src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSh_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_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  69     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  78     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  79     __m128           dummy_mask,cutoff_mask;
  80     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  81     __m128           one     = _mm_set1_ps(1.0);
  82     __m128           two     = _mm_set1_ps(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm_set1_ps(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96     krf              = _mm_set1_ps(fr->ic->k_rf);
  97     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
  98     crf              = _mm_set1_ps(fr->ic->c_rf);
  99     nvdwtype         = fr->ntype;
 100     vdwparam         = fr->nbfp;
 101     vdwtype          = mdatoms->typeA;
 102
 103     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 104     rcutoff_scalar   = fr->rcoulomb;
 105     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 106     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 107
 108     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 109     rvdw             = _mm_set1_ps(fr->rvdw);
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = jnrC = jnrD = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115     j_coord_offsetC = 0;
 116     j_coord_offsetD = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 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         shX              = shiftvec[i_shift_offset+XX];
 127         shY              = shiftvec[i_shift_offset+YY];
 128         shZ              = shiftvec[i_shift_offset+ZZ];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 140         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 141         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 142
 143         fix0             = _mm_setzero_ps();
 144         fiy0             = _mm_setzero_ps();
 145         fiz0             = _mm_setzero_ps();
 146
 147         /* Load parameters for i particles */
 148         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 149         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 150
 151         /* Reset potential sums */
 152         velecsum         = _mm_setzero_ps();
 153         vvdwsum          = _mm_setzero_ps();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             jnrC             = jjnr[jidx+2];
 163             jnrD             = jjnr[jidx+3];
 164
 165             j_coord_offsetA  = DIM*jnrA;
 166             j_coord_offsetB  = DIM*jnrB;
 167             j_coord_offsetC  = DIM*jnrC;
 168             j_coord_offsetD  = DIM*jnrD;
 169
 170             /* load j atom coordinates */
 171             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                               x+j_coord_offsetC,x+j_coord_offsetD,
 173                                               &jx0,&jy0,&jz0);
 174
 175             /* Calculate displacement vector */
 176             dx00             = _mm_sub_ps(ix0,jx0);
 177             dy00             = _mm_sub_ps(iy0,jy0);
 178             dz00             = _mm_sub_ps(iz0,jz0);
 179
 180             /* Calculate squared distance and things based on it */
 181             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 182
 183             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 184
 185             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 186
 187             /* Load parameters for j particles */
 188             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 189                                                               charge+jnrC+0,charge+jnrD+0);
 190             vdwjidx0A        = 2*vdwtype[jnrA+0];
 191             vdwjidx0B        = 2*vdwtype[jnrB+0];
 192             vdwjidx0C        = 2*vdwtype[jnrC+0];
 193             vdwjidx0D        = 2*vdwtype[jnrD+0];
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             if (gmx_mm_any_lt(rsq00,rcutoff2))
 200             {
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             qq00             = _mm_mul_ps(iq0,jq0);
 204             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 205                                          vdwparam+vdwioffset0+vdwjidx0B,
 206                                          vdwparam+vdwioffset0+vdwjidx0C,
 207                                          vdwparam+vdwioffset0+vdwjidx0D,
 208                                          &c6_00,&c12_00);
 209
 210             /* REACTION-FIELD ELECTROSTATICS */
 211             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 212             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 213
 214             /* LENNARD-JONES DISPERSION/REPULSION */
 215
 216             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 217             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 218             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 219             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 220                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 221             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 222
 223             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 224
 225             /* Update potential sum for this i atom from the interaction with this j atom. */
 226             velec            = _mm_and_ps(velec,cutoff_mask);
 227             velecsum         = _mm_add_ps(velecsum,velec);
 228             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 229             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 230
 231             fscal            = _mm_add_ps(felec,fvdw);
 232
 233             fscal            = _mm_and_ps(fscal,cutoff_mask);
 234
 235             /* Calculate temporary vectorial force */
 236             tx               = _mm_mul_ps(fscal,dx00);
 237             ty               = _mm_mul_ps(fscal,dy00);
 238             tz               = _mm_mul_ps(fscal,dz00);
 239
 240             /* Update vectorial force */
 241             fix0             = _mm_add_ps(fix0,tx);
 242             fiy0             = _mm_add_ps(fiy0,ty);
 243             fiz0             = _mm_add_ps(fiz0,tz);
 244
 245             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 246                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 247                                                    tx,ty,tz);
 248
 249             }
 250
 251             /* Inner loop uses 54 flops */
 252         }
 253
 254         if(jidx<j_index_end)
 255         {
 256
 257             /* Get j neighbor index, and coordinate index */
 258             jnrA             = jjnr[jidx];
 259             jnrB             = jjnr[jidx+1];
 260             jnrC             = jjnr[jidx+2];
 261             jnrD             = jjnr[jidx+3];
 262
 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       = (jnrA>=0) ? jnrA : 0;
 269             jnrB       = (jnrB>=0) ? jnrB : 0;
 270             jnrC       = (jnrC>=0) ? jnrC : 0;
 271             jnrD       = (jnrD>=0) ? jnrD : 0;
 272
 273             j_coord_offsetA  = DIM*jnrA;
 274             j_coord_offsetB  = DIM*jnrB;
 275             j_coord_offsetC  = DIM*jnrC;
 276             j_coord_offsetD  = DIM*jnrD;
 277
 278             /* load j atom coordinates */
 279             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 280                                               x+j_coord_offsetC,x+j_coord_offsetD,
 281                                               &jx0,&jy0,&jz0);
 282
 283             /* Calculate displacement vector */
 284             dx00             = _mm_sub_ps(ix0,jx0);
 285             dy00             = _mm_sub_ps(iy0,jy0);
 286             dz00             = _mm_sub_ps(iz0,jz0);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 290
 291             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 292
 293             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 294
 295             /* Load parameters for j particles */
 296             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 297                                                               charge+jnrC+0,charge+jnrD+0);
 298             vdwjidx0A        = 2*vdwtype[jnrA+0];
 299             vdwjidx0B        = 2*vdwtype[jnrB+0];
 300             vdwjidx0C        = 2*vdwtype[jnrC+0];
 301             vdwjidx0D        = 2*vdwtype[jnrD+0];
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm_any_lt(rsq00,rcutoff2))
 308             {
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq00             = _mm_mul_ps(iq0,jq0);
 312             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 313                                          vdwparam+vdwioffset0+vdwjidx0B,
 314                                          vdwparam+vdwioffset0+vdwjidx0C,
 315                                          vdwparam+vdwioffset0+vdwjidx0D,
 316                                          &c6_00,&c12_00);
 317
 318             /* REACTION-FIELD ELECTROSTATICS */
 319             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 320             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 321
 322             /* LENNARD-JONES DISPERSION/REPULSION */
 323
 324             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 325             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 326             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 327             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 328                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 329             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 330
 331             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velec            = _mm_and_ps(velec,cutoff_mask);
 335             velec            = _mm_andnot_ps(dummy_mask,velec);
 336             velecsum         = _mm_add_ps(velecsum,velec);
 337             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 338             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 339             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 340
 341             fscal            = _mm_add_ps(felec,fvdw);
 342
 343             fscal            = _mm_and_ps(fscal,cutoff_mask);
 344
 345             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 346
 347             /* Calculate temporary vectorial force */
 348             tx               = _mm_mul_ps(fscal,dx00);
 349             ty               = _mm_mul_ps(fscal,dy00);
 350             tz               = _mm_mul_ps(fscal,dz00);
 351
 352             /* Update vectorial force */
 353             fix0             = _mm_add_ps(fix0,tx);
 354             fiy0             = _mm_add_ps(fiy0,ty);
 355             fiz0             = _mm_add_ps(fiz0,tz);
 356
 357             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 358                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 359                                                    tx,ty,tz);
 360
 361             }
 362
 363             /* Inner loop uses 54 flops */
 364         }
 365
 366         /* End of innermost loop */
 367
 368         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 369                                               f+i_coord_offset,fshift+i_shift_offset);
 370
 371         ggid                        = gid[iidx];
 372         /* Update potential energies */
 373         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 374         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 375
 376         /* Increment number of inner iterations */
 377         inneriter                  += j_index_end - j_index_start;
 378
 379         /* Outer loop uses 12 flops */
 380     }
 381
 382     /* Increment number of outer iterations */
 383     outeriter        += nri;
 384
 385     /* Update outer/inner flops */
 386
 387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*54);
 388 }
 389 /*
 390  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_single
 391  * Electrostatics interaction: ReactionField
 392  * VdW interaction:            LennardJones
 393  * Geometry:                   Particle-Particle
 394  * Calculate force/pot:        Force
 395  */
 396 void
 397 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_single
 398                     (t_nblist * gmx_restrict                nlist,
 399                      rvec * gmx_restrict                    xx,
 400                      rvec * gmx_restrict                    ff,
 401                      t_forcerec * gmx_restrict              fr,
 402                      t_mdatoms * gmx_restrict               mdatoms,
 403                      nb_kernel_data_t * gmx_restrict        kernel_data,
 404                      t_nrnb * gmx_restrict                  nrnb)
 405 {
 406     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 407      * just 0 for non-waters.
 408      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 409      * jnr indices corresponding to data put in the four positions in the SIMD register.
 410      */
 411     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 412     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 413     int              jnrA,jnrB,jnrC,jnrD;
 414     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 415     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 416     real             shX,shY,shZ,rcutoff_scalar;
 417     real             *shiftvec,*fshift,*x,*f;
 418     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 419     int              vdwioffset0;
 420     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 421     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 422     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 423     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 424     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 425     real             *charge;
 426     int              nvdwtype;
 427     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 428     int              *vdwtype;
 429     real             *vdwparam;
 430     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 431     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 432     __m128           dummy_mask,cutoff_mask;
 433     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 434     __m128           one     = _mm_set1_ps(1.0);
 435     __m128           two     = _mm_set1_ps(2.0);
 436     x                = xx[0];
 437     f                = ff[0];
 438
 439     nri              = nlist->nri;
 440     iinr             = nlist->iinr;
 441     jindex           = nlist->jindex;
 442     jjnr             = nlist->jjnr;
 443     shiftidx         = nlist->shift;
 444     gid              = nlist->gid;
 445     shiftvec         = fr->shift_vec[0];
 446     fshift           = fr->fshift[0];
 447     facel            = _mm_set1_ps(fr->epsfac);
 448     charge           = mdatoms->chargeA;
 449     krf              = _mm_set1_ps(fr->ic->k_rf);
 450     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 451     crf              = _mm_set1_ps(fr->ic->c_rf);
 452     nvdwtype         = fr->ntype;
 453     vdwparam         = fr->nbfp;
 454     vdwtype          = mdatoms->typeA;
 455
 456     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 457     rcutoff_scalar   = fr->rcoulomb;
 458     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 459     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 460
 461     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 462     rvdw             = _mm_set1_ps(fr->rvdw);
 463
 464     /* Avoid stupid compiler warnings */
 465     jnrA = jnrB = jnrC = jnrD = 0;
 466     j_coord_offsetA = 0;
 467     j_coord_offsetB = 0;
 468     j_coord_offsetC = 0;
 469     j_coord_offsetD = 0;
 470
 471     outeriter        = 0;
 472     inneriter        = 0;
 473
 474     /* Start outer loop over neighborlists */
 475     for(iidx=0; iidx<nri; iidx++)
 476     {
 477         /* Load shift vector for this list */
 478         i_shift_offset   = DIM*shiftidx[iidx];
 479         shX              = shiftvec[i_shift_offset+XX];
 480         shY              = shiftvec[i_shift_offset+YY];
 481         shZ              = shiftvec[i_shift_offset+ZZ];
 482
 483         /* Load limits for loop over neighbors */
 484         j_index_start    = jindex[iidx];
 485         j_index_end      = jindex[iidx+1];
 486
 487         /* Get outer coordinate index */
 488         inr              = iinr[iidx];
 489         i_coord_offset   = DIM*inr;
 490
 491         /* Load i particle coords and add shift vector */
 492         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 493         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 494         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 495
 496         fix0             = _mm_setzero_ps();
 497         fiy0             = _mm_setzero_ps();
 498         fiz0             = _mm_setzero_ps();
 499
 500         /* Load parameters for i particles */
 501         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 502         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 503
 504         /* Start inner kernel loop */
 505         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 506         {
 507
 508             /* Get j neighbor index, and coordinate index */
 509             jnrA             = jjnr[jidx];
 510             jnrB             = jjnr[jidx+1];
 511             jnrC             = jjnr[jidx+2];
 512             jnrD             = jjnr[jidx+3];
 513
 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             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 535
 536             /* Load parameters for j particles */
 537             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 538                                                               charge+jnrC+0,charge+jnrD+0);
 539             vdwjidx0A        = 2*vdwtype[jnrA+0];
 540             vdwjidx0B        = 2*vdwtype[jnrB+0];
 541             vdwjidx0C        = 2*vdwtype[jnrC+0];
 542             vdwjidx0D        = 2*vdwtype[jnrD+0];
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             if (gmx_mm_any_lt(rsq00,rcutoff2))
 549             {
 550
 551             /* Compute parameters for interactions between i and j atoms */
 552             qq00             = _mm_mul_ps(iq0,jq0);
 553             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 554                                          vdwparam+vdwioffset0+vdwjidx0B,
 555                                          vdwparam+vdwioffset0+vdwjidx0C,
 556                                          vdwparam+vdwioffset0+vdwjidx0D,
 557                                          &c6_00,&c12_00);
 558
 559             /* REACTION-FIELD ELECTROSTATICS */
 560             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 561
 562             /* LENNARD-JONES DISPERSION/REPULSION */
 563
 564             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 565             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 566
 567             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 568
 569             fscal            = _mm_add_ps(felec,fvdw);
 570
 571             fscal            = _mm_and_ps(fscal,cutoff_mask);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm_mul_ps(fscal,dx00);
 575             ty               = _mm_mul_ps(fscal,dy00);
 576             tz               = _mm_mul_ps(fscal,dz00);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_add_ps(fix0,tx);
 580             fiy0             = _mm_add_ps(fiy0,ty);
 581             fiz0             = _mm_add_ps(fiz0,tz);
 582
 583             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 584                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 585                                                    tx,ty,tz);
 586
 587             }
 588
 589             /* Inner loop uses 37 flops */
 590         }
 591
 592         if(jidx<j_index_end)
 593         {
 594
 595             /* Get j neighbor index, and coordinate index */
 596             jnrA             = jjnr[jidx];
 597             jnrB             = jjnr[jidx+1];
 598             jnrC             = jjnr[jidx+2];
 599             jnrD             = jjnr[jidx+3];
 600
 601             /* Sign of each element will be negative for non-real atoms.
 602              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 603              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 604              */
 605             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 606             jnrA       = (jnrA>=0) ? jnrA : 0;
 607             jnrB       = (jnrB>=0) ? jnrB : 0;
 608             jnrC       = (jnrC>=0) ? jnrC : 0;
 609             jnrD       = (jnrD>=0) ? jnrD : 0;
 610
 611             j_coord_offsetA  = DIM*jnrA;
 612             j_coord_offsetB  = DIM*jnrB;
 613             j_coord_offsetC  = DIM*jnrC;
 614             j_coord_offsetD  = DIM*jnrD;
 615
 616             /* load j atom coordinates */
 617             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 618                                               x+j_coord_offsetC,x+j_coord_offsetD,
 619                                               &jx0,&jy0,&jz0);
 620
 621             /* Calculate displacement vector */
 622             dx00             = _mm_sub_ps(ix0,jx0);
 623             dy00             = _mm_sub_ps(iy0,jy0);
 624             dz00             = _mm_sub_ps(iz0,jz0);
 625
 626             /* Calculate squared distance and things based on it */
 627             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 628
 629             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 630
 631             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 632
 633             /* Load parameters for j particles */
 634             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 635                                                               charge+jnrC+0,charge+jnrD+0);
 636             vdwjidx0A        = 2*vdwtype[jnrA+0];
 637             vdwjidx0B        = 2*vdwtype[jnrB+0];
 638             vdwjidx0C        = 2*vdwtype[jnrC+0];
 639             vdwjidx0D        = 2*vdwtype[jnrD+0];
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             if (gmx_mm_any_lt(rsq00,rcutoff2))
 646             {
 647
 648             /* Compute parameters for interactions between i and j atoms */
 649             qq00             = _mm_mul_ps(iq0,jq0);
 650             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 651                                          vdwparam+vdwioffset0+vdwjidx0B,
 652                                          vdwparam+vdwioffset0+vdwjidx0C,
 653                                          vdwparam+vdwioffset0+vdwjidx0D,
 654                                          &c6_00,&c12_00);
 655
 656             /* REACTION-FIELD ELECTROSTATICS */
 657             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 658
 659             /* LENNARD-JONES DISPERSION/REPULSION */
 660
 661             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 662             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 663
 664             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 665
 666             fscal            = _mm_add_ps(felec,fvdw);
 667
 668             fscal            = _mm_and_ps(fscal,cutoff_mask);
 669
 670             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 671
 672             /* Calculate temporary vectorial force */
 673             tx               = _mm_mul_ps(fscal,dx00);
 674             ty               = _mm_mul_ps(fscal,dy00);
 675             tz               = _mm_mul_ps(fscal,dz00);
 676
 677             /* Update vectorial force */
 678             fix0             = _mm_add_ps(fix0,tx);
 679             fiy0             = _mm_add_ps(fiy0,ty);
 680             fiz0             = _mm_add_ps(fiz0,tz);
 681
 682             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 683                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 684                                                    tx,ty,tz);
 685
 686             }
 687
 688             /* Inner loop uses 37 flops */
 689         }
 690
 691         /* End of innermost loop */
 692
 693         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 694                                               f+i_coord_offset,fshift+i_shift_offset);
 695
 696         /* Increment number of inner iterations */
 697         inneriter                  += j_index_end - j_index_start;
 698
 699         /* Outer loop uses 10 flops */
 700     }
 701
 702     /* Increment number of outer iterations */
 703     outeriter        += nri;
 704
 705     /* Update outer/inner flops */
 706
 707     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*37);
 708 }