src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128i          ewitab;
  97     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     real             *ewtab;
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 121     ewtab            = fr->ic->tabq_coul_FDV0;
 122     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 123     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 124
 125     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 126     rcutoff_scalar   = fr->rcoulomb;
 127     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 128     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 129
 130     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 131     rvdw             = _mm_set1_ps(fr->rvdw);
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168
 169         /* Load parameters for i particles */
 170         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 171         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm_setzero_ps();
 175         vvdwsum          = _mm_setzero_ps();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             jnrC             = jjnr[jidx+2];
 185             jnrD             = jjnr[jidx+3];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               x+j_coord_offsetC,x+j_coord_offsetD,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_ps(ix0,jx0);
 198             dy00             = _mm_sub_ps(iy0,jy0);
 199             dz00             = _mm_sub_ps(iz0,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 203
 204             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 205
 206             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 210                                                               charge+jnrC+0,charge+jnrD+0);
 211             vdwjidx0A        = 2*vdwtype[jnrA+0];
 212             vdwjidx0B        = 2*vdwtype[jnrB+0];
 213             vdwjidx0C        = 2*vdwtype[jnrC+0];
 214             vdwjidx0D        = 2*vdwtype[jnrD+0];
 215
 216             /**************************
 217              * CALCULATE INTERACTIONS *
 218              **************************/
 219
 220             if (gmx_mm_any_lt(rsq00,rcutoff2))
 221             {
 222
 223             r00              = _mm_mul_ps(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_ps(iq0,jq0);
 227             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 228                                          vdwparam+vdwioffset0+vdwjidx0B,
 229                                          vdwparam+vdwioffset0+vdwjidx0C,
 230                                          vdwparam+vdwioffset0+vdwjidx0D,
 231                                          &c6_00,&c12_00);
 232
 233             /* EWALD ELECTROSTATICS */
 234
 235             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 236             ewrt             = _mm_mul_ps(r00,ewtabscale);
 237             ewitab           = _mm_cvttps_epi32(ewrt);
 238             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 239             ewitab           = _mm_slli_epi32(ewitab,2);
 240             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 241             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 242             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 243             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 244             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 245             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 246             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 247             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 248             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 249
 250             /* LENNARD-JONES DISPERSION/REPULSION */
 251
 252             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 253             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 254             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 255             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) ,
 256                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 257             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 258
 259             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velec            = _mm_and_ps(velec,cutoff_mask);
 263             velecsum         = _mm_add_ps(velecsum,velec);
 264             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 265             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 266
 267             fscal            = _mm_add_ps(felec,fvdw);
 268
 269             fscal            = _mm_and_ps(fscal,cutoff_mask);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_ps(fscal,dx00);
 273             ty               = _mm_mul_ps(fscal,dy00);
 274             tz               = _mm_mul_ps(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_ps(fix0,tx);
 278             fiy0             = _mm_add_ps(fiy0,ty);
 279             fiz0             = _mm_add_ps(fiz0,tz);
 280
 281             fjptrA             = f+j_coord_offsetA;
 282             fjptrB             = f+j_coord_offsetB;
 283             fjptrC             = f+j_coord_offsetC;
 284             fjptrD             = f+j_coord_offsetD;
 285             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 286
 287             }
 288
 289             /* Inner loop uses 64 flops */
 290         }
 291
 292         if(jidx<j_index_end)
 293         {
 294
 295             /* Get j neighbor index, and coordinate index */
 296             jnrlistA         = jjnr[jidx];
 297             jnrlistB         = jjnr[jidx+1];
 298             jnrlistC         = jjnr[jidx+2];
 299             jnrlistD         = jjnr[jidx+3];
 300             /* Sign of each element will be negative for non-real atoms.
 301              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 302              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 303              */
 304             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 305             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 306             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 307             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 308             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 309             j_coord_offsetA  = DIM*jnrA;
 310             j_coord_offsetB  = DIM*jnrB;
 311             j_coord_offsetC  = DIM*jnrC;
 312             j_coord_offsetD  = DIM*jnrD;
 313
 314             /* load j atom coordinates */
 315             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 316                                               x+j_coord_offsetC,x+j_coord_offsetD,
 317                                               &jx0,&jy0,&jz0);
 318
 319             /* Calculate displacement vector */
 320             dx00             = _mm_sub_ps(ix0,jx0);
 321             dy00             = _mm_sub_ps(iy0,jy0);
 322             dz00             = _mm_sub_ps(iz0,jz0);
 323
 324             /* Calculate squared distance and things based on it */
 325             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 326
 327             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 328
 329             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 330
 331             /* Load parameters for j particles */
 332             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 333                                                               charge+jnrC+0,charge+jnrD+0);
 334             vdwjidx0A        = 2*vdwtype[jnrA+0];
 335             vdwjidx0B        = 2*vdwtype[jnrB+0];
 336             vdwjidx0C        = 2*vdwtype[jnrC+0];
 337             vdwjidx0D        = 2*vdwtype[jnrD+0];
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             if (gmx_mm_any_lt(rsq00,rcutoff2))
 344             {
 345
 346             r00              = _mm_mul_ps(rsq00,rinv00);
 347             r00              = _mm_andnot_ps(dummy_mask,r00);
 348
 349             /* Compute parameters for interactions between i and j atoms */
 350             qq00             = _mm_mul_ps(iq0,jq0);
 351             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 352                                          vdwparam+vdwioffset0+vdwjidx0B,
 353                                          vdwparam+vdwioffset0+vdwjidx0C,
 354                                          vdwparam+vdwioffset0+vdwjidx0D,
 355                                          &c6_00,&c12_00);
 356
 357             /* EWALD ELECTROSTATICS */
 358
 359             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 360             ewrt             = _mm_mul_ps(r00,ewtabscale);
 361             ewitab           = _mm_cvttps_epi32(ewrt);
 362             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 363             ewitab           = _mm_slli_epi32(ewitab,2);
 364             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 365             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 366             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 367             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 368             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 369             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 370             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 371             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 372             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 373
 374             /* LENNARD-JONES DISPERSION/REPULSION */
 375
 376             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 377             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 378             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 379             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) ,
 380                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 381             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 382
 383             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velec            = _mm_and_ps(velec,cutoff_mask);
 387             velec            = _mm_andnot_ps(dummy_mask,velec);
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 390             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 391             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 392
 393             fscal            = _mm_add_ps(felec,fvdw);
 394
 395             fscal            = _mm_and_ps(fscal,cutoff_mask);
 396
 397             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm_mul_ps(fscal,dx00);
 401             ty               = _mm_mul_ps(fscal,dy00);
 402             tz               = _mm_mul_ps(fscal,dz00);
 403
 404             /* Update vectorial force */
 405             fix0             = _mm_add_ps(fix0,tx);
 406             fiy0             = _mm_add_ps(fiy0,ty);
 407             fiz0             = _mm_add_ps(fiz0,tz);
 408
 409             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 410             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 411             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 412             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 413             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 414
 415             }
 416
 417             /* Inner loop uses 65 flops */
 418         }
 419
 420         /* End of innermost loop */
 421
 422         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 423                                               f+i_coord_offset,fshift+i_shift_offset);
 424
 425         ggid                        = gid[iidx];
 426         /* Update potential energies */
 427         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 428         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 429
 430         /* Increment number of inner iterations */
 431         inneriter                  += j_index_end - j_index_start;
 432
 433         /* Outer loop uses 9 flops */
 434     }
 435
 436     /* Increment number of outer iterations */
 437     outeriter        += nri;
 438
 439     /* Update outer/inner flops */
 440
 441     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*65);
 442 }
 443 /*
 444  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse2_single
 445  * Electrostatics interaction: Ewald
 446  * VdW interaction:            LennardJones
 447  * Geometry:                   Particle-Particle
 448  * Calculate force/pot:        Force
 449  */
 450 void
 451 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sse2_single
 452                     (t_nblist                    * gmx_restrict       nlist,
 453                      rvec                        * gmx_restrict          xx,
 454                      rvec                        * gmx_restrict          ff,
 455                      t_forcerec                  * gmx_restrict          fr,
 456                      t_mdatoms                   * gmx_restrict     mdatoms,
 457                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 458                      t_nrnb                      * gmx_restrict        nrnb)
 459 {
 460     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 461      * just 0 for non-waters.
 462      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 463      * jnr indices corresponding to data put in the four positions in the SIMD register.
 464      */
 465     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 466     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 467     int              jnrA,jnrB,jnrC,jnrD;
 468     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 469     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 470     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 471     real             rcutoff_scalar;
 472     real             *shiftvec,*fshift,*x,*f;
 473     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 474     real             scratch[4*DIM];
 475     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 476     int              vdwioffset0;
 477     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 478     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 479     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 480     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 481     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 482     real             *charge;
 483     int              nvdwtype;
 484     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 485     int              *vdwtype;
 486     real             *vdwparam;
 487     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 488     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 489     __m128i          ewitab;
 490     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 491     real             *ewtab;
 492     __m128           dummy_mask,cutoff_mask;
 493     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 494     __m128           one     = _mm_set1_ps(1.0);
 495     __m128           two     = _mm_set1_ps(2.0);
 496     x                = xx[0];
 497     f                = ff[0];
 498
 499     nri              = nlist->nri;
 500     iinr             = nlist->iinr;
 501     jindex           = nlist->jindex;
 502     jjnr             = nlist->jjnr;
 503     shiftidx         = nlist->shift;
 504     gid              = nlist->gid;
 505     shiftvec         = fr->shift_vec[0];
 506     fshift           = fr->fshift[0];
 507     facel            = _mm_set1_ps(fr->epsfac);
 508     charge           = mdatoms->chargeA;
 509     nvdwtype         = fr->ntype;
 510     vdwparam         = fr->nbfp;
 511     vdwtype          = mdatoms->typeA;
 512
 513     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 514     ewtab            = fr->ic->tabq_coul_F;
 515     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 516     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 517
 518     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 519     rcutoff_scalar   = fr->rcoulomb;
 520     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 521     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 522
 523     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 524     rvdw             = _mm_set1_ps(fr->rvdw);
 525
 526     /* Avoid stupid compiler warnings */
 527     jnrA = jnrB = jnrC = jnrD = 0;
 528     j_coord_offsetA = 0;
 529     j_coord_offsetB = 0;
 530     j_coord_offsetC = 0;
 531     j_coord_offsetD = 0;
 532
 533     outeriter        = 0;
 534     inneriter        = 0;
 535
 536     for(iidx=0;iidx<4*DIM;iidx++)
 537     {
 538         scratch[iidx] = 0.0;
 539     }
 540
 541     /* Start outer loop over neighborlists */
 542     for(iidx=0; iidx<nri; iidx++)
 543     {
 544         /* Load shift vector for this list */
 545         i_shift_offset   = DIM*shiftidx[iidx];
 546
 547         /* Load limits for loop over neighbors */
 548         j_index_start    = jindex[iidx];
 549         j_index_end      = jindex[iidx+1];
 550
 551         /* Get outer coordinate index */
 552         inr              = iinr[iidx];
 553         i_coord_offset   = DIM*inr;
 554
 555         /* Load i particle coords and add shift vector */
 556         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 557
 558         fix0             = _mm_setzero_ps();
 559         fiy0             = _mm_setzero_ps();
 560         fiz0             = _mm_setzero_ps();
 561
 562         /* Load parameters for i particles */
 563         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 564         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 565
 566         /* Start inner kernel loop */
 567         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 568         {
 569
 570             /* Get j neighbor index, and coordinate index */
 571             jnrA             = jjnr[jidx];
 572             jnrB             = jjnr[jidx+1];
 573             jnrC             = jjnr[jidx+2];
 574             jnrD             = jjnr[jidx+3];
 575             j_coord_offsetA  = DIM*jnrA;
 576             j_coord_offsetB  = DIM*jnrB;
 577             j_coord_offsetC  = DIM*jnrC;
 578             j_coord_offsetD  = DIM*jnrD;
 579
 580             /* load j atom coordinates */
 581             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 582                                               x+j_coord_offsetC,x+j_coord_offsetD,
 583                                               &jx0,&jy0,&jz0);
 584
 585             /* Calculate displacement vector */
 586             dx00             = _mm_sub_ps(ix0,jx0);
 587             dy00             = _mm_sub_ps(iy0,jy0);
 588             dz00             = _mm_sub_ps(iz0,jz0);
 589
 590             /* Calculate squared distance and things based on it */
 591             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 592
 593             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 594
 595             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 596
 597             /* Load parameters for j particles */
 598             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 599                                                               charge+jnrC+0,charge+jnrD+0);
 600             vdwjidx0A        = 2*vdwtype[jnrA+0];
 601             vdwjidx0B        = 2*vdwtype[jnrB+0];
 602             vdwjidx0C        = 2*vdwtype[jnrC+0];
 603             vdwjidx0D        = 2*vdwtype[jnrD+0];
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             if (gmx_mm_any_lt(rsq00,rcutoff2))
 610             {
 611
 612             r00              = _mm_mul_ps(rsq00,rinv00);
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq00             = _mm_mul_ps(iq0,jq0);
 616             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 617                                          vdwparam+vdwioffset0+vdwjidx0B,
 618                                          vdwparam+vdwioffset0+vdwjidx0C,
 619                                          vdwparam+vdwioffset0+vdwjidx0D,
 620                                          &c6_00,&c12_00);
 621
 622             /* EWALD ELECTROSTATICS */
 623
 624             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 625             ewrt             = _mm_mul_ps(r00,ewtabscale);
 626             ewitab           = _mm_cvttps_epi32(ewrt);
 627             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 628             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 629                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 630                                          &ewtabF,&ewtabFn);
 631             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 632             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 633
 634             /* LENNARD-JONES DISPERSION/REPULSION */
 635
 636             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 637             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 638
 639             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 640
 641             fscal            = _mm_add_ps(felec,fvdw);
 642
 643             fscal            = _mm_and_ps(fscal,cutoff_mask);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm_mul_ps(fscal,dx00);
 647             ty               = _mm_mul_ps(fscal,dy00);
 648             tz               = _mm_mul_ps(fscal,dz00);
 649
 650             /* Update vectorial force */
 651             fix0             = _mm_add_ps(fix0,tx);
 652             fiy0             = _mm_add_ps(fiy0,ty);
 653             fiz0             = _mm_add_ps(fiz0,tz);
 654
 655             fjptrA             = f+j_coord_offsetA;
 656             fjptrB             = f+j_coord_offsetB;
 657             fjptrC             = f+j_coord_offsetC;
 658             fjptrD             = f+j_coord_offsetD;
 659             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 660
 661             }
 662
 663             /* Inner loop uses 46 flops */
 664         }
 665
 666         if(jidx<j_index_end)
 667         {
 668
 669             /* Get j neighbor index, and coordinate index */
 670             jnrlistA         = jjnr[jidx];
 671             jnrlistB         = jjnr[jidx+1];
 672             jnrlistC         = jjnr[jidx+2];
 673             jnrlistD         = jjnr[jidx+3];
 674             /* Sign of each element will be negative for non-real atoms.
 675              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 676              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 677              */
 678             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 679             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 680             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 681             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 682             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 683             j_coord_offsetA  = DIM*jnrA;
 684             j_coord_offsetB  = DIM*jnrB;
 685             j_coord_offsetC  = DIM*jnrC;
 686             j_coord_offsetD  = DIM*jnrD;
 687
 688             /* load j atom coordinates */
 689             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 690                                               x+j_coord_offsetC,x+j_coord_offsetD,
 691                                               &jx0,&jy0,&jz0);
 692
 693             /* Calculate displacement vector */
 694             dx00             = _mm_sub_ps(ix0,jx0);
 695             dy00             = _mm_sub_ps(iy0,jy0);
 696             dz00             = _mm_sub_ps(iz0,jz0);
 697
 698             /* Calculate squared distance and things based on it */
 699             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 700
 701             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 702
 703             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 704
 705             /* Load parameters for j particles */
 706             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 707                                                               charge+jnrC+0,charge+jnrD+0);
 708             vdwjidx0A        = 2*vdwtype[jnrA+0];
 709             vdwjidx0B        = 2*vdwtype[jnrB+0];
 710             vdwjidx0C        = 2*vdwtype[jnrC+0];
 711             vdwjidx0D        = 2*vdwtype[jnrD+0];
 712
 713             /**************************
 714              * CALCULATE INTERACTIONS *
 715              **************************/
 716
 717             if (gmx_mm_any_lt(rsq00,rcutoff2))
 718             {
 719
 720             r00              = _mm_mul_ps(rsq00,rinv00);
 721             r00              = _mm_andnot_ps(dummy_mask,r00);
 722
 723             /* Compute parameters for interactions between i and j atoms */
 724             qq00             = _mm_mul_ps(iq0,jq0);
 725             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 726                                          vdwparam+vdwioffset0+vdwjidx0B,
 727                                          vdwparam+vdwioffset0+vdwjidx0C,
 728                                          vdwparam+vdwioffset0+vdwjidx0D,
 729                                          &c6_00,&c12_00);
 730
 731             /* EWALD ELECTROSTATICS */
 732
 733             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 734             ewrt             = _mm_mul_ps(r00,ewtabscale);
 735             ewitab           = _mm_cvttps_epi32(ewrt);
 736             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 737             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 738                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 739                                          &ewtabF,&ewtabFn);
 740             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 741             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 742
 743             /* LENNARD-JONES DISPERSION/REPULSION */
 744
 745             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 746             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 747
 748             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 749
 750             fscal            = _mm_add_ps(felec,fvdw);
 751
 752             fscal            = _mm_and_ps(fscal,cutoff_mask);
 753
 754             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 755
 756             /* Calculate temporary vectorial force */
 757             tx               = _mm_mul_ps(fscal,dx00);
 758             ty               = _mm_mul_ps(fscal,dy00);
 759             tz               = _mm_mul_ps(fscal,dz00);
 760
 761             /* Update vectorial force */
 762             fix0             = _mm_add_ps(fix0,tx);
 763             fiy0             = _mm_add_ps(fiy0,ty);
 764             fiz0             = _mm_add_ps(fiz0,tz);
 765
 766             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 767             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 768             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 769             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 770             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 771
 772             }
 773
 774             /* Inner loop uses 47 flops */
 775         }
 776
 777         /* End of innermost loop */
 778
 779         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 780                                               f+i_coord_offset,fshift+i_shift_offset);
 781
 782         /* Increment number of inner iterations */
 783         inneriter                  += j_index_end - j_index_start;
 784
 785         /* Outer loop uses 7 flops */
 786     }
 787
 788     /* Increment number of outer iterations */
 789     outeriter        += nri;
 790
 791     /* Update outer/inner flops */
 792
 793     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*47);
 794 }