src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              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     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     int              nvdwtype;
  77     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  78     int              *vdwtype;
  79     real             *vdwparam;
  80     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  81     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  82     __m128i          vfitab;
  83     __m128i          ifour       = _mm_set1_epi32(4);
  84     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  85     real             *vftab;
  86     __m128           dummy_mask,cutoff_mask;
  87     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  88     __m128           one     = _mm_set1_ps(1.0);
  89     __m128           two     = _mm_set1_ps(2.0);
  90     x                = xx[0];
  91     f                = ff[0];
  92
  93     nri              = nlist->nri;
  94     iinr             = nlist->iinr;
  95     jindex           = nlist->jindex;
  96     jjnr             = nlist->jjnr;
  97     shiftidx         = nlist->shift;
  98     gid              = nlist->gid;
  99     shiftvec         = fr->shift_vec[0];
 100     fshift           = fr->fshift[0];
 101     facel            = _mm_set1_ps(fr->epsfac);
 102     charge           = mdatoms->chargeA;
 103     nvdwtype         = fr->ntype;
 104     vdwparam         = fr->nbfp;
 105     vdwtype          = mdatoms->typeA;
 106
 107     vftab            = kernel_data->table_elec->data;
 108     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 109
 110     /* Avoid stupid compiler warnings */
 111     jnrA = jnrB = jnrC = jnrD = 0;
 112     j_coord_offsetA = 0;
 113     j_coord_offsetB = 0;
 114     j_coord_offsetC = 0;
 115     j_coord_offsetD = 0;
 116
 117     outeriter        = 0;
 118     inneriter        = 0;
 119
 120     for(iidx=0;iidx<4*DIM;iidx++)
 121     {
 122         scratch[iidx] = 0.0;
 123     }
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _mm_setzero_ps();
 143         fiy0             = _mm_setzero_ps();
 144         fiz0             = _mm_setzero_ps();
 145
 146         /* Load parameters for i particles */
 147         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 148         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm_setzero_ps();
 152         vvdwsum          = _mm_setzero_ps();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             jnrC             = jjnr[jidx+2];
 162             jnrD             = jjnr[jidx+3];
 163             j_coord_offsetA  = DIM*jnrA;
 164             j_coord_offsetB  = DIM*jnrB;
 165             j_coord_offsetC  = DIM*jnrC;
 166             j_coord_offsetD  = DIM*jnrD;
 167
 168             /* load j atom coordinates */
 169             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 170                                               x+j_coord_offsetC,x+j_coord_offsetD,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _mm_sub_ps(ix0,jx0);
 175             dy00             = _mm_sub_ps(iy0,jy0);
 176             dz00             = _mm_sub_ps(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 182
 183             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 187                                                               charge+jnrC+0,charge+jnrD+0);
 188             vdwjidx0A        = 2*vdwtype[jnrA+0];
 189             vdwjidx0B        = 2*vdwtype[jnrB+0];
 190             vdwjidx0C        = 2*vdwtype[jnrC+0];
 191             vdwjidx0D        = 2*vdwtype[jnrD+0];
 192
 193             /**************************
 194              * CALCULATE INTERACTIONS *
 195              **************************/
 196
 197             r00              = _mm_mul_ps(rsq00,rinv00);
 198
 199             /* Compute parameters for interactions between i and j atoms */
 200             qq00             = _mm_mul_ps(iq0,jq0);
 201             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 202                                          vdwparam+vdwioffset0+vdwjidx0B,
 203                                          vdwparam+vdwioffset0+vdwjidx0C,
 204                                          vdwparam+vdwioffset0+vdwjidx0D,
 205                                          &c6_00,&c12_00);
 206
 207             /* Calculate table index by multiplying r with table scale and truncate to integer */
 208             rt               = _mm_mul_ps(r00,vftabscale);
 209             vfitab           = _mm_cvttps_epi32(rt);
 210             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 211             vfitab           = _mm_slli_epi32(vfitab,2);
 212
 213             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 214             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 215             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 216             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 217             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 218             _MM_TRANSPOSE4_PS(Y,F,G,H);
 219             Heps             = _mm_mul_ps(vfeps,H);
 220             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 221             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 222             velec            = _mm_mul_ps(qq00,VV);
 223             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 224             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 225
 226             /* LENNARD-JONES DISPERSION/REPULSION */
 227
 228             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 229             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 230             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 231             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 232             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 233
 234             /* Update potential sum for this i atom from the interaction with this j atom. */
 235             velecsum         = _mm_add_ps(velecsum,velec);
 236             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 237
 238             fscal            = _mm_add_ps(felec,fvdw);
 239
 240             /* Calculate temporary vectorial force */
 241             tx               = _mm_mul_ps(fscal,dx00);
 242             ty               = _mm_mul_ps(fscal,dy00);
 243             tz               = _mm_mul_ps(fscal,dz00);
 244
 245             /* Update vectorial force */
 246             fix0             = _mm_add_ps(fix0,tx);
 247             fiy0             = _mm_add_ps(fiy0,ty);
 248             fiz0             = _mm_add_ps(fiz0,tz);
 249
 250             fjptrA             = f+j_coord_offsetA;
 251             fjptrB             = f+j_coord_offsetB;
 252             fjptrC             = f+j_coord_offsetC;
 253             fjptrD             = f+j_coord_offsetD;
 254             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 255
 256             /* Inner loop uses 56 flops */
 257         }
 258
 259         if(jidx<j_index_end)
 260         {
 261
 262             /* Get j neighbor index, and coordinate index */
 263             jnrlistA         = jjnr[jidx];
 264             jnrlistB         = jjnr[jidx+1];
 265             jnrlistC         = jjnr[jidx+2];
 266             jnrlistD         = jjnr[jidx+3];
 267             /* Sign of each element will be negative for non-real atoms.
 268              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 269              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 270              */
 271             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 272             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 273             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 274             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 275             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 276             j_coord_offsetA  = DIM*jnrA;
 277             j_coord_offsetB  = DIM*jnrB;
 278             j_coord_offsetC  = DIM*jnrC;
 279             j_coord_offsetD  = DIM*jnrD;
 280
 281             /* load j atom coordinates */
 282             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 283                                               x+j_coord_offsetC,x+j_coord_offsetD,
 284                                               &jx0,&jy0,&jz0);
 285
 286             /* Calculate displacement vector */
 287             dx00             = _mm_sub_ps(ix0,jx0);
 288             dy00             = _mm_sub_ps(iy0,jy0);
 289             dz00             = _mm_sub_ps(iz0,jz0);
 290
 291             /* Calculate squared distance and things based on it */
 292             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 293
 294             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 295
 296             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 297
 298             /* Load parameters for j particles */
 299             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 300                                                               charge+jnrC+0,charge+jnrD+0);
 301             vdwjidx0A        = 2*vdwtype[jnrA+0];
 302             vdwjidx0B        = 2*vdwtype[jnrB+0];
 303             vdwjidx0C        = 2*vdwtype[jnrC+0];
 304             vdwjidx0D        = 2*vdwtype[jnrD+0];
 305
 306             /**************************
 307              * CALCULATE INTERACTIONS *
 308              **************************/
 309
 310             r00              = _mm_mul_ps(rsq00,rinv00);
 311             r00              = _mm_andnot_ps(dummy_mask,r00);
 312
 313             /* Compute parameters for interactions between i and j atoms */
 314             qq00             = _mm_mul_ps(iq0,jq0);
 315             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 316                                          vdwparam+vdwioffset0+vdwjidx0B,
 317                                          vdwparam+vdwioffset0+vdwjidx0C,
 318                                          vdwparam+vdwioffset0+vdwjidx0D,
 319                                          &c6_00,&c12_00);
 320
 321             /* Calculate table index by multiplying r with table scale and truncate to integer */
 322             rt               = _mm_mul_ps(r00,vftabscale);
 323             vfitab           = _mm_cvttps_epi32(rt);
 324             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 325             vfitab           = _mm_slli_epi32(vfitab,2);
 326
 327             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 328             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 329             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 330             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 331             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 332             _MM_TRANSPOSE4_PS(Y,F,G,H);
 333             Heps             = _mm_mul_ps(vfeps,H);
 334             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 335             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 336             velec            = _mm_mul_ps(qq00,VV);
 337             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 338             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 339
 340             /* LENNARD-JONES DISPERSION/REPULSION */
 341
 342             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 343             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 344             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 345             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 346             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velec            = _mm_andnot_ps(dummy_mask,velec);
 350             velecsum         = _mm_add_ps(velecsum,velec);
 351             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 352             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 353
 354             fscal            = _mm_add_ps(felec,fvdw);
 355
 356             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 357
 358             /* Calculate temporary vectorial force */
 359             tx               = _mm_mul_ps(fscal,dx00);
 360             ty               = _mm_mul_ps(fscal,dy00);
 361             tz               = _mm_mul_ps(fscal,dz00);
 362
 363             /* Update vectorial force */
 364             fix0             = _mm_add_ps(fix0,tx);
 365             fiy0             = _mm_add_ps(fiy0,ty);
 366             fiz0             = _mm_add_ps(fiz0,tz);
 367
 368             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 369             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 370             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 371             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 372             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 373
 374             /* Inner loop uses 57 flops */
 375         }
 376
 377         /* End of innermost loop */
 378
 379         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 380                                               f+i_coord_offset,fshift+i_shift_offset);
 381
 382         ggid                        = gid[iidx];
 383         /* Update potential energies */
 384         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 385         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 386
 387         /* Increment number of inner iterations */
 388         inneriter                  += j_index_end - j_index_start;
 389
 390         /* Outer loop uses 9 flops */
 391     }
 392
 393     /* Increment number of outer iterations */
 394     outeriter        += nri;
 395
 396     /* Update outer/inner flops */
 397
 398     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 399 }
 400 /*
 401  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_single
 402  * Electrostatics interaction: CubicSplineTable
 403  * VdW interaction:            LennardJones
 404  * Geometry:                   Particle-Particle
 405  * Calculate force/pot:        Force
 406  */
 407 void
 408 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_single
 409                     (t_nblist * gmx_restrict                nlist,
 410                      rvec * gmx_restrict                    xx,
 411                      rvec * gmx_restrict                    ff,
 412                      t_forcerec * gmx_restrict              fr,
 413                      t_mdatoms * gmx_restrict               mdatoms,
 414                      nb_kernel_data_t * gmx_restrict        kernel_data,
 415                      t_nrnb * gmx_restrict                  nrnb)
 416 {
 417     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 418      * just 0 for non-waters.
 419      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 420      * jnr indices corresponding to data put in the four positions in the SIMD register.
 421      */
 422     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 423     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 424     int              jnrA,jnrB,jnrC,jnrD;
 425     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 426     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 427     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 428     real             rcutoff_scalar;
 429     real             *shiftvec,*fshift,*x,*f;
 430     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 431     real             scratch[4*DIM];
 432     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 433     int              vdwioffset0;
 434     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 435     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 436     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 437     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 438     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 439     real             *charge;
 440     int              nvdwtype;
 441     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 442     int              *vdwtype;
 443     real             *vdwparam;
 444     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 445     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 446     __m128i          vfitab;
 447     __m128i          ifour       = _mm_set1_epi32(4);
 448     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 449     real             *vftab;
 450     __m128           dummy_mask,cutoff_mask;
 451     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 452     __m128           one     = _mm_set1_ps(1.0);
 453     __m128           two     = _mm_set1_ps(2.0);
 454     x                = xx[0];
 455     f                = ff[0];
 456
 457     nri              = nlist->nri;
 458     iinr             = nlist->iinr;
 459     jindex           = nlist->jindex;
 460     jjnr             = nlist->jjnr;
 461     shiftidx         = nlist->shift;
 462     gid              = nlist->gid;
 463     shiftvec         = fr->shift_vec[0];
 464     fshift           = fr->fshift[0];
 465     facel            = _mm_set1_ps(fr->epsfac);
 466     charge           = mdatoms->chargeA;
 467     nvdwtype         = fr->ntype;
 468     vdwparam         = fr->nbfp;
 469     vdwtype          = mdatoms->typeA;
 470
 471     vftab            = kernel_data->table_elec->data;
 472     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 473
 474     /* Avoid stupid compiler warnings */
 475     jnrA = jnrB = jnrC = jnrD = 0;
 476     j_coord_offsetA = 0;
 477     j_coord_offsetB = 0;
 478     j_coord_offsetC = 0;
 479     j_coord_offsetD = 0;
 480
 481     outeriter        = 0;
 482     inneriter        = 0;
 483
 484     for(iidx=0;iidx<4*DIM;iidx++)
 485     {
 486         scratch[iidx] = 0.0;
 487     }
 488
 489     /* Start outer loop over neighborlists */
 490     for(iidx=0; iidx<nri; iidx++)
 491     {
 492         /* Load shift vector for this list */
 493         i_shift_offset   = DIM*shiftidx[iidx];
 494
 495         /* Load limits for loop over neighbors */
 496         j_index_start    = jindex[iidx];
 497         j_index_end      = jindex[iidx+1];
 498
 499         /* Get outer coordinate index */
 500         inr              = iinr[iidx];
 501         i_coord_offset   = DIM*inr;
 502
 503         /* Load i particle coords and add shift vector */
 504         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 505
 506         fix0             = _mm_setzero_ps();
 507         fiy0             = _mm_setzero_ps();
 508         fiz0             = _mm_setzero_ps();
 509
 510         /* Load parameters for i particles */
 511         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 512         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 513
 514         /* Start inner kernel loop */
 515         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 516         {
 517
 518             /* Get j neighbor index, and coordinate index */
 519             jnrA             = jjnr[jidx];
 520             jnrB             = jjnr[jidx+1];
 521             jnrC             = jjnr[jidx+2];
 522             jnrD             = jjnr[jidx+3];
 523             j_coord_offsetA  = DIM*jnrA;
 524             j_coord_offsetB  = DIM*jnrB;
 525             j_coord_offsetC  = DIM*jnrC;
 526             j_coord_offsetD  = DIM*jnrD;
 527
 528             /* load j atom coordinates */
 529             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 530                                               x+j_coord_offsetC,x+j_coord_offsetD,
 531                                               &jx0,&jy0,&jz0);
 532
 533             /* Calculate displacement vector */
 534             dx00             = _mm_sub_ps(ix0,jx0);
 535             dy00             = _mm_sub_ps(iy0,jy0);
 536             dz00             = _mm_sub_ps(iz0,jz0);
 537
 538             /* Calculate squared distance and things based on it */
 539             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 540
 541             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 542
 543             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 544
 545             /* Load parameters for j particles */
 546             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 547                                                               charge+jnrC+0,charge+jnrD+0);
 548             vdwjidx0A        = 2*vdwtype[jnrA+0];
 549             vdwjidx0B        = 2*vdwtype[jnrB+0];
 550             vdwjidx0C        = 2*vdwtype[jnrC+0];
 551             vdwjidx0D        = 2*vdwtype[jnrD+0];
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             r00              = _mm_mul_ps(rsq00,rinv00);
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq00             = _mm_mul_ps(iq0,jq0);
 561             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 562                                          vdwparam+vdwioffset0+vdwjidx0B,
 563                                          vdwparam+vdwioffset0+vdwjidx0C,
 564                                          vdwparam+vdwioffset0+vdwjidx0D,
 565                                          &c6_00,&c12_00);
 566
 567             /* Calculate table index by multiplying r with table scale and truncate to integer */
 568             rt               = _mm_mul_ps(r00,vftabscale);
 569             vfitab           = _mm_cvttps_epi32(rt);
 570             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 571             vfitab           = _mm_slli_epi32(vfitab,2);
 572
 573             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 574             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 575             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 576             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 577             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 578             _MM_TRANSPOSE4_PS(Y,F,G,H);
 579             Heps             = _mm_mul_ps(vfeps,H);
 580             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 581             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 582             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 583
 584             /* LENNARD-JONES DISPERSION/REPULSION */
 585
 586             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 587             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 588
 589             fscal            = _mm_add_ps(felec,fvdw);
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm_mul_ps(fscal,dx00);
 593             ty               = _mm_mul_ps(fscal,dy00);
 594             tz               = _mm_mul_ps(fscal,dz00);
 595
 596             /* Update vectorial force */
 597             fix0             = _mm_add_ps(fix0,tx);
 598             fiy0             = _mm_add_ps(fiy0,ty);
 599             fiz0             = _mm_add_ps(fiz0,tz);
 600
 601             fjptrA             = f+j_coord_offsetA;
 602             fjptrB             = f+j_coord_offsetB;
 603             fjptrC             = f+j_coord_offsetC;
 604             fjptrD             = f+j_coord_offsetD;
 605             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 606
 607             /* Inner loop uses 47 flops */
 608         }
 609
 610         if(jidx<j_index_end)
 611         {
 612
 613             /* Get j neighbor index, and coordinate index */
 614             jnrlistA         = jjnr[jidx];
 615             jnrlistB         = jjnr[jidx+1];
 616             jnrlistC         = jjnr[jidx+2];
 617             jnrlistD         = jjnr[jidx+3];
 618             /* Sign of each element will be negative for non-real atoms.
 619              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 620              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 621              */
 622             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 623             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 624             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 625             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 626             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 627             j_coord_offsetA  = DIM*jnrA;
 628             j_coord_offsetB  = DIM*jnrB;
 629             j_coord_offsetC  = DIM*jnrC;
 630             j_coord_offsetD  = DIM*jnrD;
 631
 632             /* load j atom coordinates */
 633             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 634                                               x+j_coord_offsetC,x+j_coord_offsetD,
 635                                               &jx0,&jy0,&jz0);
 636
 637             /* Calculate displacement vector */
 638             dx00             = _mm_sub_ps(ix0,jx0);
 639             dy00             = _mm_sub_ps(iy0,jy0);
 640             dz00             = _mm_sub_ps(iz0,jz0);
 641
 642             /* Calculate squared distance and things based on it */
 643             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 644
 645             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 646
 647             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 648
 649             /* Load parameters for j particles */
 650             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 651                                                               charge+jnrC+0,charge+jnrD+0);
 652             vdwjidx0A        = 2*vdwtype[jnrA+0];
 653             vdwjidx0B        = 2*vdwtype[jnrB+0];
 654             vdwjidx0C        = 2*vdwtype[jnrC+0];
 655             vdwjidx0D        = 2*vdwtype[jnrD+0];
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             r00              = _mm_mul_ps(rsq00,rinv00);
 662             r00              = _mm_andnot_ps(dummy_mask,r00);
 663
 664             /* Compute parameters for interactions between i and j atoms */
 665             qq00             = _mm_mul_ps(iq0,jq0);
 666             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 667                                          vdwparam+vdwioffset0+vdwjidx0B,
 668                                          vdwparam+vdwioffset0+vdwjidx0C,
 669                                          vdwparam+vdwioffset0+vdwjidx0D,
 670                                          &c6_00,&c12_00);
 671
 672             /* Calculate table index by multiplying r with table scale and truncate to integer */
 673             rt               = _mm_mul_ps(r00,vftabscale);
 674             vfitab           = _mm_cvttps_epi32(rt);
 675             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 676             vfitab           = _mm_slli_epi32(vfitab,2);
 677
 678             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 679             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 680             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 681             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 682             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 683             _MM_TRANSPOSE4_PS(Y,F,G,H);
 684             Heps             = _mm_mul_ps(vfeps,H);
 685             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 686             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 687             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 688
 689             /* LENNARD-JONES DISPERSION/REPULSION */
 690
 691             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 692             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 693
 694             fscal            = _mm_add_ps(felec,fvdw);
 695
 696             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 697
 698             /* Calculate temporary vectorial force */
 699             tx               = _mm_mul_ps(fscal,dx00);
 700             ty               = _mm_mul_ps(fscal,dy00);
 701             tz               = _mm_mul_ps(fscal,dz00);
 702
 703             /* Update vectorial force */
 704             fix0             = _mm_add_ps(fix0,tx);
 705             fiy0             = _mm_add_ps(fiy0,ty);
 706             fiz0             = _mm_add_ps(fiz0,tz);
 707
 708             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 709             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 710             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 711             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 712             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 713
 714             /* Inner loop uses 48 flops */
 715         }
 716
 717         /* End of innermost loop */
 718
 719         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 720                                               f+i_coord_offset,fshift+i_shift_offset);
 721
 722         /* Increment number of inner iterations */
 723         inneriter                  += j_index_end - j_index_start;
 724
 725         /* Outer loop uses 7 flops */
 726     }
 727
 728     /* Increment number of outer iterations */
 729     outeriter        += nri;
 730
 731     /* Update outer/inner flops */
 732
 733     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*48);
 734 }