src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecNone_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_ElecNone_VdwCSTab_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: None
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_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     int              nvdwtype;
  72     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  73     int              *vdwtype;
  74     real             *vdwparam;
  75     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  76     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  77     __m128i          vfitab;
  78     __m128i          ifour       = _mm_set1_epi32(4);
  79     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  80     real             *vftab;
  81     __m128           dummy_mask,cutoff_mask;
  82     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  83     __m128           one     = _mm_set1_ps(1.0);
  84     __m128           two     = _mm_set1_ps(2.0);
  85     x                = xx[0];
  86     f                = ff[0];
  87
  88     nri              = nlist->nri;
  89     iinr             = nlist->iinr;
  90     jindex           = nlist->jindex;
  91     jjnr             = nlist->jjnr;
  92     shiftidx         = nlist->shift;
  93     gid              = nlist->gid;
  94     shiftvec         = fr->shift_vec[0];
  95     fshift           = fr->fshift[0];
  96     nvdwtype         = fr->ntype;
  97     vdwparam         = fr->nbfp;
  98     vdwtype          = mdatoms->typeA;
  99
 100     vftab            = kernel_data->table_vdw->data;
 101     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 102
 103     /* Avoid stupid compiler warnings */
 104     jnrA = jnrB = jnrC = jnrD = 0;
 105     j_coord_offsetA = 0;
 106     j_coord_offsetB = 0;
 107     j_coord_offsetC = 0;
 108     j_coord_offsetD = 0;
 109
 110     outeriter        = 0;
 111     inneriter        = 0;
 112
 113     /* Start outer loop over neighborlists */
 114     for(iidx=0; iidx<nri; iidx++)
 115     {
 116         /* Load shift vector for this list */
 117         i_shift_offset   = DIM*shiftidx[iidx];
 118         shX              = shiftvec[i_shift_offset+XX];
 119         shY              = shiftvec[i_shift_offset+YY];
 120         shZ              = shiftvec[i_shift_offset+ZZ];
 121
 122         /* Load limits for loop over neighbors */
 123         j_index_start    = jindex[iidx];
 124         j_index_end      = jindex[iidx+1];
 125
 126         /* Get outer coordinate index */
 127         inr              = iinr[iidx];
 128         i_coord_offset   = DIM*inr;
 129
 130         /* Load i particle coords and add shift vector */
 131         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 132         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 133         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 134
 135         fix0             = _mm_setzero_ps();
 136         fiy0             = _mm_setzero_ps();
 137         fiz0             = _mm_setzero_ps();
 138
 139         /* Load parameters for i particles */
 140         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142         /* Reset potential sums */
 143         vvdwsum          = _mm_setzero_ps();
 144
 145         /* Start inner kernel loop */
 146         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 147         {
 148
 149             /* Get j neighbor index, and coordinate index */
 150             jnrA             = jjnr[jidx];
 151             jnrB             = jjnr[jidx+1];
 152             jnrC             = jjnr[jidx+2];
 153             jnrD             = jjnr[jidx+3];
 154
 155             j_coord_offsetA  = DIM*jnrA;
 156             j_coord_offsetB  = DIM*jnrB;
 157             j_coord_offsetC  = DIM*jnrC;
 158             j_coord_offsetD  = DIM*jnrD;
 159
 160             /* load j atom coordinates */
 161             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 162                                               x+j_coord_offsetC,x+j_coord_offsetD,
 163                                               &jx0,&jy0,&jz0);
 164
 165             /* Calculate displacement vector */
 166             dx00             = _mm_sub_ps(ix0,jx0);
 167             dy00             = _mm_sub_ps(iy0,jy0);
 168             dz00             = _mm_sub_ps(iz0,jz0);
 169
 170             /* Calculate squared distance and things based on it */
 171             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 172
 173             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 174
 175             /* Load parameters for j particles */
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178             vdwjidx0C        = 2*vdwtype[jnrC+0];
 179             vdwjidx0D        = 2*vdwtype[jnrD+0];
 180
 181             /**************************
 182              * CALCULATE INTERACTIONS *
 183              **************************/
 184
 185             r00              = _mm_mul_ps(rsq00,rinv00);
 186
 187             /* Compute parameters for interactions between i and j atoms */
 188             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 189                                          vdwparam+vdwioffset0+vdwjidx0B,
 190                                          vdwparam+vdwioffset0+vdwjidx0C,
 191                                          vdwparam+vdwioffset0+vdwjidx0D,
 192                                          &c6_00,&c12_00);
 193
 194             /* Calculate table index by multiplying r with table scale and truncate to integer */
 195             rt               = _mm_mul_ps(r00,vftabscale);
 196             vfitab           = _mm_cvttps_epi32(rt);
 197             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 198             vfitab           = _mm_slli_epi32(vfitab,3);
 199
 200             /* CUBIC SPLINE TABLE DISPERSION */
 201             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 202             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 203             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 204             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 205             _MM_TRANSPOSE4_PS(Y,F,G,H);
 206             Heps             = _mm_mul_ps(vfeps,H);
 207             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 208             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 209             vvdw6            = _mm_mul_ps(c6_00,VV);
 210             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 211             fvdw6            = _mm_mul_ps(c6_00,FF);
 212
 213             /* CUBIC SPLINE TABLE REPULSION */
 214             vfitab           = _mm_add_epi32(vfitab,ifour);
 215             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 216             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 217             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 218             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 219             _MM_TRANSPOSE4_PS(Y,F,G,H);
 220             Heps             = _mm_mul_ps(vfeps,H);
 221             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 222             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 223             vvdw12           = _mm_mul_ps(c12_00,VV);
 224             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 225             fvdw12           = _mm_mul_ps(c12_00,FF);
 226             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 227             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 228
 229             /* Update potential sum for this i atom from the interaction with this j atom. */
 230             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 231
 232             fscal            = fvdw;
 233
 234             /* Calculate temporary vectorial force */
 235             tx               = _mm_mul_ps(fscal,dx00);
 236             ty               = _mm_mul_ps(fscal,dy00);
 237             tz               = _mm_mul_ps(fscal,dz00);
 238
 239             /* Update vectorial force */
 240             fix0             = _mm_add_ps(fix0,tx);
 241             fiy0             = _mm_add_ps(fiy0,ty);
 242             fiz0             = _mm_add_ps(fiz0,tz);
 243
 244             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 245                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 246                                                    tx,ty,tz);
 247
 248             /* Inner loop uses 56 flops */
 249         }
 250
 251         if(jidx<j_index_end)
 252         {
 253
 254             /* Get j neighbor index, and coordinate index */
 255             jnrA             = jjnr[jidx];
 256             jnrB             = jjnr[jidx+1];
 257             jnrC             = jjnr[jidx+2];
 258             jnrD             = jjnr[jidx+3];
 259
 260             /* Sign of each element will be negative for non-real atoms.
 261              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 262              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 263              */
 264             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 265             jnrA       = (jnrA>=0) ? jnrA : 0;
 266             jnrB       = (jnrB>=0) ? jnrB : 0;
 267             jnrC       = (jnrC>=0) ? jnrC : 0;
 268             jnrD       = (jnrD>=0) ? jnrD : 0;
 269
 270             j_coord_offsetA  = DIM*jnrA;
 271             j_coord_offsetB  = DIM*jnrB;
 272             j_coord_offsetC  = DIM*jnrC;
 273             j_coord_offsetD  = DIM*jnrD;
 274
 275             /* load j atom coordinates */
 276             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 277                                               x+j_coord_offsetC,x+j_coord_offsetD,
 278                                               &jx0,&jy0,&jz0);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm_sub_ps(ix0,jx0);
 282             dy00             = _mm_sub_ps(iy0,jy0);
 283             dz00             = _mm_sub_ps(iz0,jz0);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 287
 288             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 289
 290             /* Load parameters for j particles */
 291             vdwjidx0A        = 2*vdwtype[jnrA+0];
 292             vdwjidx0B        = 2*vdwtype[jnrB+0];
 293             vdwjidx0C        = 2*vdwtype[jnrC+0];
 294             vdwjidx0D        = 2*vdwtype[jnrD+0];
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             r00              = _mm_mul_ps(rsq00,rinv00);
 301             r00              = _mm_andnot_ps(dummy_mask,r00);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 305                                          vdwparam+vdwioffset0+vdwjidx0B,
 306                                          vdwparam+vdwioffset0+vdwjidx0C,
 307                                          vdwparam+vdwioffset0+vdwjidx0D,
 308                                          &c6_00,&c12_00);
 309
 310             /* Calculate table index by multiplying r with table scale and truncate to integer */
 311             rt               = _mm_mul_ps(r00,vftabscale);
 312             vfitab           = _mm_cvttps_epi32(rt);
 313             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 314             vfitab           = _mm_slli_epi32(vfitab,3);
 315
 316             /* CUBIC SPLINE TABLE DISPERSION */
 317             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 318             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 319             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 320             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 321             _MM_TRANSPOSE4_PS(Y,F,G,H);
 322             Heps             = _mm_mul_ps(vfeps,H);
 323             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 324             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 325             vvdw6            = _mm_mul_ps(c6_00,VV);
 326             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 327             fvdw6            = _mm_mul_ps(c6_00,FF);
 328
 329             /* CUBIC SPLINE TABLE REPULSION */
 330             vfitab           = _mm_add_epi32(vfitab,ifour);
 331             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 332             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 333             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 334             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 335             _MM_TRANSPOSE4_PS(Y,F,G,H);
 336             Heps             = _mm_mul_ps(vfeps,H);
 337             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 338             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 339             vvdw12           = _mm_mul_ps(c12_00,VV);
 340             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 341             fvdw12           = _mm_mul_ps(c12_00,FF);
 342             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 343             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 347             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 348
 349             fscal            = fvdw;
 350
 351             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm_mul_ps(fscal,dx00);
 355             ty               = _mm_mul_ps(fscal,dy00);
 356             tz               = _mm_mul_ps(fscal,dz00);
 357
 358             /* Update vectorial force */
 359             fix0             = _mm_add_ps(fix0,tx);
 360             fiy0             = _mm_add_ps(fiy0,ty);
 361             fiz0             = _mm_add_ps(fiz0,tz);
 362
 363             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 364                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 365                                                    tx,ty,tz);
 366
 367             /* Inner loop uses 57 flops */
 368         }
 369
 370         /* End of innermost loop */
 371
 372         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 373                                               f+i_coord_offset,fshift+i_shift_offset);
 374
 375         ggid                        = gid[iidx];
 376         /* Update potential energies */
 377         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 378
 379         /* Increment number of inner iterations */
 380         inneriter                  += j_index_end - j_index_start;
 381
 382         /* Outer loop uses 10 flops */
 383     }
 384
 385     /* Increment number of outer iterations */
 386     outeriter        += nri;
 387
 388     /* Update outer/inner flops */
 389
 390     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*10 + inneriter*57);
 391 }
 392 /*
 393  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse2_single
 394  * Electrostatics interaction: None
 395  * VdW interaction:            CubicSplineTable
 396  * Geometry:                   Particle-Particle
 397  * Calculate force/pot:        Force
 398  */
 399 void
 400 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse2_single
 401                     (t_nblist * gmx_restrict                nlist,
 402                      rvec * gmx_restrict                    xx,
 403                      rvec * gmx_restrict                    ff,
 404                      t_forcerec * gmx_restrict              fr,
 405                      t_mdatoms * gmx_restrict               mdatoms,
 406                      nb_kernel_data_t * gmx_restrict        kernel_data,
 407                      t_nrnb * gmx_restrict                  nrnb)
 408 {
 409     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 410      * just 0 for non-waters.
 411      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 412      * jnr indices corresponding to data put in the four positions in the SIMD register.
 413      */
 414     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 415     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 416     int              jnrA,jnrB,jnrC,jnrD;
 417     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 418     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 419     real             shX,shY,shZ,rcutoff_scalar;
 420     real             *shiftvec,*fshift,*x,*f;
 421     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 422     int              vdwioffset0;
 423     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 424     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 425     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 426     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 427     int              nvdwtype;
 428     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 429     int              *vdwtype;
 430     real             *vdwparam;
 431     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 432     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 433     __m128i          vfitab;
 434     __m128i          ifour       = _mm_set1_epi32(4);
 435     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 436     real             *vftab;
 437     __m128           dummy_mask,cutoff_mask;
 438     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 439     __m128           one     = _mm_set1_ps(1.0);
 440     __m128           two     = _mm_set1_ps(2.0);
 441     x                = xx[0];
 442     f                = ff[0];
 443
 444     nri              = nlist->nri;
 445     iinr             = nlist->iinr;
 446     jindex           = nlist->jindex;
 447     jjnr             = nlist->jjnr;
 448     shiftidx         = nlist->shift;
 449     gid              = nlist->gid;
 450     shiftvec         = fr->shift_vec[0];
 451     fshift           = fr->fshift[0];
 452     nvdwtype         = fr->ntype;
 453     vdwparam         = fr->nbfp;
 454     vdwtype          = mdatoms->typeA;
 455
 456     vftab            = kernel_data->table_vdw->data;
 457     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 458
 459     /* Avoid stupid compiler warnings */
 460     jnrA = jnrB = jnrC = jnrD = 0;
 461     j_coord_offsetA = 0;
 462     j_coord_offsetB = 0;
 463     j_coord_offsetC = 0;
 464     j_coord_offsetD = 0;
 465
 466     outeriter        = 0;
 467     inneriter        = 0;
 468
 469     /* Start outer loop over neighborlists */
 470     for(iidx=0; iidx<nri; iidx++)
 471     {
 472         /* Load shift vector for this list */
 473         i_shift_offset   = DIM*shiftidx[iidx];
 474         shX              = shiftvec[i_shift_offset+XX];
 475         shY              = shiftvec[i_shift_offset+YY];
 476         shZ              = shiftvec[i_shift_offset+ZZ];
 477
 478         /* Load limits for loop over neighbors */
 479         j_index_start    = jindex[iidx];
 480         j_index_end      = jindex[iidx+1];
 481
 482         /* Get outer coordinate index */
 483         inr              = iinr[iidx];
 484         i_coord_offset   = DIM*inr;
 485
 486         /* Load i particle coords and add shift vector */
 487         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 488         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 489         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 490
 491         fix0             = _mm_setzero_ps();
 492         fiy0             = _mm_setzero_ps();
 493         fiz0             = _mm_setzero_ps();
 494
 495         /* Load parameters for i particles */
 496         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 497
 498         /* Start inner kernel loop */
 499         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 500         {
 501
 502             /* Get j neighbor index, and coordinate index */
 503             jnrA             = jjnr[jidx];
 504             jnrB             = jjnr[jidx+1];
 505             jnrC             = jjnr[jidx+2];
 506             jnrD             = jjnr[jidx+3];
 507
 508             j_coord_offsetA  = DIM*jnrA;
 509             j_coord_offsetB  = DIM*jnrB;
 510             j_coord_offsetC  = DIM*jnrC;
 511             j_coord_offsetD  = DIM*jnrD;
 512
 513             /* load j atom coordinates */
 514             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 515                                               x+j_coord_offsetC,x+j_coord_offsetD,
 516                                               &jx0,&jy0,&jz0);
 517
 518             /* Calculate displacement vector */
 519             dx00             = _mm_sub_ps(ix0,jx0);
 520             dy00             = _mm_sub_ps(iy0,jy0);
 521             dz00             = _mm_sub_ps(iz0,jz0);
 522
 523             /* Calculate squared distance and things based on it */
 524             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 525
 526             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 527
 528             /* Load parameters for j particles */
 529             vdwjidx0A        = 2*vdwtype[jnrA+0];
 530             vdwjidx0B        = 2*vdwtype[jnrB+0];
 531             vdwjidx0C        = 2*vdwtype[jnrC+0];
 532             vdwjidx0D        = 2*vdwtype[jnrD+0];
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             r00              = _mm_mul_ps(rsq00,rinv00);
 539
 540             /* Compute parameters for interactions between i and j atoms */
 541             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 542                                          vdwparam+vdwioffset0+vdwjidx0B,
 543                                          vdwparam+vdwioffset0+vdwjidx0C,
 544                                          vdwparam+vdwioffset0+vdwjidx0D,
 545                                          &c6_00,&c12_00);
 546
 547             /* Calculate table index by multiplying r with table scale and truncate to integer */
 548             rt               = _mm_mul_ps(r00,vftabscale);
 549             vfitab           = _mm_cvttps_epi32(rt);
 550             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 551             vfitab           = _mm_slli_epi32(vfitab,3);
 552
 553             /* CUBIC SPLINE TABLE DISPERSION */
 554             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 555             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 556             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 557             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 558             _MM_TRANSPOSE4_PS(Y,F,G,H);
 559             Heps             = _mm_mul_ps(vfeps,H);
 560             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 561             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 562             fvdw6            = _mm_mul_ps(c6_00,FF);
 563
 564             /* CUBIC SPLINE TABLE REPULSION */
 565             vfitab           = _mm_add_epi32(vfitab,ifour);
 566             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 567             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 568             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 569             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 570             _MM_TRANSPOSE4_PS(Y,F,G,H);
 571             Heps             = _mm_mul_ps(vfeps,H);
 572             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 573             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 574             fvdw12           = _mm_mul_ps(c12_00,FF);
 575             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 576
 577             fscal            = fvdw;
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_ps(fscal,dx00);
 581             ty               = _mm_mul_ps(fscal,dy00);
 582             tz               = _mm_mul_ps(fscal,dz00);
 583
 584             /* Update vectorial force */
 585             fix0             = _mm_add_ps(fix0,tx);
 586             fiy0             = _mm_add_ps(fiy0,ty);
 587             fiz0             = _mm_add_ps(fiz0,tz);
 588
 589             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 590                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 591                                                    tx,ty,tz);
 592
 593             /* Inner loop uses 48 flops */
 594         }
 595
 596         if(jidx<j_index_end)
 597         {
 598
 599             /* Get j neighbor index, and coordinate index */
 600             jnrA             = jjnr[jidx];
 601             jnrB             = jjnr[jidx+1];
 602             jnrC             = jjnr[jidx+2];
 603             jnrD             = jjnr[jidx+3];
 604
 605             /* Sign of each element will be negative for non-real atoms.
 606              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 607              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 608              */
 609             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 610             jnrA       = (jnrA>=0) ? jnrA : 0;
 611             jnrB       = (jnrB>=0) ? jnrB : 0;
 612             jnrC       = (jnrC>=0) ? jnrC : 0;
 613             jnrD       = (jnrD>=0) ? jnrD : 0;
 614
 615             j_coord_offsetA  = DIM*jnrA;
 616             j_coord_offsetB  = DIM*jnrB;
 617             j_coord_offsetC  = DIM*jnrC;
 618             j_coord_offsetD  = DIM*jnrD;
 619
 620             /* load j atom coordinates */
 621             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 622                                               x+j_coord_offsetC,x+j_coord_offsetD,
 623                                               &jx0,&jy0,&jz0);
 624
 625             /* Calculate displacement vector */
 626             dx00             = _mm_sub_ps(ix0,jx0);
 627             dy00             = _mm_sub_ps(iy0,jy0);
 628             dz00             = _mm_sub_ps(iz0,jz0);
 629
 630             /* Calculate squared distance and things based on it */
 631             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 632
 633             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 634
 635             /* Load parameters for j particles */
 636             vdwjidx0A        = 2*vdwtype[jnrA+0];
 637             vdwjidx0B        = 2*vdwtype[jnrB+0];
 638             vdwjidx0C        = 2*vdwtype[jnrC+0];
 639             vdwjidx0D        = 2*vdwtype[jnrD+0];
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             r00              = _mm_mul_ps(rsq00,rinv00);
 646             r00              = _mm_andnot_ps(dummy_mask,r00);
 647
 648             /* Compute parameters for interactions between i and j atoms */
 649             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 650                                          vdwparam+vdwioffset0+vdwjidx0B,
 651                                          vdwparam+vdwioffset0+vdwjidx0C,
 652                                          vdwparam+vdwioffset0+vdwjidx0D,
 653                                          &c6_00,&c12_00);
 654
 655             /* Calculate table index by multiplying r with table scale and truncate to integer */
 656             rt               = _mm_mul_ps(r00,vftabscale);
 657             vfitab           = _mm_cvttps_epi32(rt);
 658             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 659             vfitab           = _mm_slli_epi32(vfitab,3);
 660
 661             /* CUBIC SPLINE TABLE DISPERSION */
 662             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 663             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 664             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 665             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 666             _MM_TRANSPOSE4_PS(Y,F,G,H);
 667             Heps             = _mm_mul_ps(vfeps,H);
 668             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 669             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 670             fvdw6            = _mm_mul_ps(c6_00,FF);
 671
 672             /* CUBIC SPLINE TABLE REPULSION */
 673             vfitab           = _mm_add_epi32(vfitab,ifour);
 674             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 675             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 676             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 677             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 678             _MM_TRANSPOSE4_PS(Y,F,G,H);
 679             Heps             = _mm_mul_ps(vfeps,H);
 680             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 681             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 682             fvdw12           = _mm_mul_ps(c12_00,FF);
 683             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 684
 685             fscal            = fvdw;
 686
 687             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 688
 689             /* Calculate temporary vectorial force */
 690             tx               = _mm_mul_ps(fscal,dx00);
 691             ty               = _mm_mul_ps(fscal,dy00);
 692             tz               = _mm_mul_ps(fscal,dz00);
 693
 694             /* Update vectorial force */
 695             fix0             = _mm_add_ps(fix0,tx);
 696             fiy0             = _mm_add_ps(fiy0,ty);
 697             fiz0             = _mm_add_ps(fiz0,tz);
 698
 699             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 700                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 701                                                    tx,ty,tz);
 702
 703             /* Inner loop uses 49 flops */
 704         }
 705
 706         /* End of innermost loop */
 707
 708         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 709                                               f+i_coord_offset,fshift+i_shift_offset);
 710
 711         /* Increment number of inner iterations */
 712         inneriter                  += j_index_end - j_index_start;
 713
 714         /* Outer loop uses 9 flops */
 715     }
 716
 717     /* Increment number of outer iterations */
 718     outeriter        += nri;
 719
 720     /* Update outer/inner flops */
 721
 722     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*9 + inneriter*49);
 723 }