = _mm_setzero_ps();
 149         vvdwsum          = _mm_setzero_ps();
 150
 151         /* Start inner kernel loop */
 152         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 153         {
 154
 155             /* Get j neighbor index, and coordinate index */
 156             jnrA             = jjnr[jidx];
 157             jnrB             = jjnr[jidx+1];
 158             jnrC             = jjnr[jidx+2];
 159             jnrD             = jjnr[jidx+3];
 160
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163             j_coord_offsetC  = DIM*jnrC;
 164             j_coord_offsetD  = DIM*jnrD;
 165
 166             /* load j atom coordinates */
 167             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                               x+j_coord_offsetC,x+j_coord_offsetD,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_ps(ix0,jx0);
 173             dy00             = _mm_sub_ps(iy0,jy0);
 174             dz00             = _mm_sub_ps(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 180
 181             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 185                                                               charge+jnrC+0,charge+jnrD+0);
 186             vdwjidx0A        = 2*vdwtype[jnrA+0];
 187             vdwjidx0B        = 2*vdwtype[jnrB+0];
 188             vdwjidx0C        = 2*vdwtype[jnrC+0];
 189             vdwjidx0D        = 2*vdwtype[jnrD+0];
 190
 191             /**************************
 192              * CALCULATE INTERACTIONS *
 193              **************************/
 194
 195             r00              = _mm_mul_ps(rsq00,rinv00);
 196
 197             /* Compute parameters for interactions between i and j atoms */
 198             qq00             = _mm_mul_ps(iq0,jq0);
 199             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 200                                          vdwparam+vdwioffset0+vdwjidx0B,
 201                                          vdwparam+vdwioffset0+vdwjidx0C,
 202                                          vdwparam+vdwioffset0+vdwjidx0D,
 203                                          &c6_00,&c12_00);
 204
 205             /* Calculate table index by multiplying r with table scale and truncate to integer */
 206             rt               = _mm_mul_ps(r00,vftabscale);
 207             vfitab           = _mm_cvttps_epi32(rt);
 208             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 209             vfitab           = _mm_slli_epi32(vfitab,2);
 210
 211             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 212             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 213             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 214             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 215             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 216             _MM_TRANSPOSE4_PS(Y,F,G,H);
 217             Heps             = _mm_mul_ps(vfeps,H);
 218             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 219             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 220             velec            = _mm_mul_ps(qq00,VV);
 221             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 222             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 223
 224             /* LENNARD-JONES DISPERSION/REPULSION */
 225
 226             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 227             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 228             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 229             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 230             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             velecsum         = _mm_add_ps(velecsum,velec);
 234             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 235
 236             fscal            = _mm_add_ps(felec,fvdw);
 237
 238             /* Calculate temporary vectorial force */
 239             tx               = _mm_mul_ps(fscal,dx00);
 240             ty               = _mm_mul_ps(fscal,dy00);
 241             tz               = _mm_mul_ps(fscal,dz00);
 242
 243             /* Update vectorial force */
 244             fix0             = _mm_add_ps(fix0,tx);
 245             fiy0             = _mm_add_ps(fiy0,ty);
 246             fiz0             = _mm_add_ps(fiz0,tz);
 247
 248             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 249                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 250                                                    tx,ty,tz);
 251
 252             /* Inner loop uses 56 flops */
 253         }
 254
 255         if(jidx<j_index_end)
 256         {
 257
 258             /* Get j neighbor index, and coordinate index */
 259             jnrA             = jjnr[jidx];
 260             jnrB             = jjnr[jidx+1];
 261             jnrC             = jjnr[jidx+2];
 262             jnrD             = jjnr[jidx+3];
 263
 264             /* Sign of each element will be negative for non-real atoms.
 265              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 266              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 267              */
 268             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 269             jnrA       = (jnrA>=0) ? jnrA : 0;
 270             jnrB       = (jnrB>=0) ? jnrB : 0;
 271             jnrC       = (jnrC>=0) ? jnrC : 0;
 272             jnrD       = (jnrD>=0) ? jnrD : 0;
 273
 274             j_coord_offsetA  = DIM*jnrA;
 275             j_coord_offsetB  = DIM*jnrB;
 276             j_coord_offsetC  = DIM*jnrC;
 277             j_coord_offsetD  = DIM*jnrD;
 278
 279             /* load j atom coordinates */
 280             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 281                                               x+j_coord_offsetC,x+j_coord_offsetD,
 282                                               &jx0,&jy0,&jz0);
 283
 284             /* Calculate displacement vector */
 285             dx00             = _mm_sub_ps(ix0,jx0);
 286             dy00             = _mm_sub_ps(iy0,jy0);
 287             dz00             = _mm_sub_ps(iz0,jz0);
 288
 289             /* Calculate squared distance and things based on it */
 290             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 291
 292             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 293
 294             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 295
 296             /* Load parameters for j particles */
 297             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 298                                                               charge+jnrC+0,charge+jnrD+0);
 299             vdwjidx0A        = 2*vdwtype[jnrA+0];
 300             vdwjidx0B        = 2*vdwtype[jnrB+0];
 301             vdwjidx0C        = 2*vdwtype[jnrC+0];
 302             vdwjidx0D        = 2*vdwtype[jnrD+0];
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             r00              = _mm_mul_ps(rsq00,rinv00);
 309             r00              = _mm_andnot_ps(dummy_mask,r00);
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq00             = _mm_mul_ps(iq0,jq0);
 313             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 314                                          vdwparam+vdwioffset0+vdwjidx0B,
 315                                          vdwparam+vdwioffset0+vdwjidx0C,
 316                                          vdwparam+vdwioffset0+vdwjidx0D,
 317                                          &c6_00,&c12_00);
 318
 319             /* Calculate table index by multiplying r with table scale and truncate to integer */
 320             rt               = _mm_mul_ps(r00,vftabscale);
 321             vfitab           = _mm_cvttps_epi32(rt);
 322             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 323             vfitab           = _mm_slli_epi32(vfitab,2);
 324
 325             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 326             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 327             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 328             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 329             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 330             _MM_TRANSPOSE4_PS(Y,F,G,H);
 331             Heps             = _mm_mul_ps(vfeps,H);
 332             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 333             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 334             velec            = _mm_mul_ps(qq00,VV);
 335             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 336             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 337
 338             /* LENNARD-JONES DISPERSION/REPULSION */
 339
 340             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 341             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 342             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 343             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 344             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velec            = _mm_andnot_ps(dummy_mask,velec);
 348             velecsum         = _mm_add_ps(velecsum,velec);
 349             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 350             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 351
 352             fscal            = _mm_add_ps(felec,fvdw);
 353
 354             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm_mul_ps(fscal,dx00);
 358             ty               = _mm_mul_ps(fscal,dy00);
 359             tz               = _mm_mul_ps(fscal,dz00);
 360
 361             /* Update vectorial force */
 362             fix0             = _mm_add_ps(fix0,tx);
 363             fiy0             = _mm_add_ps(fiy0,ty);
 364             fiz0             = _mm_add_ps(fiz0,tz);
 365
 366             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 367                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 368                                                    tx,ty,tz);
 369
 370             /* Inner loop uses 57 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(velecsum,kernel_data->energygrp_elec+ggid);
 381         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 382
 383         /* Increment number of inner iterations */
 384         inneriter                  += j_index_end - j_index_start;
 385
 386         /* Outer loop uses 12 flops */
 387     }
 388
 389     /* Increment number of outer iterations */
 390     outeriter        += nri;
 391
 392     /* Update outer/inner flops */
 393
 394     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*57);
 395 }
 396 /*
 397  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_single
 398  * Electrostatics interaction: CubicSplineTable
 399  * VdW interaction:            LennardJones
 400  * Geometry:                   Particle-Particle
 401  * Calculate force/pot:        Force
 402  */
 403 void
 404 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_single
 405                     (t_nblist * gmx_restrict                nlist,
 406                      rvec * gmx_restrict                    xx,
 407                      rvec * gmx_restrict                    ff,
 408                      t_forcerec * gmx_restrict              fr,
 409                      t_mdatoms * gmx_restrict               mdatoms,
 410                      nb_kernel_data_t * gmx_restrict        kernel_data,
 411                      t_nrnb * gmx_restrict                  nrnb)
 412 {
 413     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 414      * just 0 for non-waters.
 415      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 416      * jnr indices corresponding to data put in the four positions in the SIMD register.
 417      */
 418     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 419     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 420     int              jnrA,jnrB,jnrC,jnrD;
 421     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 422     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 423     real             shX,shY,shZ,rcutoff_scalar;
 424     real             *shiftvec,*fshift,*x,*f;
 425     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 426     int              vdwioffset0;
 427     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 428     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 429     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 430     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 431     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 432     real             *charge;
 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     __m128i          vfitab;
 440     __m128i          ifour       = _mm_set1_epi32(4);
 441     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 442     real             *vftab;
 443     __m128           dummy_mask,cutoff_mask;
 444     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 445     __m128           one     = _mm_set1_ps(1.0);
 446     __m128           two     = _mm_set1_ps(2.0);
 447     x                = xx[0];
 448     f                = ff[0];
 449
 450     nri              = nlist->nri;
 451     iinr             = nlist->iinr;
 452     jindex           = nlist->jindex;
 453     jjnr             = nlist->jjnr;
 454     shiftidx         = nlist->shift;
 455     gid              = nlist->gid;
 456     shiftvec         = fr->shift_vec[0];
 457     fshift           = fr->fshift[0];
 458     facel            = _mm_set1_ps(fr->epsfac);
 459     charge           = mdatoms->chargeA;
 460     nvdwtype         = fr->ntype;
 461     vdwparam         = fr->nbfp;
 462     vdwtype          = mdatoms->typeA;
 463
 464     vftab            = kernel_data->table_elec->data;
 465     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 466
 467     /* Avoid stupid compiler warnings */
 468     jnrA = jnrB = jnrC = jnrD = 0;
 469     j_coord_offsetA = 0;
 470     j_coord_offsetB = 0;
 471     j_coord_offsetC = 0;
 472     j_coord_offsetD = 0;
 473
 474     outeriter        = 0;
 475     inneriter        = 0;
 476
 477     /* Start outer loop over neighborlists */
 478     for(iidx=0; iidx<nri; iidx++)
 479     {
 480         /* Load shift vector for this list */
 481         i_shift_offset   = DIM*shiftidx[iidx];
 482         shX              = shiftvec[i_shift_offset+XX];
 483         shY              = shiftvec[i_shift_offset+YY];
 484         shZ              = shiftvec[i_shift_offset+ZZ];
 485
 486         /* Load limits for loop over neighbors */
 487         j_index_start    = jindex[iidx];
 488         j_index_end      = jindex[iidx+1];
 489
 490         /* Get outer coordinate index */
 491         inr              = iinr[iidx];
 492         i_coord_offset   = DIM*inr;
 493
 494         /* Load i particle coords and add shift vector */
 495         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 496         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 497         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 498
 499         fix0             = _mm_setzero_ps();
 500         fiy0             = _mm_setzero_ps();
 501         fiz0             = _mm_setzero_ps();
 502
 503         /* Load parameters for i particles */
 504         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 505         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 506
 507         /* Start inner kernel loop */
 508         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 509         {
 510
 511             /* Get j neighbor index, and coordinate index */
 512             jnrA             = jjnr[jidx];
 513             jnrB             = jjnr[jidx+1];
 514             jnrC             = jjnr[jidx+2];
 515             jnrD             = jjnr[jidx+3];
 516
 517             j_coord_offsetA  = DIM*jnrA;
 518             j_coord_offsetB  = DIM*jnrB;
 519             j_coord_offsetC  = DIM*jnrC;
 520             j_coord_offsetD  = DIM*jnrD;
 521
 522             /* load j atom coordinates */
 523             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 524                                               x+j_coord_offsetC,x+j_coord_offsetD,
 525                                               &jx0,&jy0,&jz0);
 526
 527             /* Calculate displacement vector */
 528             dx00             = _mm_sub_ps(ix0,jx0);
 529             dy00             = _mm_sub_ps(iy0,jy0);
 530             dz00             = _mm_sub_ps(iz0,jz0);
 531
 532             /* Calculate squared distance and things based on it */
 533             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 534
 535             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 536
 537             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 538
 539             /* Load parameters for j particles */
 540             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 541                                                               charge+jnrC+0,charge+jnrD+0);
 542             vdwjidx0A        = 2*vdwtype[jnrA+0];
 543             vdwjidx0B        = 2*vdwtype[jnrB+0];
 544             vdwjidx0C        = 2*vdwtype[jnrC+0];
 545             vdwjidx0D        = 2*vdwtype[jnrD+0];
 546
 547             /**************************
 548              * CALCULATE INTERACTIONS *
 549              **************************/
 550
 551             r00              = _mm_mul_ps(rsq00,rinv00);
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq00             = _mm_mul_ps(iq0,jq0);
 555             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 556                                          vdwparam+vdwioffset0+vdwjidx0B,
 557                                          vdwparam+vdwioffset0+vdwjidx0C,
 558                                          vdwparam+vdwioffset0+vdwjidx0D,
 559                                          &c6_00,&c12_00);
 560
 561             /* Calculate table index by multiplying r with table scale and truncate to integer */
 562             rt               = _mm_mul_ps(r00,vftabscale);
 563             vfitab           = _mm_cvttps_epi32(rt);
 564             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 565             vfitab           = _mm_slli_epi32(vfitab,2);
 566
 567             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 568             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 569             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 570             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 571             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 572             _MM_TRANSPOSE4_PS(Y,F,G,H);
 573             Heps             = _mm_mul_ps(vfeps,H);
 574             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 575             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 576             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 577
 578             /* LENNARD-JONES DISPERSION/REPULSION */
 579
 580             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 581             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 582
 583             fscal            = _mm_add_ps(felec,fvdw);
 584
 585             /* Calculate temporary vectorial force */
 586             tx               = _mm_mul_ps(fscal,dx00);
 587             ty               = _mm_mul_ps(fscal,dy00);
 588             tz               = _mm_mul_ps(fscal,dz00);
 589
 590             /* Update vectorial force */
 591             fix0             = _mm_add_ps(fix0,tx);
 592             fiy0             = _mm_add_ps(fiy0,ty);
 593             fiz0             = _mm_add_ps(fiz0,tz);
 594
 595             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 596                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 597                                                    tx,ty,tz);
 598
 599             /* Inner loop uses 47 flops */
 600         }
 601
 602         if(jidx<j_index_end)
 603         {
 604
 605             /* Get j neighbor index, and coordinate index */
 606             jnrA             = jjnr[jidx];
 607             jnrB             = jjnr[jidx+1];
 608             jnrC             = jjnr[jidx+2];
 609             jnrD             = jjnr[jidx+3];
 610
 611             /* Sign of each element will be negative for non-real atoms.
 612              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 613              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 614              */
 615             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 616             jnrA       = (jnrA>=0) ? jnrA : 0;
 617             jnrB       = (jnrB>=0) ? jnrB : 0;
 618             jnrC       = (jnrC>=0) ? jnrC : 0;
 619             jnrD       = (jnrD>=0) ? jnrD : 0;
 620
 621             j_coord_offsetA  = DIM*jnrA;
 622             j_coord_offsetB  = DIM*jnrB;
 623             j_coord_offsetC  = DIM*jnrC;
 624             j_coord_offsetD  = DIM*jnrD;
 625
 626             /* load j atom coordinates */
 627             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 628                                               x+j_coord_offsetC,x+j_coord_offsetD,
 629                                               &jx0,&jy0,&jz0);
 630
 631             /* Calculate displacement vector */
 632             dx00             = _mm_sub_ps(ix0,jx0);
 633             dy00             = _mm_sub_ps(iy0,jy0);
 634             dz00             = _mm_sub_ps(iz0,jz0);
 635
 636             /* Calculate squared distance and things based on it */
 637             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 638
 639             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 640
 641             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 642
 643             /* Load parameters for j particles */
 644             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 645                                                               charge+jnrC+0,charge+jnrD+0);
 646             vdwjidx0A        = 2*vdwtype[jnrA+0];
 647             vdwjidx0B        = 2*vdwtype[jnrB+0];
 648             vdwjidx0C        = 2*vdwtype[jnrC+0];
 649             vdwjidx0D        = 2*vdwtype[jnrD+0];
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             r00              = _mm_mul_ps(rsq00,rinv00);
 656             r00              = _mm_andnot_ps(dummy_mask,r00);
 657
 658             /* Compute parameters for interactions between i and j atoms */
 659             qq00             = _mm_mul_ps(iq0,jq0);
 660             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 661                                          vdwparam+vdwioffset0+vdwjidx0B,
 662                                          vdwparam+vdwioffset0+vdwjidx0C,
 663                                          vdwparam+vdwioffset0+vdwjidx0D,
 664                                          &c6_00,&c12_00);
 665
 666             /* Calculate table index by multiplying r with table scale and truncate to integer */
 667             rt               = _mm_mul_ps(r00,vftabscale);
 668             vfitab           = _mm_cvttps_epi32(rt);
 669             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 670             vfitab           = _mm_slli_epi32(vfitab,2);
 671
 672             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 673             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 674             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 675             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 676             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 677             _MM_TRANSPOSE4_PS(Y,F,G,H);
 678             Heps             = _mm_mul_ps(vfeps,H);
 679             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 680             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 681             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 682
 683             /* LENNARD-JONES DISPERSION/REPULSION */
 684
 685             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 686             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 687
 688             fscal            = _mm_add_ps(felec,fvdw);
 689
 690             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 691
 692             /* Calculate temporary vectorial force */
 693             tx               = _mm_mul_ps(fscal,dx00);
 694             ty               = _mm_mul_ps(fscal,dy00);
 695             tz               = _mm_mul_ps(fscal,dz00);
 696
 697             /* Update vectorial force */
 698             fix0             = _mm_add_ps(fix0,tx);
 699             fiy0             = _mm_add_ps(fiy0,ty);
 700             fiz0             = _mm_add_ps(fiz0,tz);
 701
 702             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 703                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 704                                                    tx,ty,tz);
 705
 706             /* Inner loop uses 48 flops */
 707         }
 708
 709         /* End of innermost loop */
 710
 711         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 712                                               f+i_coord_offset,fshift+i_shift_offset);
 713
 714         /* Increment number of inner iterations */
 715         inneriter                  += j_index_end - j_index_start;
 716
 717         /* Outer loop uses 10 flops */
 718     }
 719
 720     /* Increment number of outer iterations */
 721     outeriter        += nri;
 722
 723     /* Update outer/inner flops */
 724
 725     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*48);
 726 }