src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128i          vfitab;
  97     __m128i          ifour       = _mm_set1_epi32(4);
  98     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  99     real             *vftab;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     vftab            = kernel_data->table_elec->data;
 122     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     for(iidx=0;iidx<4*DIM;iidx++)
 135     {
 136         scratch[iidx] = 0.0;
 137     }
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 155
 156         fix0             = _mm_setzero_ps();
 157         fiy0             = _mm_setzero_ps();
 158         fiz0             = _mm_setzero_ps();
 159
 160         /* Load parameters for i particles */
 161         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 162         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_ps();
 166         vvdwsum          = _mm_setzero_ps();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             jnrC             = jjnr[jidx+2];
 176             jnrD             = jjnr[jidx+3];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179             j_coord_offsetC  = DIM*jnrC;
 180             j_coord_offsetD  = DIM*jnrD;
 181
 182             /* load j atom coordinates */
 183             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 184                                               x+j_coord_offsetC,x+j_coord_offsetD,
 185                                               &jx0,&jy0,&jz0);
 186
 187             /* Calculate displacement vector */
 188             dx00             = _mm_sub_ps(ix0,jx0);
 189             dy00             = _mm_sub_ps(iy0,jy0);
 190             dz00             = _mm_sub_ps(iz0,jz0);
 191
 192             /* Calculate squared distance and things based on it */
 193             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 194
 195             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 196
 197             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 198
 199             /* Load parameters for j particles */
 200             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 201                                                               charge+jnrC+0,charge+jnrD+0);
 202             vdwjidx0A        = 2*vdwtype[jnrA+0];
 203             vdwjidx0B        = 2*vdwtype[jnrB+0];
 204             vdwjidx0C        = 2*vdwtype[jnrC+0];
 205             vdwjidx0D        = 2*vdwtype[jnrD+0];
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             r00              = _mm_mul_ps(rsq00,rinv00);
 212
 213             /* Compute parameters for interactions between i and j atoms */
 214             qq00             = _mm_mul_ps(iq0,jq0);
 215             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 216                                          vdwparam+vdwioffset0+vdwjidx0B,
 217                                          vdwparam+vdwioffset0+vdwjidx0C,
 218                                          vdwparam+vdwioffset0+vdwjidx0D,
 219                                          &c6_00,&c12_00);
 220
 221             /* Calculate table index by multiplying r with table scale and truncate to integer */
 222             rt               = _mm_mul_ps(r00,vftabscale);
 223             vfitab           = _mm_cvttps_epi32(rt);
 224             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 225             vfitab           = _mm_slli_epi32(vfitab,2);
 226
 227             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 228             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 229             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 230             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 231             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 232             _MM_TRANSPOSE4_PS(Y,F,G,H);
 233             Heps             = _mm_mul_ps(vfeps,H);
 234             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 235             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 236             velec            = _mm_mul_ps(qq00,VV);
 237             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 238             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 239
 240             /* LENNARD-JONES DISPERSION/REPULSION */
 241
 242             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 243             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 244             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 245             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 246             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             velecsum         = _mm_add_ps(velecsum,velec);
 250             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 251
 252             fscal            = _mm_add_ps(felec,fvdw);
 253
 254             /* Calculate temporary vectorial force */
 255             tx               = _mm_mul_ps(fscal,dx00);
 256             ty               = _mm_mul_ps(fscal,dy00);
 257             tz               = _mm_mul_ps(fscal,dz00);
 258
 259             /* Update vectorial force */
 260             fix0             = _mm_add_ps(fix0,tx);
 261             fiy0             = _mm_add_ps(fiy0,ty);
 262             fiz0             = _mm_add_ps(fiz0,tz);
 263
 264             fjptrA             = f+j_coord_offsetA;
 265             fjptrB             = f+j_coord_offsetB;
 266             fjptrC             = f+j_coord_offsetC;
 267             fjptrD             = f+j_coord_offsetD;
 268             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 269
 270             /* Inner loop uses 56 flops */
 271         }
 272
 273         if(jidx<j_index_end)
 274         {
 275
 276             /* Get j neighbor index, and coordinate index */
 277             jnrlistA         = jjnr[jidx];
 278             jnrlistB         = jjnr[jidx+1];
 279             jnrlistC         = jjnr[jidx+2];
 280             jnrlistD         = jjnr[jidx+3];
 281             /* Sign of each element will be negative for non-real atoms.
 282              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 283              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 284              */
 285             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 286             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 287             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 288             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 289             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 290             j_coord_offsetA  = DIM*jnrA;
 291             j_coord_offsetB  = DIM*jnrB;
 292             j_coord_offsetC  = DIM*jnrC;
 293             j_coord_offsetD  = DIM*jnrD;
 294
 295             /* load j atom coordinates */
 296             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 297                                               x+j_coord_offsetC,x+j_coord_offsetD,
 298                                               &jx0,&jy0,&jz0);
 299
 300             /* Calculate displacement vector */
 301             dx00             = _mm_sub_ps(ix0,jx0);
 302             dy00             = _mm_sub_ps(iy0,jy0);
 303             dz00             = _mm_sub_ps(iz0,jz0);
 304
 305             /* Calculate squared distance and things based on it */
 306             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 307
 308             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 309
 310             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 311
 312             /* Load parameters for j particles */
 313             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 314                                                               charge+jnrC+0,charge+jnrD+0);
 315             vdwjidx0A        = 2*vdwtype[jnrA+0];
 316             vdwjidx0B        = 2*vdwtype[jnrB+0];
 317             vdwjidx0C        = 2*vdwtype[jnrC+0];
 318             vdwjidx0D        = 2*vdwtype[jnrD+0];
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r00              = _mm_mul_ps(rsq00,rinv00);
 325             r00              = _mm_andnot_ps(dummy_mask,r00);
 326
 327             /* Compute parameters for interactions between i and j atoms */
 328             qq00             = _mm_mul_ps(iq0,jq0);
 329             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 330                                          vdwparam+vdwioffset0+vdwjidx0B,
 331                                          vdwparam+vdwioffset0+vdwjidx0C,
 332                                          vdwparam+vdwioffset0+vdwjidx0D,
 333                                          &c6_00,&c12_00);
 334
 335             /* Calculate table index by multiplying r with table scale and truncate to integer */
 336             rt               = _mm_mul_ps(r00,vftabscale);
 337             vfitab           = _mm_cvttps_epi32(rt);
 338             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 339             vfitab           = _mm_slli_epi32(vfitab,2);
 340
 341             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 342             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 343             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 344             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 345             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 346             _MM_TRANSPOSE4_PS(Y,F,G,H);
 347             Heps             = _mm_mul_ps(vfeps,H);
 348             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 349             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 350             velec            = _mm_mul_ps(qq00,VV);
 351             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 352             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 353
 354             /* LENNARD-JONES DISPERSION/REPULSION */
 355
 356             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 357             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 358             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 359             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 360             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velec            = _mm_andnot_ps(dummy_mask,velec);
 364             velecsum         = _mm_add_ps(velecsum,velec);
 365             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 366             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 367
 368             fscal            = _mm_add_ps(felec,fvdw);
 369
 370             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm_mul_ps(fscal,dx00);
 374             ty               = _mm_mul_ps(fscal,dy00);
 375             tz               = _mm_mul_ps(fscal,dz00);
 376
 377             /* Update vectorial force */
 378             fix0             = _mm_add_ps(fix0,tx);
 379             fiy0             = _mm_add_ps(fiy0,ty);
 380             fiz0             = _mm_add_ps(fiz0,tz);
 381
 382             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 383             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 384             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 385             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 386             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 387
 388             /* Inner loop uses 57 flops */
 389         }
 390
 391         /* End of innermost loop */
 392
 393         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 394                                               f+i_coord_offset,fshift+i_shift_offset);
 395
 396         ggid                        = gid[iidx];
 397         /* Update potential energies */
 398         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 399         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 400
 401         /* Increment number of inner iterations */
 402         inneriter                  += j_index_end - j_index_start;
 403
 404         /* Outer loop uses 9 flops */
 405     }
 406
 407     /* Increment number of outer iterations */
 408     outeriter        += nri;
 409
 410     /* Update outer/inner flops */
 411
 412     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 413 }
 414 /*
 415  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_single
 416  * Electrostatics interaction: CubicSplineTable
 417  * VdW interaction:            LennardJones
 418  * Geometry:                   Particle-Particle
 419  * Calculate force/pot:        Force
 420  */
 421 void
 422 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sse2_single
 423                     (t_nblist                    * gmx_restrict       nlist,
 424                      rvec                        * gmx_restrict          xx,
 425                      rvec                        * gmx_restrict          ff,
 426                      t_forcerec                  * gmx_restrict          fr,
 427                      t_mdatoms                   * gmx_restrict     mdatoms,
 428                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 429                      t_nrnb                      * gmx_restrict        nrnb)
 430 {
 431     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 432      * just 0 for non-waters.
 433      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 434      * jnr indices corresponding to data put in the four positions in the SIMD register.
 435      */
 436     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 437     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 438     int              jnrA,jnrB,jnrC,jnrD;
 439     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 440     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 441     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 442     real             rcutoff_scalar;
 443     real             *shiftvec,*fshift,*x,*f;
 444     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 445     real             scratch[4*DIM];
 446     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 447     int              vdwioffset0;
 448     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 449     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 450     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 451     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 452     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 453     real             *charge;
 454     int              nvdwtype;
 455     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 456     int              *vdwtype;
 457     real             *vdwparam;
 458     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 459     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 460     __m128i          vfitab;
 461     __m128i          ifour       = _mm_set1_epi32(4);
 462     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 463     real             *vftab;
 464     __m128           dummy_mask,cutoff_mask;
 465     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 466     __m128           one     = _mm_set1_ps(1.0);
 467     __m128           two     = _mm_set1_ps(2.0);
 468     x                = xx[0];
 469     f                = ff[0];
 470
 471     nri              = nlist->nri;
 472     iinr             = nlist->iinr;
 473     jindex           = nlist->jindex;
 474     jjnr             = nlist->jjnr;
 475     shiftidx         = nlist->shift;
 476     gid              = nlist->gid;
 477     shiftvec         = fr->shift_vec[0];
 478     fshift           = fr->fshift[0];
 479     facel            = _mm_set1_ps(fr->epsfac);
 480     charge           = mdatoms->chargeA;
 481     nvdwtype         = fr->ntype;
 482     vdwparam         = fr->nbfp;
 483     vdwtype          = mdatoms->typeA;
 484
 485     vftab            = kernel_data->table_elec->data;
 486     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 487
 488     /* Avoid stupid compiler warnings */
 489     jnrA = jnrB = jnrC = jnrD = 0;
 490     j_coord_offsetA = 0;
 491     j_coord_offsetB = 0;
 492     j_coord_offsetC = 0;
 493     j_coord_offsetD = 0;
 494
 495     outeriter        = 0;
 496     inneriter        = 0;
 497
 498     for(iidx=0;iidx<4*DIM;iidx++)
 499     {
 500         scratch[iidx] = 0.0;
 501     }
 502
 503     /* Start outer loop over neighborlists */
 504     for(iidx=0; iidx<nri; iidx++)
 505     {
 506         /* Load shift vector for this list */
 507         i_shift_offset   = DIM*shiftidx[iidx];
 508
 509         /* Load limits for loop over neighbors */
 510         j_index_start    = jindex[iidx];
 511         j_index_end      = jindex[iidx+1];
 512
 513         /* Get outer coordinate index */
 514         inr              = iinr[iidx];
 515         i_coord_offset   = DIM*inr;
 516
 517         /* Load i particle coords and add shift vector */
 518         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 519
 520         fix0             = _mm_setzero_ps();
 521         fiy0             = _mm_setzero_ps();
 522         fiz0             = _mm_setzero_ps();
 523
 524         /* Load parameters for i particles */
 525         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 526         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 527
 528         /* Start inner kernel loop */
 529         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 530         {
 531
 532             /* Get j neighbor index, and coordinate index */
 533             jnrA             = jjnr[jidx];
 534             jnrB             = jjnr[jidx+1];
 535             jnrC             = jjnr[jidx+2];
 536             jnrD             = jjnr[jidx+3];
 537             j_coord_offsetA  = DIM*jnrA;
 538             j_coord_offsetB  = DIM*jnrB;
 539             j_coord_offsetC  = DIM*jnrC;
 540             j_coord_offsetD  = DIM*jnrD;
 541
 542             /* load j atom coordinates */
 543             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 544                                               x+j_coord_offsetC,x+j_coord_offsetD,
 545                                               &jx0,&jy0,&jz0);
 546
 547             /* Calculate displacement vector */
 548             dx00             = _mm_sub_ps(ix0,jx0);
 549             dy00             = _mm_sub_ps(iy0,jy0);
 550             dz00             = _mm_sub_ps(iz0,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 554
 555             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 556
 557             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 558
 559             /* Load parameters for j particles */
 560             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 561                                                               charge+jnrC+0,charge+jnrD+0);
 562             vdwjidx0A        = 2*vdwtype[jnrA+0];
 563             vdwjidx0B        = 2*vdwtype[jnrB+0];
 564             vdwjidx0C        = 2*vdwtype[jnrC+0];
 565             vdwjidx0D        = 2*vdwtype[jnrD+0];
 566
 567             /**************************
 568              * CALCULATE INTERACTIONS *
 569              **************************/
 570
 571             r00              = _mm_mul_ps(rsq00,rinv00);
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq00             = _mm_mul_ps(iq0,jq0);
 575             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 576                                          vdwparam+vdwioffset0+vdwjidx0B,
 577                                          vdwparam+vdwioffset0+vdwjidx0C,
 578                                          vdwparam+vdwioffset0+vdwjidx0D,
 579                                          &c6_00,&c12_00);
 580
 581             /* Calculate table index by multiplying r with table scale and truncate to integer */
 582             rt               = _mm_mul_ps(r00,vftabscale);
 583             vfitab           = _mm_cvttps_epi32(rt);
 584             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 585             vfitab           = _mm_slli_epi32(vfitab,2);
 586
 587             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 588             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 589             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 590             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 591             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 592             _MM_TRANSPOSE4_PS(Y,F,G,H);
 593             Heps             = _mm_mul_ps(vfeps,H);
 594             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 595             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 596             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 597
 598             /* LENNARD-JONES DISPERSION/REPULSION */
 599
 600             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 601             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 602
 603             fscal            = _mm_add_ps(felec,fvdw);
 604
 605             /* Calculate temporary vectorial force */
 606             tx               = _mm_mul_ps(fscal,dx00);
 607             ty               = _mm_mul_ps(fscal,dy00);
 608             tz               = _mm_mul_ps(fscal,dz00);
 609
 610             /* Update vectorial force */
 611             fix0             = _mm_add_ps(fix0,tx);
 612             fiy0             = _mm_add_ps(fiy0,ty);
 613             fiz0             = _mm_add_ps(fiz0,tz);
 614
 615             fjptrA             = f+j_coord_offsetA;
 616             fjptrB             = f+j_coord_offsetB;
 617             fjptrC             = f+j_coord_offsetC;
 618             fjptrD             = f+j_coord_offsetD;
 619             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 620
 621             /* Inner loop uses 47 flops */
 622         }
 623
 624         if(jidx<j_index_end)
 625         {
 626
 627             /* Get j neighbor index, and coordinate index */
 628             jnrlistA         = jjnr[jidx];
 629             jnrlistB         = jjnr[jidx+1];
 630             jnrlistC         = jjnr[jidx+2];
 631             jnrlistD         = jjnr[jidx+3];
 632             /* Sign of each element will be negative for non-real atoms.
 633              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 634              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 635              */
 636             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 637             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 638             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 639             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 640             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 641             j_coord_offsetA  = DIM*jnrA;
 642             j_coord_offsetB  = DIM*jnrB;
 643             j_coord_offsetC  = DIM*jnrC;
 644             j_coord_offsetD  = DIM*jnrD;
 645
 646             /* load j atom coordinates */
 647             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 648                                               x+j_coord_offsetC,x+j_coord_offsetD,
 649                                               &jx0,&jy0,&jz0);
 650
 651             /* Calculate displacement vector */
 652             dx00             = _mm_sub_ps(ix0,jx0);
 653             dy00             = _mm_sub_ps(iy0,jy0);
 654             dz00             = _mm_sub_ps(iz0,jz0);
 655
 656             /* Calculate squared distance and things based on it */
 657             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 658
 659             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 660
 661             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 662
 663             /* Load parameters for j particles */
 664             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 665                                                               charge+jnrC+0,charge+jnrD+0);
 666             vdwjidx0A        = 2*vdwtype[jnrA+0];
 667             vdwjidx0B        = 2*vdwtype[jnrB+0];
 668             vdwjidx0C        = 2*vdwtype[jnrC+0];
 669             vdwjidx0D        = 2*vdwtype[jnrD+0];
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             r00              = _mm_mul_ps(rsq00,rinv00);
 676             r00              = _mm_andnot_ps(dummy_mask,r00);
 677
 678             /* Compute parameters for interactions between i and j atoms */
 679             qq00             = _mm_mul_ps(iq0,jq0);
 680             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 681                                          vdwparam+vdwioffset0+vdwjidx0B,
 682                                          vdwparam+vdwioffset0+vdwjidx0C,
 683                                          vdwparam+vdwioffset0+vdwjidx0D,
 684                                          &c6_00,&c12_00);
 685
 686             /* Calculate table index by multiplying r with table scale and truncate to integer */
 687             rt               = _mm_mul_ps(r00,vftabscale);
 688             vfitab           = _mm_cvttps_epi32(rt);
 689             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 690             vfitab           = _mm_slli_epi32(vfitab,2);
 691
 692             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 693             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 694             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 695             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 696             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 697             _MM_TRANSPOSE4_PS(Y,F,G,H);
 698             Heps             = _mm_mul_ps(vfeps,H);
 699             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 700             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 701             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 702
 703             /* LENNARD-JONES DISPERSION/REPULSION */
 704
 705             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 706             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 707
 708             fscal            = _mm_add_ps(felec,fvdw);
 709
 710             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 711
 712             /* Calculate temporary vectorial force */
 713             tx               = _mm_mul_ps(fscal,dx00);
 714             ty               = _mm_mul_ps(fscal,dy00);
 715             tz               = _mm_mul_ps(fscal,dz00);
 716
 717             /* Update vectorial force */
 718             fix0             = _mm_add_ps(fix0,tx);
 719             fiy0             = _mm_add_ps(fiy0,ty);
 720             fiz0             = _mm_add_ps(fiz0,tz);
 721
 722             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 723             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 724             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 725             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 726             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 727
 728             /* Inner loop uses 48 flops */
 729         }
 730
 731         /* End of innermost loop */
 732
 733         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 734                                               f+i_coord_offset,fshift+i_shift_offset);
 735
 736         /* Increment number of inner iterations */
 737         inneriter                  += j_index_end - j_index_start;
 738
 739         /* Outer loop uses 7 flops */
 740     }
 741
 742     /* Increment number of outer iterations */
 743     outeriter        += nri;
 744
 745     /* Update outer/inner flops */
 746
 747     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*48);
 748 }