src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_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 "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_ElecEw_VdwLJ_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_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     __m128i          ewitab;
  99     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     __m128           dummy_mask,cutoff_mask;
 102     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 103     __m128           one     = _mm_set1_ps(1.0);
 104     __m128           two     = _mm_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_ps(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162
 163         /* Load parameters for i particles */
 164         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 165         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 166
 167         /* Reset potential sums */
 168         velecsum         = _mm_setzero_ps();
 169         vvdwsum          = _mm_setzero_ps();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               x+j_coord_offsetC,x+j_coord_offsetD,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_ps(ix0,jx0);
 192             dy00             = _mm_sub_ps(iy0,jy0);
 193             dz00             = _mm_sub_ps(iz0,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 197
 198             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 199
 200             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 201
 202             /* Load parameters for j particles */
 203             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 204                                                               charge+jnrC+0,charge+jnrD+0);
 205             vdwjidx0A        = 2*vdwtype[jnrA+0];
 206             vdwjidx0B        = 2*vdwtype[jnrB+0];
 207             vdwjidx0C        = 2*vdwtype[jnrC+0];
 208             vdwjidx0D        = 2*vdwtype[jnrD+0];
 209
 210             /**************************
 211              * CALCULATE INTERACTIONS *
 212              **************************/
 213
 214             r00              = _mm_mul_ps(rsq00,rinv00);
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq00             = _mm_mul_ps(iq0,jq0);
 218             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 219                                          vdwparam+vdwioffset0+vdwjidx0B,
 220                                          vdwparam+vdwioffset0+vdwjidx0C,
 221                                          vdwparam+vdwioffset0+vdwjidx0D,
 222                                          &c6_00,&c12_00);
 223
 224             /* EWALD ELECTROSTATICS */
 225
 226             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 227             ewrt             = _mm_mul_ps(r00,ewtabscale);
 228             ewitab           = _mm_cvttps_epi32(ewrt);
 229             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 230             ewitab           = _mm_slli_epi32(ewitab,2);
 231             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 232             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 233             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 234             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 235             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 236             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 237             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 238             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 239             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 240
 241             /* LENNARD-JONES DISPERSION/REPULSION */
 242
 243             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 244             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 245             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 246             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 247             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             velecsum         = _mm_add_ps(velecsum,velec);
 251             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 252
 253             fscal            = _mm_add_ps(felec,fvdw);
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm_mul_ps(fscal,dx00);
 257             ty               = _mm_mul_ps(fscal,dy00);
 258             tz               = _mm_mul_ps(fscal,dz00);
 259
 260             /* Update vectorial force */
 261             fix0             = _mm_add_ps(fix0,tx);
 262             fiy0             = _mm_add_ps(fiy0,ty);
 263             fiz0             = _mm_add_ps(fiz0,tz);
 264
 265             fjptrA             = f+j_coord_offsetA;
 266             fjptrB             = f+j_coord_offsetB;
 267             fjptrC             = f+j_coord_offsetC;
 268             fjptrD             = f+j_coord_offsetD;
 269             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 270
 271             /* Inner loop uses 53 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             r00              = _mm_mul_ps(rsq00,rinv00);
 326             r00              = _mm_andnot_ps(dummy_mask,r00);
 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             /* EWALD ELECTROSTATICS */
 337
 338             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 339             ewrt             = _mm_mul_ps(r00,ewtabscale);
 340             ewitab           = _mm_cvttps_epi32(ewrt);
 341             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 342             ewitab           = _mm_slli_epi32(ewitab,2);
 343             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 344             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 345             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 346             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 347             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 348             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 349             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 350             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 351             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 352
 353             /* LENNARD-JONES DISPERSION/REPULSION */
 354
 355             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 356             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 357             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 358             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 359             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_andnot_ps(dummy_mask,velec);
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 365             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 366
 367             fscal            = _mm_add_ps(felec,fvdw);
 368
 369             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm_mul_ps(fscal,dx00);
 373             ty               = _mm_mul_ps(fscal,dy00);
 374             tz               = _mm_mul_ps(fscal,dz00);
 375
 376             /* Update vectorial force */
 377             fix0             = _mm_add_ps(fix0,tx);
 378             fiy0             = _mm_add_ps(fiy0,ty);
 379             fiz0             = _mm_add_ps(fiz0,tz);
 380
 381             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 382             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 383             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 384             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 385             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 386
 387             /* Inner loop uses 54 flops */
 388         }
 389
 390         /* End of innermost loop */
 391
 392         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 393                                               f+i_coord_offset,fshift+i_shift_offset);
 394
 395         ggid                        = gid[iidx];
 396         /* Update potential energies */
 397         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 398         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 399
 400         /* Increment number of inner iterations */
 401         inneriter                  += j_index_end - j_index_start;
 402
 403         /* Outer loop uses 9 flops */
 404     }
 405
 406     /* Increment number of outer iterations */
 407     outeriter        += nri;
 408
 409     /* Update outer/inner flops */
 410
 411     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*54);
 412 }
 413 /*
 414  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse2_single
 415  * Electrostatics interaction: Ewald
 416  * VdW interaction:            LennardJones
 417  * Geometry:                   Particle-Particle
 418  * Calculate force/pot:        Force
 419  */
 420 void
 421 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse2_single
 422                     (t_nblist                    * gmx_restrict       nlist,
 423                      rvec                        * gmx_restrict          xx,
 424                      rvec                        * gmx_restrict          ff,
 425                      t_forcerec                  * gmx_restrict          fr,
 426                      t_mdatoms                   * gmx_restrict     mdatoms,
 427                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 428                      t_nrnb                      * gmx_restrict        nrnb)
 429 {
 430     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 431      * just 0 for non-waters.
 432      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 433      * jnr indices corresponding to data put in the four positions in the SIMD register.
 434      */
 435     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 436     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 437     int              jnrA,jnrB,jnrC,jnrD;
 438     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 439     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 440     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 441     real             rcutoff_scalar;
 442     real             *shiftvec,*fshift,*x,*f;
 443     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 444     real             scratch[4*DIM];
 445     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 446     int              vdwioffset0;
 447     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 448     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 449     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 450     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 451     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 452     real             *charge;
 453     int              nvdwtype;
 454     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 455     int              *vdwtype;
 456     real             *vdwparam;
 457     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 458     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 459     __m128i          ewitab;
 460     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 461     real             *ewtab;
 462     __m128           dummy_mask,cutoff_mask;
 463     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 464     __m128           one     = _mm_set1_ps(1.0);
 465     __m128           two     = _mm_set1_ps(2.0);
 466     x                = xx[0];
 467     f                = ff[0];
 468
 469     nri              = nlist->nri;
 470     iinr             = nlist->iinr;
 471     jindex           = nlist->jindex;
 472     jjnr             = nlist->jjnr;
 473     shiftidx         = nlist->shift;
 474     gid              = nlist->gid;
 475     shiftvec         = fr->shift_vec[0];
 476     fshift           = fr->fshift[0];
 477     facel            = _mm_set1_ps(fr->epsfac);
 478     charge           = mdatoms->chargeA;
 479     nvdwtype         = fr->ntype;
 480     vdwparam         = fr->nbfp;
 481     vdwtype          = mdatoms->typeA;
 482
 483     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 484     ewtab            = fr->ic->tabq_coul_F;
 485     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 486     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_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             /* EWALD ELECTROSTATICS */
 582
 583             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 584             ewrt             = _mm_mul_ps(r00,ewtabscale);
 585             ewitab           = _mm_cvttps_epi32(ewrt);
 586             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 587             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 588                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 589                                          &ewtabF,&ewtabFn);
 590             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 591             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 592
 593             /* LENNARD-JONES DISPERSION/REPULSION */
 594
 595             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 596             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 597
 598             fscal            = _mm_add_ps(felec,fvdw);
 599
 600             /* Calculate temporary vectorial force */
 601             tx               = _mm_mul_ps(fscal,dx00);
 602             ty               = _mm_mul_ps(fscal,dy00);
 603             tz               = _mm_mul_ps(fscal,dz00);
 604
 605             /* Update vectorial force */
 606             fix0             = _mm_add_ps(fix0,tx);
 607             fiy0             = _mm_add_ps(fiy0,ty);
 608             fiz0             = _mm_add_ps(fiz0,tz);
 609
 610             fjptrA             = f+j_coord_offsetA;
 611             fjptrB             = f+j_coord_offsetB;
 612             fjptrC             = f+j_coord_offsetC;
 613             fjptrD             = f+j_coord_offsetD;
 614             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 615
 616             /* Inner loop uses 43 flops */
 617         }
 618
 619         if(jidx<j_index_end)
 620         {
 621
 622             /* Get j neighbor index, and coordinate index */
 623             jnrlistA         = jjnr[jidx];
 624             jnrlistB         = jjnr[jidx+1];
 625             jnrlistC         = jjnr[jidx+2];
 626             jnrlistD         = jjnr[jidx+3];
 627             /* Sign of each element will be negative for non-real atoms.
 628              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 629              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 630              */
 631             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 632             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 633             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 634             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 635             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 636             j_coord_offsetA  = DIM*jnrA;
 637             j_coord_offsetB  = DIM*jnrB;
 638             j_coord_offsetC  = DIM*jnrC;
 639             j_coord_offsetD  = DIM*jnrD;
 640
 641             /* load j atom coordinates */
 642             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 643                                               x+j_coord_offsetC,x+j_coord_offsetD,
 644                                               &jx0,&jy0,&jz0);
 645
 646             /* Calculate displacement vector */
 647             dx00             = _mm_sub_ps(ix0,jx0);
 648             dy00             = _mm_sub_ps(iy0,jy0);
 649             dz00             = _mm_sub_ps(iz0,jz0);
 650
 651             /* Calculate squared distance and things based on it */
 652             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 653
 654             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 655
 656             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 657
 658             /* Load parameters for j particles */
 659             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 660                                                               charge+jnrC+0,charge+jnrD+0);
 661             vdwjidx0A        = 2*vdwtype[jnrA+0];
 662             vdwjidx0B        = 2*vdwtype[jnrB+0];
 663             vdwjidx0C        = 2*vdwtype[jnrC+0];
 664             vdwjidx0D        = 2*vdwtype[jnrD+0];
 665
 666             /**************************
 667              * CALCULATE INTERACTIONS *
 668              **************************/
 669
 670             r00              = _mm_mul_ps(rsq00,rinv00);
 671             r00              = _mm_andnot_ps(dummy_mask,r00);
 672
 673             /* Compute parameters for interactions between i and j atoms */
 674             qq00             = _mm_mul_ps(iq0,jq0);
 675             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 676                                          vdwparam+vdwioffset0+vdwjidx0B,
 677                                          vdwparam+vdwioffset0+vdwjidx0C,
 678                                          vdwparam+vdwioffset0+vdwjidx0D,
 679                                          &c6_00,&c12_00);
 680
 681             /* EWALD ELECTROSTATICS */
 682
 683             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 684             ewrt             = _mm_mul_ps(r00,ewtabscale);
 685             ewitab           = _mm_cvttps_epi32(ewrt);
 686             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 687             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 688                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 689                                          &ewtabF,&ewtabFn);
 690             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 691             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 692
 693             /* LENNARD-JONES DISPERSION/REPULSION */
 694
 695             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 696             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 697
 698             fscal            = _mm_add_ps(felec,fvdw);
 699
 700             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 701
 702             /* Calculate temporary vectorial force */
 703             tx               = _mm_mul_ps(fscal,dx00);
 704             ty               = _mm_mul_ps(fscal,dy00);
 705             tz               = _mm_mul_ps(fscal,dz00);
 706
 707             /* Update vectorial force */
 708             fix0             = _mm_add_ps(fix0,tx);
 709             fiy0             = _mm_add_ps(fiy0,ty);
 710             fiz0             = _mm_add_ps(fiz0,tz);
 711
 712             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 713             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 714             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 715             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 716             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 717
 718             /* Inner loop uses 44 flops */
 719         }
 720
 721         /* End of innermost loop */
 722
 723         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 724                                               f+i_coord_offset,fshift+i_shift_offset);
 725
 726         /* Increment number of inner iterations */
 727         inneriter                  += j_index_end - j_index_start;
 728
 729         /* Outer loop uses 7 flops */
 730     }
 731
 732     /* Increment number of outer iterations */
 733     outeriter        += nri;
 734
 735     /* Update outer/inner flops */
 736
 737     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*44);
 738 }