src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwCSTab_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_ElecCoul_VdwCSTab_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_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_vdw->data;
 105     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->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,3);
 210
 211             /* COULOMB ELECTROSTATICS */
 212             velec            = _mm_mul_ps(qq00,rinv00);
 213             felec            = _mm_mul_ps(velec,rinvsq00);
 214
 215             /* CUBIC SPLINE TABLE DISPERSION */
 216             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 217             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 218             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 219             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 220             _MM_TRANSPOSE4_PS(Y,F,G,H);
 221             Heps             = _mm_mul_ps(vfeps,H);
 222             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 223             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 224             vvdw6            = _mm_mul_ps(c6_00,VV);
 225             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 226             fvdw6            = _mm_mul_ps(c6_00,FF);
 227
 228             /* CUBIC SPLINE TABLE REPULSION */
 229             vfitab           = _mm_add_epi32(vfitab,ifour);
 230             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 231             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 232             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 233             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 234             _MM_TRANSPOSE4_PS(Y,F,G,H);
 235             Heps             = _mm_mul_ps(vfeps,H);
 236             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 237             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 238             vvdw12           = _mm_mul_ps(c12_00,VV);
 239             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 240             fvdw12           = _mm_mul_ps(c12_00,FF);
 241             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 242             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velecsum         = _mm_add_ps(velecsum,velec);
 246             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 247
 248             fscal            = _mm_add_ps(felec,fvdw);
 249
 250             /* Calculate temporary vectorial force */
 251             tx               = _mm_mul_ps(fscal,dx00);
 252             ty               = _mm_mul_ps(fscal,dy00);
 253             tz               = _mm_mul_ps(fscal,dz00);
 254
 255             /* Update vectorial force */
 256             fix0             = _mm_add_ps(fix0,tx);
 257             fiy0             = _mm_add_ps(fiy0,ty);
 258             fiz0             = _mm_add_ps(fiz0,tz);
 259
 260             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 261                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 262                                                    tx,ty,tz);
 263
 264             /* Inner loop uses 63 flops */
 265         }
 266
 267         if(jidx<j_index_end)
 268         {
 269
 270             /* Get j neighbor index, and coordinate index */
 271             jnrA             = jjnr[jidx];
 272             jnrB             = jjnr[jidx+1];
 273             jnrC             = jjnr[jidx+2];
 274             jnrD             = jjnr[jidx+3];
 275
 276             /* Sign of each element will be negative for non-real atoms.
 277              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 278              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 279              */
 280             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 281             jnrA       = (jnrA>=0) ? jnrA : 0;
 282             jnrB       = (jnrB>=0) ? jnrB : 0;
 283             jnrC       = (jnrC>=0) ? jnrC : 0;
 284             jnrD       = (jnrD>=0) ? jnrD : 0;
 285
 286             j_coord_offsetA  = DIM*jnrA;
 287             j_coord_offsetB  = DIM*jnrB;
 288             j_coord_offsetC  = DIM*jnrC;
 289             j_coord_offsetD  = DIM*jnrD;
 290
 291             /* load j atom coordinates */
 292             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 293                                               x+j_coord_offsetC,x+j_coord_offsetD,
 294                                               &jx0,&jy0,&jz0);
 295
 296             /* Calculate displacement vector */
 297             dx00             = _mm_sub_ps(ix0,jx0);
 298             dy00             = _mm_sub_ps(iy0,jy0);
 299             dz00             = _mm_sub_ps(iz0,jz0);
 300
 301             /* Calculate squared distance and things based on it */
 302             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 303
 304             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 305
 306             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 307
 308             /* Load parameters for j particles */
 309             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 310                                                               charge+jnrC+0,charge+jnrD+0);
 311             vdwjidx0A        = 2*vdwtype[jnrA+0];
 312             vdwjidx0B        = 2*vdwtype[jnrB+0];
 313             vdwjidx0C        = 2*vdwtype[jnrC+0];
 314             vdwjidx0D        = 2*vdwtype[jnrD+0];
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r00              = _mm_mul_ps(rsq00,rinv00);
 321             r00              = _mm_andnot_ps(dummy_mask,r00);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq00             = _mm_mul_ps(iq0,jq0);
 325             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 326                                          vdwparam+vdwioffset0+vdwjidx0B,
 327                                          vdwparam+vdwioffset0+vdwjidx0C,
 328                                          vdwparam+vdwioffset0+vdwjidx0D,
 329                                          &c6_00,&c12_00);
 330
 331             /* Calculate table index by multiplying r with table scale and truncate to integer */
 332             rt               = _mm_mul_ps(r00,vftabscale);
 333             vfitab           = _mm_cvttps_epi32(rt);
 334             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 335             vfitab           = _mm_slli_epi32(vfitab,3);
 336
 337             /* COULOMB ELECTROSTATICS */
 338             velec            = _mm_mul_ps(qq00,rinv00);
 339             felec            = _mm_mul_ps(velec,rinvsq00);
 340
 341             /* CUBIC SPLINE TABLE DISPERSION */
 342             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 343             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 344             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 345             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 346             _MM_TRANSPOSE4_PS(Y,F,G,H);
 347             Heps             = _mm_mul_ps(vfeps,H);
 348             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 349             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 350             vvdw6            = _mm_mul_ps(c6_00,VV);
 351             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 352             fvdw6            = _mm_mul_ps(c6_00,FF);
 353
 354             /* CUBIC SPLINE TABLE REPULSION */
 355             vfitab           = _mm_add_epi32(vfitab,ifour);
 356             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 357             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 358             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 359             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 360             _MM_TRANSPOSE4_PS(Y,F,G,H);
 361             Heps             = _mm_mul_ps(vfeps,H);
 362             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 363             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 364             vvdw12           = _mm_mul_ps(c12_00,VV);
 365             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 366             fvdw12           = _mm_mul_ps(c12_00,FF);
 367             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 368             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm_andnot_ps(dummy_mask,velec);
 372             velecsum         = _mm_add_ps(velecsum,velec);
 373             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 374             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 375
 376             fscal            = _mm_add_ps(felec,fvdw);
 377
 378             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm_mul_ps(fscal,dx00);
 382             ty               = _mm_mul_ps(fscal,dy00);
 383             tz               = _mm_mul_ps(fscal,dz00);
 384
 385             /* Update vectorial force */
 386             fix0             = _mm_add_ps(fix0,tx);
 387             fiy0             = _mm_add_ps(fiy0,ty);
 388             fiz0             = _mm_add_ps(fiz0,tz);
 389
 390             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 391                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 392                                                    tx,ty,tz);
 393
 394             /* Inner loop uses 64 flops */
 395         }
 396
 397         /* End of innermost loop */
 398
 399         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 400                                               f+i_coord_offset,fshift+i_shift_offset);
 401
 402         ggid                        = gid[iidx];
 403         /* Update potential energies */
 404         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 405         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 406
 407         /* Increment number of inner iterations */
 408         inneriter                  += j_index_end - j_index_start;
 409
 410         /* Outer loop uses 12 flops */
 411     }
 412
 413     /* Increment number of outer iterations */
 414     outeriter        += nri;
 415
 416     /* Update outer/inner flops */
 417
 418     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*64);
 419 }
 420 /*
 421  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_single
 422  * Electrostatics interaction: Coulomb
 423  * VdW interaction:            CubicSplineTable
 424  * Geometry:                   Particle-Particle
 425  * Calculate force/pot:        Force
 426  */
 427 void
 428 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_single
 429                     (t_nblist * gmx_restrict                nlist,
 430                      rvec * gmx_restrict                    xx,
 431                      rvec * gmx_restrict                    ff,
 432                      t_forcerec * gmx_restrict              fr,
 433                      t_mdatoms * gmx_restrict               mdatoms,
 434                      nb_kernel_data_t * gmx_restrict        kernel_data,
 435                      t_nrnb * gmx_restrict                  nrnb)
 436 {
 437     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 438      * just 0 for non-waters.
 439      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 440      * jnr indices corresponding to data put in the four positions in the SIMD register.
 441      */
 442     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 443     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 444     int              jnrA,jnrB,jnrC,jnrD;
 445     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 446     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 447     real             shX,shY,shZ,rcutoff_scalar;
 448     real             *shiftvec,*fshift,*x,*f;
 449     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 450     int              vdwioffset0;
 451     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 452     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 453     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 454     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 455     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 456     real             *charge;
 457     int              nvdwtype;
 458     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 459     int              *vdwtype;
 460     real             *vdwparam;
 461     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 462     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 463     __m128i          vfitab;
 464     __m128i          ifour       = _mm_set1_epi32(4);
 465     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 466     real             *vftab;
 467     __m128           dummy_mask,cutoff_mask;
 468     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 469     __m128           one     = _mm_set1_ps(1.0);
 470     __m128           two     = _mm_set1_ps(2.0);
 471     x                = xx[0];
 472     f                = ff[0];
 473
 474     nri              = nlist->nri;
 475     iinr             = nlist->iinr;
 476     jindex           = nlist->jindex;
 477     jjnr             = nlist->jjnr;
 478     shiftidx         = nlist->shift;
 479     gid              = nlist->gid;
 480     shiftvec         = fr->shift_vec[0];
 481     fshift           = fr->fshift[0];
 482     facel            = _mm_set1_ps(fr->epsfac);
 483     charge           = mdatoms->chargeA;
 484     nvdwtype         = fr->ntype;
 485     vdwparam         = fr->nbfp;
 486     vdwtype          = mdatoms->typeA;
 487
 488     vftab            = kernel_data->table_vdw->data;
 489     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 490
 491     /* Avoid stupid compiler warnings */
 492     jnrA = jnrB = jnrC = jnrD = 0;
 493     j_coord_offsetA = 0;
 494     j_coord_offsetB = 0;
 495     j_coord_offsetC = 0;
 496     j_coord_offsetD = 0;
 497
 498     outeriter        = 0;
 499     inneriter        = 0;
 500
 501     /* Start outer loop over neighborlists */
 502     for(iidx=0; iidx<nri; iidx++)
 503     {
 504         /* Load shift vector for this list */
 505         i_shift_offset   = DIM*shiftidx[iidx];
 506         shX              = shiftvec[i_shift_offset+XX];
 507         shY              = shiftvec[i_shift_offset+YY];
 508         shZ              = shiftvec[i_shift_offset+ZZ];
 509
 510         /* Load limits for loop over neighbors */
 511         j_index_start    = jindex[iidx];
 512         j_index_end      = jindex[iidx+1];
 513
 514         /* Get outer coordinate index */
 515         inr              = iinr[iidx];
 516         i_coord_offset   = DIM*inr;
 517
 518         /* Load i particle coords and add shift vector */
 519         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 520         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 521         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 522
 523         fix0             = _mm_setzero_ps();
 524         fiy0             = _mm_setzero_ps();
 525         fiz0             = _mm_setzero_ps();
 526
 527         /* Load parameters for i particles */
 528         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 529         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 530
 531         /* Start inner kernel loop */
 532         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 533         {
 534
 535             /* Get j neighbor index, and coordinate index */
 536             jnrA             = jjnr[jidx];
 537             jnrB             = jjnr[jidx+1];
 538             jnrC             = jjnr[jidx+2];
 539             jnrD             = jjnr[jidx+3];
 540
 541             j_coord_offsetA  = DIM*jnrA;
 542             j_coord_offsetB  = DIM*jnrB;
 543             j_coord_offsetC  = DIM*jnrC;
 544             j_coord_offsetD  = DIM*jnrD;
 545
 546             /* load j atom coordinates */
 547             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 548                                               x+j_coord_offsetC,x+j_coord_offsetD,
 549                                               &jx0,&jy0,&jz0);
 550
 551             /* Calculate displacement vector */
 552             dx00             = _mm_sub_ps(ix0,jx0);
 553             dy00             = _mm_sub_ps(iy0,jy0);
 554             dz00             = _mm_sub_ps(iz0,jz0);
 555
 556             /* Calculate squared distance and things based on it */
 557             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 558
 559             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 560
 561             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 562
 563             /* Load parameters for j particles */
 564             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 565                                                               charge+jnrC+0,charge+jnrD+0);
 566             vdwjidx0A        = 2*vdwtype[jnrA+0];
 567             vdwjidx0B        = 2*vdwtype[jnrB+0];
 568             vdwjidx0C        = 2*vdwtype[jnrC+0];
 569             vdwjidx0D        = 2*vdwtype[jnrD+0];
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             r00              = _mm_mul_ps(rsq00,rinv00);
 576
 577             /* Compute parameters for interactions between i and j atoms */
 578             qq00             = _mm_mul_ps(iq0,jq0);
 579             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 580                                          vdwparam+vdwioffset0+vdwjidx0B,
 581                                          vdwparam+vdwioffset0+vdwjidx0C,
 582                                          vdwparam+vdwioffset0+vdwjidx0D,
 583                                          &c6_00,&c12_00);
 584
 585             /* Calculate table index by multiplying r with table scale and truncate to integer */
 586             rt               = _mm_mul_ps(r00,vftabscale);
 587             vfitab           = _mm_cvttps_epi32(rt);
 588             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 589             vfitab           = _mm_slli_epi32(vfitab,3);
 590
 591             /* COULOMB ELECTROSTATICS */
 592             velec            = _mm_mul_ps(qq00,rinv00);
 593             felec            = _mm_mul_ps(velec,rinvsq00);
 594
 595             /* CUBIC SPLINE TABLE DISPERSION */
 596             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 597             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 598             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 599             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 600             _MM_TRANSPOSE4_PS(Y,F,G,H);
 601             Heps             = _mm_mul_ps(vfeps,H);
 602             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 603             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 604             fvdw6            = _mm_mul_ps(c6_00,FF);
 605
 606             /* CUBIC SPLINE TABLE REPULSION */
 607             vfitab           = _mm_add_epi32(vfitab,ifour);
 608             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 609             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 610             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 611             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 612             _MM_TRANSPOSE4_PS(Y,F,G,H);
 613             Heps             = _mm_mul_ps(vfeps,H);
 614             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 615             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 616             fvdw12           = _mm_mul_ps(c12_00,FF);
 617             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 618
 619             fscal            = _mm_add_ps(felec,fvdw);
 620
 621             /* Calculate temporary vectorial force */
 622             tx               = _mm_mul_ps(fscal,dx00);
 623             ty               = _mm_mul_ps(fscal,dy00);
 624             tz               = _mm_mul_ps(fscal,dz00);
 625
 626             /* Update vectorial force */
 627             fix0             = _mm_add_ps(fix0,tx);
 628             fiy0             = _mm_add_ps(fiy0,ty);
 629             fiz0             = _mm_add_ps(fiz0,tz);
 630
 631             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 632                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 633                                                    tx,ty,tz);
 634
 635             /* Inner loop uses 54 flops */
 636         }
 637
 638         if(jidx<j_index_end)
 639         {
 640
 641             /* Get j neighbor index, and coordinate index */
 642             jnrA             = jjnr[jidx];
 643             jnrB             = jjnr[jidx+1];
 644             jnrC             = jjnr[jidx+2];
 645             jnrD             = jjnr[jidx+3];
 646
 647             /* Sign of each element will be negative for non-real atoms.
 648              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 649              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 650              */
 651             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 652             jnrA       = (jnrA>=0) ? jnrA : 0;
 653             jnrB       = (jnrB>=0) ? jnrB : 0;
 654             jnrC       = (jnrC>=0) ? jnrC : 0;
 655             jnrD       = (jnrD>=0) ? jnrD : 0;
 656
 657             j_coord_offsetA  = DIM*jnrA;
 658             j_coord_offsetB  = DIM*jnrB;
 659             j_coord_offsetC  = DIM*jnrC;
 660             j_coord_offsetD  = DIM*jnrD;
 661
 662             /* load j atom coordinates */
 663             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 664                                               x+j_coord_offsetC,x+j_coord_offsetD,
 665                                               &jx0,&jy0,&jz0);
 666
 667             /* Calculate displacement vector */
 668             dx00             = _mm_sub_ps(ix0,jx0);
 669             dy00             = _mm_sub_ps(iy0,jy0);
 670             dz00             = _mm_sub_ps(iz0,jz0);
 671
 672             /* Calculate squared distance and things based on it */
 673             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 674
 675             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 676
 677             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 678
 679             /* Load parameters for j particles */
 680             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 681                                                               charge+jnrC+0,charge+jnrD+0);
 682             vdwjidx0A        = 2*vdwtype[jnrA+0];
 683             vdwjidx0B        = 2*vdwtype[jnrB+0];
 684             vdwjidx0C        = 2*vdwtype[jnrC+0];
 685             vdwjidx0D        = 2*vdwtype[jnrD+0];
 686
 687             /**************************
 688              * CALCULATE INTERACTIONS *
 689              **************************/
 690
 691             r00              = _mm_mul_ps(rsq00,rinv00);
 692             r00              = _mm_andnot_ps(dummy_mask,r00);
 693
 694             /* Compute parameters for interactions between i and j atoms */
 695             qq00             = _mm_mul_ps(iq0,jq0);
 696             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 697                                          vdwparam+vdwioffset0+vdwjidx0B,
 698                                          vdwparam+vdwioffset0+vdwjidx0C,
 699                                          vdwparam+vdwioffset0+vdwjidx0D,
 700                                          &c6_00,&c12_00);
 701
 702             /* Calculate table index by multiplying r with table scale and truncate to integer */
 703             rt               = _mm_mul_ps(r00,vftabscale);
 704             vfitab           = _mm_cvttps_epi32(rt);
 705             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 706             vfitab           = _mm_slli_epi32(vfitab,3);
 707
 708             /* COULOMB ELECTROSTATICS */
 709             velec            = _mm_mul_ps(qq00,rinv00);
 710             felec            = _mm_mul_ps(velec,rinvsq00);
 711
 712             /* CUBIC SPLINE TABLE DISPERSION */
 713             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 714             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 715             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 716             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 717             _MM_TRANSPOSE4_PS(Y,F,G,H);
 718             Heps             = _mm_mul_ps(vfeps,H);
 719             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 720             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 721             fvdw6            = _mm_mul_ps(c6_00,FF);
 722
 723             /* CUBIC SPLINE TABLE REPULSION */
 724             vfitab           = _mm_add_epi32(vfitab,ifour);
 725             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 726             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 727             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 728             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 729             _MM_TRANSPOSE4_PS(Y,F,G,H);
 730             Heps             = _mm_mul_ps(vfeps,H);
 731             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 732             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 733             fvdw12           = _mm_mul_ps(c12_00,FF);
 734             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 735
 736             fscal            = _mm_add_ps(felec,fvdw);
 737
 738             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 739
 740             /* Calculate temporary vectorial force */
 741             tx               = _mm_mul_ps(fscal,dx00);
 742             ty               = _mm_mul_ps(fscal,dy00);
 743             tz               = _mm_mul_ps(fscal,dz00);
 744
 745             /* Update vectorial force */
 746             fix0             = _mm_add_ps(fix0,tx);
 747             fiy0             = _mm_add_ps(fiy0,ty);
 748             fiz0             = _mm_add_ps(fiz0,tz);
 749
 750             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 751                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 752                                                    tx,ty,tz);
 753
 754             /* Inner loop uses 55 flops */
 755         }
 756
 757         /* End of innermost loop */
 758
 759         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 760                                               f+i_coord_offset,fshift+i_shift_offset);
 761
 762         /* Increment number of inner iterations */
 763         inneriter                  += j_index_end - j_index_start;
 764
 765         /* Outer loop uses 10 flops */
 766     }
 767
 768     /* Increment number of outer iterations */
 769     outeriter        += nri;
 770
 771     /* Update outer/inner flops */
 772
 773     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*55);
 774 }