src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSh_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm_set1_ps(fr->ic->k_rf);
 116     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 117     crf              = _mm_set1_ps(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 126
 127     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 128     rvdw             = _mm_set1_ps(fr->rvdw);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165
 166         /* Load parameters for i particles */
 167         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 168         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_ps();
 172         vvdwsum          = _mm_setzero_ps();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             jnrC             = jjnr[jidx+2];
 182             jnrD             = jjnr[jidx+3];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               x+j_coord_offsetC,x+j_coord_offsetD,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_ps(ix0,jx0);
 195             dy00             = _mm_sub_ps(iy0,jy0);
 196             dz00             = _mm_sub_ps(iz0,jz0);
 197
 198             /* Calculate squared distance and things based on it */
 199             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 200
 201             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 202
 203             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 204
 205             /* Load parameters for j particles */
 206             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 207                                                               charge+jnrC+0,charge+jnrD+0);
 208             vdwjidx0A        = 2*vdwtype[jnrA+0];
 209             vdwjidx0B        = 2*vdwtype[jnrB+0];
 210             vdwjidx0C        = 2*vdwtype[jnrC+0];
 211             vdwjidx0D        = 2*vdwtype[jnrD+0];
 212
 213             /**************************
 214              * CALCULATE INTERACTIONS *
 215              **************************/
 216
 217             if (gmx_mm_any_lt(rsq00,rcutoff2))
 218             {
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             qq00             = _mm_mul_ps(iq0,jq0);
 222             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 223                                          vdwparam+vdwioffset0+vdwjidx0B,
 224                                          vdwparam+vdwioffset0+vdwjidx0C,
 225                                          vdwparam+vdwioffset0+vdwjidx0D,
 226                                          &c6_00,&c12_00);
 227
 228             /* REACTION-FIELD ELECTROSTATICS */
 229             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 230             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 231
 232             /* LENNARD-JONES DISPERSION/REPULSION */
 233
 234             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 235             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 236             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 237             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) ,
 238                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 239             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 240
 241             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velec            = _mm_and_ps(velec,cutoff_mask);
 245             velecsum         = _mm_add_ps(velecsum,velec);
 246             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 247             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 248
 249             fscal            = _mm_add_ps(felec,fvdw);
 250
 251             fscal            = _mm_and_ps(fscal,cutoff_mask);
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_ps(fscal,dx00);
 255             ty               = _mm_mul_ps(fscal,dy00);
 256             tz               = _mm_mul_ps(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_ps(fix0,tx);
 260             fiy0             = _mm_add_ps(fiy0,ty);
 261             fiz0             = _mm_add_ps(fiz0,tz);
 262
 263             fjptrA             = f+j_coord_offsetA;
 264             fjptrB             = f+j_coord_offsetB;
 265             fjptrC             = f+j_coord_offsetC;
 266             fjptrD             = f+j_coord_offsetD;
 267             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 268
 269             }
 270
 271             /* Inner loop uses 54 flops */
 272         }
 273
 274         if(jidx<j_index_end)
 275         {
 276
 277             /* Get j neighbor index, and coordinate index */
 278             jnrlistA         = jjnr[jidx];
 279             jnrlistB         = jjnr[jidx+1];
 280             jnrlistC         = jjnr[jidx+2];
 281             jnrlistD         = jjnr[jidx+3];
 282             /* Sign of each element will be negative for non-real atoms.
 283              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 284              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 285              */
 286             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 287             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 288             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 289             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 290             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 291             j_coord_offsetA  = DIM*jnrA;
 292             j_coord_offsetB  = DIM*jnrB;
 293             j_coord_offsetC  = DIM*jnrC;
 294             j_coord_offsetD  = DIM*jnrD;
 295
 296             /* load j atom coordinates */
 297             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 298                                               x+j_coord_offsetC,x+j_coord_offsetD,
 299                                               &jx0,&jy0,&jz0);
 300
 301             /* Calculate displacement vector */
 302             dx00             = _mm_sub_ps(ix0,jx0);
 303             dy00             = _mm_sub_ps(iy0,jy0);
 304             dz00             = _mm_sub_ps(iz0,jz0);
 305
 306             /* Calculate squared distance and things based on it */
 307             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 308
 309             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 310
 311             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 312
 313             /* Load parameters for j particles */
 314             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 315                                                               charge+jnrC+0,charge+jnrD+0);
 316             vdwjidx0A        = 2*vdwtype[jnrA+0];
 317             vdwjidx0B        = 2*vdwtype[jnrB+0];
 318             vdwjidx0C        = 2*vdwtype[jnrC+0];
 319             vdwjidx0D        = 2*vdwtype[jnrD+0];
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             if (gmx_mm_any_lt(rsq00,rcutoff2))
 326             {
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq00             = _mm_mul_ps(iq0,jq0);
 330             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 331                                          vdwparam+vdwioffset0+vdwjidx0B,
 332                                          vdwparam+vdwioffset0+vdwjidx0C,
 333                                          vdwparam+vdwioffset0+vdwjidx0D,
 334                                          &c6_00,&c12_00);
 335
 336             /* REACTION-FIELD ELECTROSTATICS */
 337             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 338             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 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( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 346                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 347             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 348
 349             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velec            = _mm_and_ps(velec,cutoff_mask);
 353             velec            = _mm_andnot_ps(dummy_mask,velec);
 354             velecsum         = _mm_add_ps(velecsum,velec);
 355             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 356             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 357             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 358
 359             fscal            = _mm_add_ps(felec,fvdw);
 360
 361             fscal            = _mm_and_ps(fscal,cutoff_mask);
 362
 363             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm_mul_ps(fscal,dx00);
 367             ty               = _mm_mul_ps(fscal,dy00);
 368             tz               = _mm_mul_ps(fscal,dz00);
 369
 370             /* Update vectorial force */
 371             fix0             = _mm_add_ps(fix0,tx);
 372             fiy0             = _mm_add_ps(fiy0,ty);
 373             fiz0             = _mm_add_ps(fiz0,tz);
 374
 375             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 376             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 377             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 378             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 379             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 380
 381             }
 382
 383             /* Inner loop uses 54 flops */
 384         }
 385
 386         /* End of innermost loop */
 387
 388         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 389                                               f+i_coord_offset,fshift+i_shift_offset);
 390
 391         ggid                        = gid[iidx];
 392         /* Update potential energies */
 393         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 394         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 395
 396         /* Increment number of inner iterations */
 397         inneriter                  += j_index_end - j_index_start;
 398
 399         /* Outer loop uses 9 flops */
 400     }
 401
 402     /* Increment number of outer iterations */
 403     outeriter        += nri;
 404
 405     /* Update outer/inner flops */
 406
 407     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 408 }
 409 /*
 410  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_single
 411  * Electrostatics interaction: ReactionField
 412  * VdW interaction:            LennardJones
 413  * Geometry:                   Particle-Particle
 414  * Calculate force/pot:        Force
 415  */
 416 void
 417 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_sse2_single
 418                     (t_nblist                    * gmx_restrict       nlist,
 419                      rvec                        * gmx_restrict          xx,
 420                      rvec                        * gmx_restrict          ff,
 421                      t_forcerec                  * gmx_restrict          fr,
 422                      t_mdatoms                   * gmx_restrict     mdatoms,
 423                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 424                      t_nrnb                      * gmx_restrict        nrnb)
 425 {
 426     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 427      * just 0 for non-waters.
 428      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 429      * jnr indices corresponding to data put in the four positions in the SIMD register.
 430      */
 431     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 432     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 433     int              jnrA,jnrB,jnrC,jnrD;
 434     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 435     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 436     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 437     real             rcutoff_scalar;
 438     real             *shiftvec,*fshift,*x,*f;
 439     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 440     real             scratch[4*DIM];
 441     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 442     int              vdwioffset0;
 443     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 444     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 445     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 446     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 447     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 448     real             *charge;
 449     int              nvdwtype;
 450     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 451     int              *vdwtype;
 452     real             *vdwparam;
 453     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 454     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 455     __m128           dummy_mask,cutoff_mask;
 456     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 457     __m128           one     = _mm_set1_ps(1.0);
 458     __m128           two     = _mm_set1_ps(2.0);
 459     x                = xx[0];
 460     f                = ff[0];
 461
 462     nri              = nlist->nri;
 463     iinr             = nlist->iinr;
 464     jindex           = nlist->jindex;
 465     jjnr             = nlist->jjnr;
 466     shiftidx         = nlist->shift;
 467     gid              = nlist->gid;
 468     shiftvec         = fr->shift_vec[0];
 469     fshift           = fr->fshift[0];
 470     facel            = _mm_set1_ps(fr->epsfac);
 471     charge           = mdatoms->chargeA;
 472     krf              = _mm_set1_ps(fr->ic->k_rf);
 473     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 474     crf              = _mm_set1_ps(fr->ic->c_rf);
 475     nvdwtype         = fr->ntype;
 476     vdwparam         = fr->nbfp;
 477     vdwtype          = mdatoms->typeA;
 478
 479     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 480     rcutoff_scalar   = fr->rcoulomb;
 481     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 482     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 483
 484     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 485     rvdw             = _mm_set1_ps(fr->rvdw);
 486
 487     /* Avoid stupid compiler warnings */
 488     jnrA = jnrB = jnrC = jnrD = 0;
 489     j_coord_offsetA = 0;
 490     j_coord_offsetB = 0;
 491     j_coord_offsetC = 0;
 492     j_coord_offsetD = 0;
 493
 494     outeriter        = 0;
 495     inneriter        = 0;
 496
 497     for(iidx=0;iidx<4*DIM;iidx++)
 498     {
 499         scratch[iidx] = 0.0;
 500     }
 501
 502     /* Start outer loop over neighborlists */
 503     for(iidx=0; iidx<nri; iidx++)
 504     {
 505         /* Load shift vector for this list */
 506         i_shift_offset   = DIM*shiftidx[iidx];
 507
 508         /* Load limits for loop over neighbors */
 509         j_index_start    = jindex[iidx];
 510         j_index_end      = jindex[iidx+1];
 511
 512         /* Get outer coordinate index */
 513         inr              = iinr[iidx];
 514         i_coord_offset   = DIM*inr;
 515
 516         /* Load i particle coords and add shift vector */
 517         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 518
 519         fix0             = _mm_setzero_ps();
 520         fiy0             = _mm_setzero_ps();
 521         fiz0             = _mm_setzero_ps();
 522
 523         /* Load parameters for i particles */
 524         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 525         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 526
 527         /* Start inner kernel loop */
 528         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 529         {
 530
 531             /* Get j neighbor index, and coordinate index */
 532             jnrA             = jjnr[jidx];
 533             jnrB             = jjnr[jidx+1];
 534             jnrC             = jjnr[jidx+2];
 535             jnrD             = jjnr[jidx+3];
 536             j_coord_offsetA  = DIM*jnrA;
 537             j_coord_offsetB  = DIM*jnrB;
 538             j_coord_offsetC  = DIM*jnrC;
 539             j_coord_offsetD  = DIM*jnrD;
 540
 541             /* load j atom coordinates */
 542             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 543                                               x+j_coord_offsetC,x+j_coord_offsetD,
 544                                               &jx0,&jy0,&jz0);
 545
 546             /* Calculate displacement vector */
 547             dx00             = _mm_sub_ps(ix0,jx0);
 548             dy00             = _mm_sub_ps(iy0,jy0);
 549             dz00             = _mm_sub_ps(iz0,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 553
 554             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 555
 556             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 557
 558             /* Load parameters for j particles */
 559             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 560                                                               charge+jnrC+0,charge+jnrD+0);
 561             vdwjidx0A        = 2*vdwtype[jnrA+0];
 562             vdwjidx0B        = 2*vdwtype[jnrB+0];
 563             vdwjidx0C        = 2*vdwtype[jnrC+0];
 564             vdwjidx0D        = 2*vdwtype[jnrD+0];
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             if (gmx_mm_any_lt(rsq00,rcutoff2))
 571             {
 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             /* REACTION-FIELD ELECTROSTATICS */
 582             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 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             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 590
 591             fscal            = _mm_add_ps(felec,fvdw);
 592
 593             fscal            = _mm_and_ps(fscal,cutoff_mask);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_ps(fscal,dx00);
 597             ty               = _mm_mul_ps(fscal,dy00);
 598             tz               = _mm_mul_ps(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm_add_ps(fix0,tx);
 602             fiy0             = _mm_add_ps(fiy0,ty);
 603             fiz0             = _mm_add_ps(fiz0,tz);
 604
 605             fjptrA             = f+j_coord_offsetA;
 606             fjptrB             = f+j_coord_offsetB;
 607             fjptrC             = f+j_coord_offsetC;
 608             fjptrD             = f+j_coord_offsetD;
 609             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 610
 611             }
 612
 613             /* Inner loop uses 37 flops */
 614         }
 615
 616         if(jidx<j_index_end)
 617         {
 618
 619             /* Get j neighbor index, and coordinate index */
 620             jnrlistA         = jjnr[jidx];
 621             jnrlistB         = jjnr[jidx+1];
 622             jnrlistC         = jjnr[jidx+2];
 623             jnrlistD         = jjnr[jidx+3];
 624             /* Sign of each element will be negative for non-real atoms.
 625              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 626              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 627              */
 628             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 629             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 630             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 631             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 632             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 633             j_coord_offsetA  = DIM*jnrA;
 634             j_coord_offsetB  = DIM*jnrB;
 635             j_coord_offsetC  = DIM*jnrC;
 636             j_coord_offsetD  = DIM*jnrD;
 637
 638             /* load j atom coordinates */
 639             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 640                                               x+j_coord_offsetC,x+j_coord_offsetD,
 641                                               &jx0,&jy0,&jz0);
 642
 643             /* Calculate displacement vector */
 644             dx00             = _mm_sub_ps(ix0,jx0);
 645             dy00             = _mm_sub_ps(iy0,jy0);
 646             dz00             = _mm_sub_ps(iz0,jz0);
 647
 648             /* Calculate squared distance and things based on it */
 649             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 650
 651             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 652
 653             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 654
 655             /* Load parameters for j particles */
 656             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 657                                                               charge+jnrC+0,charge+jnrD+0);
 658             vdwjidx0A        = 2*vdwtype[jnrA+0];
 659             vdwjidx0B        = 2*vdwtype[jnrB+0];
 660             vdwjidx0C        = 2*vdwtype[jnrC+0];
 661             vdwjidx0D        = 2*vdwtype[jnrD+0];
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             if (gmx_mm_any_lt(rsq00,rcutoff2))
 668             {
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             qq00             = _mm_mul_ps(iq0,jq0);
 672             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 673                                          vdwparam+vdwioffset0+vdwjidx0B,
 674                                          vdwparam+vdwioffset0+vdwjidx0C,
 675                                          vdwparam+vdwioffset0+vdwjidx0D,
 676                                          &c6_00,&c12_00);
 677
 678             /* REACTION-FIELD ELECTROSTATICS */
 679             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 680
 681             /* LENNARD-JONES DISPERSION/REPULSION */
 682
 683             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 684             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 685
 686             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 687
 688             fscal            = _mm_add_ps(felec,fvdw);
 689
 690             fscal            = _mm_and_ps(fscal,cutoff_mask);
 691
 692             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 693
 694             /* Calculate temporary vectorial force */
 695             tx               = _mm_mul_ps(fscal,dx00);
 696             ty               = _mm_mul_ps(fscal,dy00);
 697             tz               = _mm_mul_ps(fscal,dz00);
 698
 699             /* Update vectorial force */
 700             fix0             = _mm_add_ps(fix0,tx);
 701             fiy0             = _mm_add_ps(fiy0,ty);
 702             fiz0             = _mm_add_ps(fiz0,tz);
 703
 704             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 705             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 706             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 707             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 708             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 709
 710             }
 711
 712             /* Inner loop uses 37 flops */
 713         }
 714
 715         /* End of innermost loop */
 716
 717         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 718                                               f+i_coord_offset,fshift+i_shift_offset);
 719
 720         /* Increment number of inner iterations */
 721         inneriter                  += j_index_end - j_index_start;
 722
 723         /* Outer loop uses 7 flops */
 724     }
 725
 726     /* Increment number of outer iterations */
 727     outeriter        += nri;
 728
 729     /* Update outer/inner flops */
 730
 731     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*37);
 732 }