src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecNone_VdwCSTab_GeomP1P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_VF_sse4_1_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_sse4_1_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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     int              nvdwtype;
  75     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  76     int              *vdwtype;
  77     real             *vdwparam;
  78     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  79     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  80     __m128i          vfitab;
  81     __m128i          ifour       = _mm_set1_epi32(4);
  82     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  83     real             *vftab;
  84     __m128           dummy_mask,cutoff_mask;
  85     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  86     __m128           one     = _mm_set1_ps(1.0);
  87     __m128           two     = _mm_set1_ps(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     nvdwtype         = fr->ntype;
 100     vdwparam         = fr->nbfp;
 101     vdwtype          = mdatoms->typeA;
 102
 103     vftab            = kernel_data->table_vdw->data;
 104     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 105
 106     /* Avoid stupid compiler warnings */
 107     jnrA = jnrB = jnrC = jnrD = 0;
 108     j_coord_offsetA = 0;
 109     j_coord_offsetB = 0;
 110     j_coord_offsetC = 0;
 111     j_coord_offsetD = 0;
 112
 113     outeriter        = 0;
 114     inneriter        = 0;
 115
 116     for(iidx=0;iidx<4*DIM;iidx++)
 117     {
 118         scratch[iidx] = 0.0;
 119     }
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 137
 138         fix0             = _mm_setzero_ps();
 139         fiy0             = _mm_setzero_ps();
 140         fiz0             = _mm_setzero_ps();
 141
 142         /* Load parameters for i particles */
 143         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145         /* Reset potential sums */
 146         vvdwsum          = _mm_setzero_ps();
 147
 148         /* Start inner kernel loop */
 149         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 150         {
 151
 152             /* Get j neighbor index, and coordinate index */
 153             jnrA             = jjnr[jidx];
 154             jnrB             = jjnr[jidx+1];
 155             jnrC             = jjnr[jidx+2];
 156             jnrD             = jjnr[jidx+3];
 157             j_coord_offsetA  = DIM*jnrA;
 158             j_coord_offsetB  = DIM*jnrB;
 159             j_coord_offsetC  = DIM*jnrC;
 160             j_coord_offsetD  = DIM*jnrD;
 161
 162             /* load j atom coordinates */
 163             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 164                                               x+j_coord_offsetC,x+j_coord_offsetD,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_ps(ix0,jx0);
 169             dy00             = _mm_sub_ps(iy0,jy0);
 170             dz00             = _mm_sub_ps(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 174
 175             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 176
 177             /* Load parameters for j particles */
 178             vdwjidx0A        = 2*vdwtype[jnrA+0];
 179             vdwjidx0B        = 2*vdwtype[jnrB+0];
 180             vdwjidx0C        = 2*vdwtype[jnrC+0];
 181             vdwjidx0D        = 2*vdwtype[jnrD+0];
 182
 183             /**************************
 184              * CALCULATE INTERACTIONS *
 185              **************************/
 186
 187             r00              = _mm_mul_ps(rsq00,rinv00);
 188
 189             /* Compute parameters for interactions between i and j atoms */
 190             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 191                                          vdwparam+vdwioffset0+vdwjidx0B,
 192                                          vdwparam+vdwioffset0+vdwjidx0C,
 193                                          vdwparam+vdwioffset0+vdwjidx0D,
 194                                          &c6_00,&c12_00);
 195
 196             /* Calculate table index by multiplying r with table scale and truncate to integer */
 197             rt               = _mm_mul_ps(r00,vftabscale);
 198             vfitab           = _mm_cvttps_epi32(rt);
 199             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 200             vfitab           = _mm_slli_epi32(vfitab,3);
 201
 202             /* CUBIC SPLINE TABLE DISPERSION */
 203             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 204             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 205             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 206             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 207             _MM_TRANSPOSE4_PS(Y,F,G,H);
 208             Heps             = _mm_mul_ps(vfeps,H);
 209             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 210             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 211             vvdw6            = _mm_mul_ps(c6_00,VV);
 212             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 213             fvdw6            = _mm_mul_ps(c6_00,FF);
 214
 215             /* CUBIC SPLINE TABLE REPULSION */
 216             vfitab           = _mm_add_epi32(vfitab,ifour);
 217             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 218             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 219             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 220             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 221             _MM_TRANSPOSE4_PS(Y,F,G,H);
 222             Heps             = _mm_mul_ps(vfeps,H);
 223             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 224             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 225             vvdw12           = _mm_mul_ps(c12_00,VV);
 226             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 227             fvdw12           = _mm_mul_ps(c12_00,FF);
 228             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 229             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 233
 234             fscal            = fvdw;
 235
 236             /* Calculate temporary vectorial force */
 237             tx               = _mm_mul_ps(fscal,dx00);
 238             ty               = _mm_mul_ps(fscal,dy00);
 239             tz               = _mm_mul_ps(fscal,dz00);
 240
 241             /* Update vectorial force */
 242             fix0             = _mm_add_ps(fix0,tx);
 243             fiy0             = _mm_add_ps(fiy0,ty);
 244             fiz0             = _mm_add_ps(fiz0,tz);
 245
 246             fjptrA             = f+j_coord_offsetA;
 247             fjptrB             = f+j_coord_offsetB;
 248             fjptrC             = f+j_coord_offsetC;
 249             fjptrD             = f+j_coord_offsetD;
 250             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,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             jnrlistA         = jjnr[jidx];
 260             jnrlistB         = jjnr[jidx+1];
 261             jnrlistC         = jjnr[jidx+2];
 262             jnrlistD         = jjnr[jidx+3];
 263             /* Sign of each element will be negative for non-real atoms.
 264              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 265              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 266              */
 267             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 268             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 269             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 270             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 271             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 272             j_coord_offsetA  = DIM*jnrA;
 273             j_coord_offsetB  = DIM*jnrB;
 274             j_coord_offsetC  = DIM*jnrC;
 275             j_coord_offsetD  = DIM*jnrD;
 276
 277             /* load j atom coordinates */
 278             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 279                                               x+j_coord_offsetC,x+j_coord_offsetD,
 280                                               &jx0,&jy0,&jz0);
 281
 282             /* Calculate displacement vector */
 283             dx00             = _mm_sub_ps(ix0,jx0);
 284             dy00             = _mm_sub_ps(iy0,jy0);
 285             dz00             = _mm_sub_ps(iz0,jz0);
 286
 287             /* Calculate squared distance and things based on it */
 288             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 289
 290             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 291
 292             /* Load parameters for j particles */
 293             vdwjidx0A        = 2*vdwtype[jnrA+0];
 294             vdwjidx0B        = 2*vdwtype[jnrB+0];
 295             vdwjidx0C        = 2*vdwtype[jnrC+0];
 296             vdwjidx0D        = 2*vdwtype[jnrD+0];
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r00              = _mm_mul_ps(rsq00,rinv00);
 303             r00              = _mm_andnot_ps(dummy_mask,r00);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 307                                          vdwparam+vdwioffset0+vdwjidx0B,
 308                                          vdwparam+vdwioffset0+vdwjidx0C,
 309                                          vdwparam+vdwioffset0+vdwjidx0D,
 310                                          &c6_00,&c12_00);
 311
 312             /* Calculate table index by multiplying r with table scale and truncate to integer */
 313             rt               = _mm_mul_ps(r00,vftabscale);
 314             vfitab           = _mm_cvttps_epi32(rt);
 315             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 316             vfitab           = _mm_slli_epi32(vfitab,3);
 317
 318             /* CUBIC SPLINE TABLE DISPERSION */
 319             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 320             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 321             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 322             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 323             _MM_TRANSPOSE4_PS(Y,F,G,H);
 324             Heps             = _mm_mul_ps(vfeps,H);
 325             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 326             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 327             vvdw6            = _mm_mul_ps(c6_00,VV);
 328             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 329             fvdw6            = _mm_mul_ps(c6_00,FF);
 330
 331             /* CUBIC SPLINE TABLE REPULSION */
 332             vfitab           = _mm_add_epi32(vfitab,ifour);
 333             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 334             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 335             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 336             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 337             _MM_TRANSPOSE4_PS(Y,F,G,H);
 338             Heps             = _mm_mul_ps(vfeps,H);
 339             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 340             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 341             vvdw12           = _mm_mul_ps(c12_00,VV);
 342             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 343             fvdw12           = _mm_mul_ps(c12_00,FF);
 344             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 345             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 349             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 350
 351             fscal            = fvdw;
 352
 353             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm_mul_ps(fscal,dx00);
 357             ty               = _mm_mul_ps(fscal,dy00);
 358             tz               = _mm_mul_ps(fscal,dz00);
 359
 360             /* Update vectorial force */
 361             fix0             = _mm_add_ps(fix0,tx);
 362             fiy0             = _mm_add_ps(fiy0,ty);
 363             fiz0             = _mm_add_ps(fiz0,tz);
 364
 365             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 366             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 367             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 368             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 369             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 370
 371             /* Inner loop uses 57 flops */
 372         }
 373
 374         /* End of innermost loop */
 375
 376         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 377                                               f+i_coord_offset,fshift+i_shift_offset);
 378
 379         ggid                        = gid[iidx];
 380         /* Update potential energies */
 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 7 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_VDW_VF,outeriter*7 + inneriter*57);
 395 }
 396 /*
 397  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_single
 398  * Electrostatics interaction: None
 399  * VdW interaction:            CubicSplineTable
 400  * Geometry:                   Particle-Particle
 401  * Calculate force/pot:        Force
 402  */
 403 void
 404 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_sse4_1_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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 422     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 423     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 424     real             rcutoff_scalar;
 425     real             *shiftvec,*fshift,*x,*f;
 426     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 427     real             scratch[4*DIM];
 428     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 429     int              vdwioffset0;
 430     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 431     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 432     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 433     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 434     int              nvdwtype;
 435     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 436     int              *vdwtype;
 437     real             *vdwparam;
 438     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 439     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 440     __m128i          vfitab;
 441     __m128i          ifour       = _mm_set1_epi32(4);
 442     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 443     real             *vftab;
 444     __m128           dummy_mask,cutoff_mask;
 445     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 446     __m128           one     = _mm_set1_ps(1.0);
 447     __m128           two     = _mm_set1_ps(2.0);
 448     x                = xx[0];
 449     f                = ff[0];
 450
 451     nri              = nlist->nri;
 452     iinr             = nlist->iinr;
 453     jindex           = nlist->jindex;
 454     jjnr             = nlist->jjnr;
 455     shiftidx         = nlist->shift;
 456     gid              = nlist->gid;
 457     shiftvec         = fr->shift_vec[0];
 458     fshift           = fr->fshift[0];
 459     nvdwtype         = fr->ntype;
 460     vdwparam         = fr->nbfp;
 461     vdwtype          = mdatoms->typeA;
 462
 463     vftab            = kernel_data->table_vdw->data;
 464     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 465
 466     /* Avoid stupid compiler warnings */
 467     jnrA = jnrB = jnrC = jnrD = 0;
 468     j_coord_offsetA = 0;
 469     j_coord_offsetB = 0;
 470     j_coord_offsetC = 0;
 471     j_coord_offsetD = 0;
 472
 473     outeriter        = 0;
 474     inneriter        = 0;
 475
 476     for(iidx=0;iidx<4*DIM;iidx++)
 477     {
 478         scratch[iidx] = 0.0;
 479     }
 480
 481     /* Start outer loop over neighborlists */
 482     for(iidx=0; iidx<nri; iidx++)
 483     {
 484         /* Load shift vector for this list */
 485         i_shift_offset   = DIM*shiftidx[iidx];
 486
 487         /* Load limits for loop over neighbors */
 488         j_index_start    = jindex[iidx];
 489         j_index_end      = jindex[iidx+1];
 490
 491         /* Get outer coordinate index */
 492         inr              = iinr[iidx];
 493         i_coord_offset   = DIM*inr;
 494
 495         /* Load i particle coords and add shift vector */
 496         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 497
 498         fix0             = _mm_setzero_ps();
 499         fiy0             = _mm_setzero_ps();
 500         fiz0             = _mm_setzero_ps();
 501
 502         /* Load parameters for i particles */
 503         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 504
 505         /* Start inner kernel loop */
 506         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 507         {
 508
 509             /* Get j neighbor index, and coordinate index */
 510             jnrA             = jjnr[jidx];
 511             jnrB             = jjnr[jidx+1];
 512             jnrC             = jjnr[jidx+2];
 513             jnrD             = jjnr[jidx+3];
 514             j_coord_offsetA  = DIM*jnrA;
 515             j_coord_offsetB  = DIM*jnrB;
 516             j_coord_offsetC  = DIM*jnrC;
 517             j_coord_offsetD  = DIM*jnrD;
 518
 519             /* load j atom coordinates */
 520             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 521                                               x+j_coord_offsetC,x+j_coord_offsetD,
 522                                               &jx0,&jy0,&jz0);
 523
 524             /* Calculate displacement vector */
 525             dx00             = _mm_sub_ps(ix0,jx0);
 526             dy00             = _mm_sub_ps(iy0,jy0);
 527             dz00             = _mm_sub_ps(iz0,jz0);
 528
 529             /* Calculate squared distance and things based on it */
 530             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 531
 532             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 533
 534             /* Load parameters for j particles */
 535             vdwjidx0A        = 2*vdwtype[jnrA+0];
 536             vdwjidx0B        = 2*vdwtype[jnrB+0];
 537             vdwjidx0C        = 2*vdwtype[jnrC+0];
 538             vdwjidx0D        = 2*vdwtype[jnrD+0];
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             r00              = _mm_mul_ps(rsq00,rinv00);
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 548                                          vdwparam+vdwioffset0+vdwjidx0B,
 549                                          vdwparam+vdwioffset0+vdwjidx0C,
 550                                          vdwparam+vdwioffset0+vdwjidx0D,
 551                                          &c6_00,&c12_00);
 552
 553             /* Calculate table index by multiplying r with table scale and truncate to integer */
 554             rt               = _mm_mul_ps(r00,vftabscale);
 555             vfitab           = _mm_cvttps_epi32(rt);
 556             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 557             vfitab           = _mm_slli_epi32(vfitab,3);
 558
 559             /* CUBIC SPLINE TABLE DISPERSION */
 560             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 561             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 562             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 563             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 564             _MM_TRANSPOSE4_PS(Y,F,G,H);
 565             Heps             = _mm_mul_ps(vfeps,H);
 566             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 567             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 568             fvdw6            = _mm_mul_ps(c6_00,FF);
 569
 570             /* CUBIC SPLINE TABLE REPULSION */
 571             vfitab           = _mm_add_epi32(vfitab,ifour);
 572             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 573             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 574             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 575             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 576             _MM_TRANSPOSE4_PS(Y,F,G,H);
 577             Heps             = _mm_mul_ps(vfeps,H);
 578             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 579             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 580             fvdw12           = _mm_mul_ps(c12_00,FF);
 581             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 582
 583             fscal            = 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             fjptrA             = f+j_coord_offsetA;
 596             fjptrB             = f+j_coord_offsetB;
 597             fjptrC             = f+j_coord_offsetC;
 598             fjptrD             = f+j_coord_offsetD;
 599             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 600
 601             /* Inner loop uses 48 flops */
 602         }
 603
 604         if(jidx<j_index_end)
 605         {
 606
 607             /* Get j neighbor index, and coordinate index */
 608             jnrlistA         = jjnr[jidx];
 609             jnrlistB         = jjnr[jidx+1];
 610             jnrlistC         = jjnr[jidx+2];
 611             jnrlistD         = jjnr[jidx+3];
 612             /* Sign of each element will be negative for non-real atoms.
 613              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 614              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 615              */
 616             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 617             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 618             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 619             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 620             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 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             /* Load parameters for j particles */
 642             vdwjidx0A        = 2*vdwtype[jnrA+0];
 643             vdwjidx0B        = 2*vdwtype[jnrB+0];
 644             vdwjidx0C        = 2*vdwtype[jnrC+0];
 645             vdwjidx0D        = 2*vdwtype[jnrD+0];
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             r00              = _mm_mul_ps(rsq00,rinv00);
 652             r00              = _mm_andnot_ps(dummy_mask,r00);
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 656                                          vdwparam+vdwioffset0+vdwjidx0B,
 657                                          vdwparam+vdwioffset0+vdwjidx0C,
 658                                          vdwparam+vdwioffset0+vdwjidx0D,
 659                                          &c6_00,&c12_00);
 660
 661             /* Calculate table index by multiplying r with table scale and truncate to integer */
 662             rt               = _mm_mul_ps(r00,vftabscale);
 663             vfitab           = _mm_cvttps_epi32(rt);
 664             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 665             vfitab           = _mm_slli_epi32(vfitab,3);
 666
 667             /* CUBIC SPLINE TABLE DISPERSION */
 668             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 669             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 670             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 671             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 672             _MM_TRANSPOSE4_PS(Y,F,G,H);
 673             Heps             = _mm_mul_ps(vfeps,H);
 674             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 675             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 676             fvdw6            = _mm_mul_ps(c6_00,FF);
 677
 678             /* CUBIC SPLINE TABLE REPULSION */
 679             vfitab           = _mm_add_epi32(vfitab,ifour);
 680             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 681             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 682             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 683             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 684             _MM_TRANSPOSE4_PS(Y,F,G,H);
 685             Heps             = _mm_mul_ps(vfeps,H);
 686             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 687             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 688             fvdw12           = _mm_mul_ps(c12_00,FF);
 689             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 690
 691             fscal            = fvdw;
 692
 693             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 694
 695             /* Calculate temporary vectorial force */
 696             tx               = _mm_mul_ps(fscal,dx00);
 697             ty               = _mm_mul_ps(fscal,dy00);
 698             tz               = _mm_mul_ps(fscal,dz00);
 699
 700             /* Update vectorial force */
 701             fix0             = _mm_add_ps(fix0,tx);
 702             fiy0             = _mm_add_ps(fiy0,ty);
 703             fiz0             = _mm_add_ps(fiz0,tz);
 704
 705             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 706             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 707             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 708             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 709             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 710
 711             /* Inner loop uses 49 flops */
 712         }
 713
 714         /* End of innermost loop */
 715
 716         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 717                                               f+i_coord_offset,fshift+i_shift_offset);
 718
 719         /* Increment number of inner iterations */
 720         inneriter                  += j_index_end - j_index_start;
 721
 722         /* Outer loop uses 6 flops */
 723     }
 724
 725     /* Increment number of outer iterations */
 726     outeriter        += nri;
 727
 728     /* Update outer/inner flops */
 729
 730     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*49);
 731 }