src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_sse4_1_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 sse4_1_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_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sse4_1_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 130     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 131
 132     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 133     rvdw             = _mm_set1_ps(fr->rvdw);
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 166
 167         fix0             = _mm_setzero_ps();
 168         fiy0             = _mm_setzero_ps();
 169         fiz0             = _mm_setzero_ps();
 170
 171         /* Load parameters for i particles */
 172         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 173         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               x+j_coord_offsetC,x+j_coord_offsetD,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_ps(ix0,jx0);
 200             dy00             = _mm_sub_ps(iy0,jy0);
 201             dz00             = _mm_sub_ps(iz0,jz0);
 202
 203             /* Calculate squared distance and things based on it */
 204             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 205
 206             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 207
 208             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 209
 210             /* Load parameters for j particles */
 211             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 212                                                               charge+jnrC+0,charge+jnrD+0);
 213             vdwjidx0A        = 2*vdwtype[jnrA+0];
 214             vdwjidx0B        = 2*vdwtype[jnrB+0];
 215             vdwjidx0C        = 2*vdwtype[jnrC+0];
 216             vdwjidx0D        = 2*vdwtype[jnrD+0];
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             if (gmx_mm_any_lt(rsq00,rcutoff2))
 223             {
 224
 225             r00              = _mm_mul_ps(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_ps(iq0,jq0);
 229             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,
 231                                          vdwparam+vdwioffset0+vdwjidx0C,
 232                                          vdwparam+vdwioffset0+vdwjidx0D,
 233                                          &c6_00,&c12_00);
 234
 235             /* EWALD ELECTROSTATICS */
 236
 237             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 238             ewrt             = _mm_mul_ps(r00,ewtabscale);
 239             ewitab           = _mm_cvttps_epi32(ewrt);
 240             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 241             ewitab           = _mm_slli_epi32(ewitab,2);
 242             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 243             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 244             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 245             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 246             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 247             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 248             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 249             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 250             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 251
 252             /* LENNARD-JONES DISPERSION/REPULSION */
 253
 254             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 255             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 256             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 257             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) ,
 258                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 259             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 260
 261             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velec            = _mm_and_ps(velec,cutoff_mask);
 265             velecsum         = _mm_add_ps(velecsum,velec);
 266             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 267             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = _mm_add_ps(felec,fvdw);
 270
 271             fscal            = _mm_and_ps(fscal,cutoff_mask);
 272
 273             /* Calculate temporary vectorial force */
 274             tx               = _mm_mul_ps(fscal,dx00);
 275             ty               = _mm_mul_ps(fscal,dy00);
 276             tz               = _mm_mul_ps(fscal,dz00);
 277
 278             /* Update vectorial force */
 279             fix0             = _mm_add_ps(fix0,tx);
 280             fiy0             = _mm_add_ps(fiy0,ty);
 281             fiz0             = _mm_add_ps(fiz0,tz);
 282
 283             fjptrA             = f+j_coord_offsetA;
 284             fjptrB             = f+j_coord_offsetB;
 285             fjptrC             = f+j_coord_offsetC;
 286             fjptrD             = f+j_coord_offsetD;
 287             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 288
 289             }
 290
 291             /* Inner loop uses 64 flops */
 292         }
 293
 294         if(jidx<j_index_end)
 295         {
 296
 297             /* Get j neighbor index, and coordinate index */
 298             jnrlistA         = jjnr[jidx];
 299             jnrlistB         = jjnr[jidx+1];
 300             jnrlistC         = jjnr[jidx+2];
 301             jnrlistD         = jjnr[jidx+3];
 302             /* Sign of each element will be negative for non-real atoms.
 303              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 304              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 305              */
 306             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 307             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 308             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 309             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 310             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 311             j_coord_offsetA  = DIM*jnrA;
 312             j_coord_offsetB  = DIM*jnrB;
 313             j_coord_offsetC  = DIM*jnrC;
 314             j_coord_offsetD  = DIM*jnrD;
 315
 316             /* load j atom coordinates */
 317             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 318                                               x+j_coord_offsetC,x+j_coord_offsetD,
 319                                               &jx0,&jy0,&jz0);
 320
 321             /* Calculate displacement vector */
 322             dx00             = _mm_sub_ps(ix0,jx0);
 323             dy00             = _mm_sub_ps(iy0,jy0);
 324             dz00             = _mm_sub_ps(iz0,jz0);
 325
 326             /* Calculate squared distance and things based on it */
 327             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 328
 329             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 330
 331             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 332
 333             /* Load parameters for j particles */
 334             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 335                                                               charge+jnrC+0,charge+jnrD+0);
 336             vdwjidx0A        = 2*vdwtype[jnrA+0];
 337             vdwjidx0B        = 2*vdwtype[jnrB+0];
 338             vdwjidx0C        = 2*vdwtype[jnrC+0];
 339             vdwjidx0D        = 2*vdwtype[jnrD+0];
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             if (gmx_mm_any_lt(rsq00,rcutoff2))
 346             {
 347
 348             r00              = _mm_mul_ps(rsq00,rinv00);
 349             r00              = _mm_andnot_ps(dummy_mask,r00);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq00             = _mm_mul_ps(iq0,jq0);
 353             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 354                                          vdwparam+vdwioffset0+vdwjidx0B,
 355                                          vdwparam+vdwioffset0+vdwjidx0C,
 356                                          vdwparam+vdwioffset0+vdwjidx0D,
 357                                          &c6_00,&c12_00);
 358
 359             /* EWALD ELECTROSTATICS */
 360
 361             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 362             ewrt             = _mm_mul_ps(r00,ewtabscale);
 363             ewitab           = _mm_cvttps_epi32(ewrt);
 364             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 365             ewitab           = _mm_slli_epi32(ewitab,2);
 366             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 367             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 368             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 369             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 370             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 371             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 372             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 373             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 374             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 375
 376             /* LENNARD-JONES DISPERSION/REPULSION */
 377
 378             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 379             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 380             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 381             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) ,
 382                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 383             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 384
 385             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm_and_ps(velec,cutoff_mask);
 389             velec            = _mm_andnot_ps(dummy_mask,velec);
 390             velecsum         = _mm_add_ps(velecsum,velec);
 391             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 392             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 393             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 394
 395             fscal            = _mm_add_ps(felec,fvdw);
 396
 397             fscal            = _mm_and_ps(fscal,cutoff_mask);
 398
 399             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm_mul_ps(fscal,dx00);
 403             ty               = _mm_mul_ps(fscal,dy00);
 404             tz               = _mm_mul_ps(fscal,dz00);
 405
 406             /* Update vectorial force */
 407             fix0             = _mm_add_ps(fix0,tx);
 408             fiy0             = _mm_add_ps(fiy0,ty);
 409             fiz0             = _mm_add_ps(fiz0,tz);
 410
 411             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 412             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 413             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 414             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 415             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 416
 417             }
 418
 419             /* Inner loop uses 65 flops */
 420         }
 421
 422         /* End of innermost loop */
 423
 424         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 425                                               f+i_coord_offset,fshift+i_shift_offset);
 426
 427         ggid                        = gid[iidx];
 428         /* Update potential energies */
 429         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 430         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 431
 432         /* Increment number of inner iterations */
 433         inneriter                  += j_index_end - j_index_start;
 434
 435         /* Outer loop uses 9 flops */
 436     }
 437
 438     /* Increment number of outer iterations */
 439     outeriter        += nri;
 440
 441     /* Update outer/inner flops */
 442
 443     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*65);
 444 }
 445 /*
 446  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse4_1_single
 447  * Electrostatics interaction: Ewald
 448  * VdW interaction:            LennardJones
 449  * Geometry:                   Particle-Particle
 450  * Calculate force/pot:        Force
 451  */
 452 void
 453 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse4_1_single
 454                     (t_nblist                    * gmx_restrict       nlist,
 455                      rvec                        * gmx_restrict          xx,
 456                      rvec                        * gmx_restrict          ff,
 457                      t_forcerec                  * gmx_restrict          fr,
 458                      t_mdatoms                   * gmx_restrict     mdatoms,
 459                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 460                      t_nrnb                      * gmx_restrict        nrnb)
 461 {
 462     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 463      * just 0 for non-waters.
 464      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 465      * jnr indices corresponding to data put in the four positions in the SIMD register.
 466      */
 467     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 468     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 469     int              jnrA,jnrB,jnrC,jnrD;
 470     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 471     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 472     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 473     real             rcutoff_scalar;
 474     real             *shiftvec,*fshift,*x,*f;
 475     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 476     real             scratch[4*DIM];
 477     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 478     int              vdwioffset0;
 479     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 480     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 481     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 482     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 483     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 484     real             *charge;
 485     int              nvdwtype;
 486     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 487     int              *vdwtype;
 488     real             *vdwparam;
 489     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 490     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 491     __m128i          ewitab;
 492     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 493     real             *ewtab;
 494     __m128           dummy_mask,cutoff_mask;
 495     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 496     __m128           one     = _mm_set1_ps(1.0);
 497     __m128           two     = _mm_set1_ps(2.0);
 498     x                = xx[0];
 499     f                = ff[0];
 500
 501     nri              = nlist->nri;
 502     iinr             = nlist->iinr;
 503     jindex           = nlist->jindex;
 504     jjnr             = nlist->jjnr;
 505     shiftidx         = nlist->shift;
 506     gid              = nlist->gid;
 507     shiftvec         = fr->shift_vec[0];
 508     fshift           = fr->fshift[0];
 509     facel            = _mm_set1_ps(fr->epsfac);
 510     charge           = mdatoms->chargeA;
 511     nvdwtype         = fr->ntype;
 512     vdwparam         = fr->nbfp;
 513     vdwtype          = mdatoms->typeA;
 514
 515     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 516     ewtab            = fr->ic->tabq_coul_F;
 517     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 518     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 519
 520     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 521     rcutoff_scalar   = fr->rcoulomb;
 522     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 523     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 524
 525     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 526     rvdw             = _mm_set1_ps(fr->rvdw);
 527
 528     /* Avoid stupid compiler warnings */
 529     jnrA = jnrB = jnrC = jnrD = 0;
 530     j_coord_offsetA = 0;
 531     j_coord_offsetB = 0;
 532     j_coord_offsetC = 0;
 533     j_coord_offsetD = 0;
 534
 535     outeriter        = 0;
 536     inneriter        = 0;
 537
 538     for(iidx=0;iidx<4*DIM;iidx++)
 539     {
 540         scratch[iidx] = 0.0;
 541     }
 542
 543     /* Start outer loop over neighborlists */
 544     for(iidx=0; iidx<nri; iidx++)
 545     {
 546         /* Load shift vector for this list */
 547         i_shift_offset   = DIM*shiftidx[iidx];
 548
 549         /* Load limits for loop over neighbors */
 550         j_index_start    = jindex[iidx];
 551         j_index_end      = jindex[iidx+1];
 552
 553         /* Get outer coordinate index */
 554         inr              = iinr[iidx];
 555         i_coord_offset   = DIM*inr;
 556
 557         /* Load i particle coords and add shift vector */
 558         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 559
 560         fix0             = _mm_setzero_ps();
 561         fiy0             = _mm_setzero_ps();
 562         fiz0             = _mm_setzero_ps();
 563
 564         /* Load parameters for i particles */
 565         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 566         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 567
 568         /* Start inner kernel loop */
 569         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 570         {
 571
 572             /* Get j neighbor index, and coordinate index */
 573             jnrA             = jjnr[jidx];
 574             jnrB             = jjnr[jidx+1];
 575             jnrC             = jjnr[jidx+2];
 576             jnrD             = jjnr[jidx+3];
 577             j_coord_offsetA  = DIM*jnrA;
 578             j_coord_offsetB  = DIM*jnrB;
 579             j_coord_offsetC  = DIM*jnrC;
 580             j_coord_offsetD  = DIM*jnrD;
 581
 582             /* load j atom coordinates */
 583             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 584                                               x+j_coord_offsetC,x+j_coord_offsetD,
 585                                               &jx0,&jy0,&jz0);
 586
 587             /* Calculate displacement vector */
 588             dx00             = _mm_sub_ps(ix0,jx0);
 589             dy00             = _mm_sub_ps(iy0,jy0);
 590             dz00             = _mm_sub_ps(iz0,jz0);
 591
 592             /* Calculate squared distance and things based on it */
 593             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 594
 595             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 596
 597             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 598
 599             /* Load parameters for j particles */
 600             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 601                                                               charge+jnrC+0,charge+jnrD+0);
 602             vdwjidx0A        = 2*vdwtype[jnrA+0];
 603             vdwjidx0B        = 2*vdwtype[jnrB+0];
 604             vdwjidx0C        = 2*vdwtype[jnrC+0];
 605             vdwjidx0D        = 2*vdwtype[jnrD+0];
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             if (gmx_mm_any_lt(rsq00,rcutoff2))
 612             {
 613
 614             r00              = _mm_mul_ps(rsq00,rinv00);
 615
 616             /* Compute parameters for interactions between i and j atoms */
 617             qq00             = _mm_mul_ps(iq0,jq0);
 618             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 619                                          vdwparam+vdwioffset0+vdwjidx0B,
 620                                          vdwparam+vdwioffset0+vdwjidx0C,
 621                                          vdwparam+vdwioffset0+vdwjidx0D,
 622                                          &c6_00,&c12_00);
 623
 624             /* EWALD ELECTROSTATICS */
 625
 626             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 627             ewrt             = _mm_mul_ps(r00,ewtabscale);
 628             ewitab           = _mm_cvttps_epi32(ewrt);
 629             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 630             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 631                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 632                                          &ewtabF,&ewtabFn);
 633             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 634             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 635
 636             /* LENNARD-JONES DISPERSION/REPULSION */
 637
 638             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 639             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 640
 641             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 642
 643             fscal            = _mm_add_ps(felec,fvdw);
 644
 645             fscal            = _mm_and_ps(fscal,cutoff_mask);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm_mul_ps(fscal,dx00);
 649             ty               = _mm_mul_ps(fscal,dy00);
 650             tz               = _mm_mul_ps(fscal,dz00);
 651
 652             /* Update vectorial force */
 653             fix0             = _mm_add_ps(fix0,tx);
 654             fiy0             = _mm_add_ps(fiy0,ty);
 655             fiz0             = _mm_add_ps(fiz0,tz);
 656
 657             fjptrA             = f+j_coord_offsetA;
 658             fjptrB             = f+j_coord_offsetB;
 659             fjptrC             = f+j_coord_offsetC;
 660             fjptrD             = f+j_coord_offsetD;
 661             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 662
 663             }
 664
 665             /* Inner loop uses 46 flops */
 666         }
 667
 668         if(jidx<j_index_end)
 669         {
 670
 671             /* Get j neighbor index, and coordinate index */
 672             jnrlistA         = jjnr[jidx];
 673             jnrlistB         = jjnr[jidx+1];
 674             jnrlistC         = jjnr[jidx+2];
 675             jnrlistD         = jjnr[jidx+3];
 676             /* Sign of each element will be negative for non-real atoms.
 677              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 678              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 679              */
 680             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 681             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 682             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 683             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 684             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 685             j_coord_offsetA  = DIM*jnrA;
 686             j_coord_offsetB  = DIM*jnrB;
 687             j_coord_offsetC  = DIM*jnrC;
 688             j_coord_offsetD  = DIM*jnrD;
 689
 690             /* load j atom coordinates */
 691             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 692                                               x+j_coord_offsetC,x+j_coord_offsetD,
 693                                               &jx0,&jy0,&jz0);
 694
 695             /* Calculate displacement vector */
 696             dx00             = _mm_sub_ps(ix0,jx0);
 697             dy00             = _mm_sub_ps(iy0,jy0);
 698             dz00             = _mm_sub_ps(iz0,jz0);
 699
 700             /* Calculate squared distance and things based on it */
 701             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 702
 703             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 704
 705             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 706
 707             /* Load parameters for j particles */
 708             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 709                                                               charge+jnrC+0,charge+jnrD+0);
 710             vdwjidx0A        = 2*vdwtype[jnrA+0];
 711             vdwjidx0B        = 2*vdwtype[jnrB+0];
 712             vdwjidx0C        = 2*vdwtype[jnrC+0];
 713             vdwjidx0D        = 2*vdwtype[jnrD+0];
 714
 715             /**************************
 716              * CALCULATE INTERACTIONS *
 717              **************************/
 718
 719             if (gmx_mm_any_lt(rsq00,rcutoff2))
 720             {
 721
 722             r00              = _mm_mul_ps(rsq00,rinv00);
 723             r00              = _mm_andnot_ps(dummy_mask,r00);
 724
 725             /* Compute parameters for interactions between i and j atoms */
 726             qq00             = _mm_mul_ps(iq0,jq0);
 727             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 728                                          vdwparam+vdwioffset0+vdwjidx0B,
 729                                          vdwparam+vdwioffset0+vdwjidx0C,
 730                                          vdwparam+vdwioffset0+vdwjidx0D,
 731                                          &c6_00,&c12_00);
 732
 733             /* EWALD ELECTROSTATICS */
 734
 735             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 736             ewrt             = _mm_mul_ps(r00,ewtabscale);
 737             ewitab           = _mm_cvttps_epi32(ewrt);
 738             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 739             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 740                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 741                                          &ewtabF,&ewtabFn);
 742             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 743             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 744
 745             /* LENNARD-JONES DISPERSION/REPULSION */
 746
 747             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 748             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 749
 750             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 751
 752             fscal            = _mm_add_ps(felec,fvdw);
 753
 754             fscal            = _mm_and_ps(fscal,cutoff_mask);
 755
 756             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 757
 758             /* Calculate temporary vectorial force */
 759             tx               = _mm_mul_ps(fscal,dx00);
 760             ty               = _mm_mul_ps(fscal,dy00);
 761             tz               = _mm_mul_ps(fscal,dz00);
 762
 763             /* Update vectorial force */
 764             fix0             = _mm_add_ps(fix0,tx);
 765             fiy0             = _mm_add_ps(fiy0,ty);
 766             fiz0             = _mm_add_ps(fiz0,tz);
 767
 768             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 769             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 770             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 771             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 772             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 773
 774             }
 775
 776             /* Inner loop uses 47 flops */
 777         }
 778
 779         /* End of innermost loop */
 780
 781         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 782                                               f+i_coord_offset,fshift+i_shift_offset);
 783
 784         /* Increment number of inner iterations */
 785         inneriter                  += j_index_end - j_index_start;
 786
 787         /* Outer loop uses 7 flops */
 788     }
 789
 790     /* Increment number of outer iterations */
 791     outeriter        += nri;
 792
 793     /* Update outer/inner flops */
 794
 795     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*47);
 796 }