src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJEw_GeomP1P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse2_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  94     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  95     __m128           c6grid_00;
  96     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  97     real             *vdwgridparam;
  98     __m128           one_half = _mm_set1_ps(0.5);
  99     __m128           minus_one = _mm_set1_ps(-1.0);
 100     __m128i          ewitab;
 101     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     real             *ewtab;
 103     __m128           dummy_mask,cutoff_mask;
 104     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 105     __m128           one     = _mm_set1_ps(1.0);
 106     __m128           two     = _mm_set1_ps(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_ps(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123     vdwgridparam     = fr->ljpme_c6grid;
 124     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 125     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 126     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 127
 128     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 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           = sse2_invsqrt_f(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             r00              = _mm_mul_ps(rsq00,rinv00);
 221
 222             /* Compute parameters for interactions between i and j atoms */
 223             qq00             = _mm_mul_ps(iq0,jq0);
 224             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 225                                          vdwparam+vdwioffset0+vdwjidx0B,
 226                                          vdwparam+vdwioffset0+vdwjidx0C,
 227                                          vdwparam+vdwioffset0+vdwjidx0D,
 228                                          &c6_00,&c12_00);
 229             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 230                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 231                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 232                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 233
 234             /* EWALD ELECTROSTATICS */
 235
 236             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 237             ewrt             = _mm_mul_ps(r00,ewtabscale);
 238             ewitab           = _mm_cvttps_epi32(ewrt);
 239             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 240             ewitab           = _mm_slli_epi32(ewitab,2);
 241             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 242             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 243             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 244             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 245             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 246             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 247             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 248             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 249             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 250
 251             /* Analytical LJ-PME */
 252             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 253             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 254             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 255             exponent         = sse2_exp_f(ewcljrsq);
 256             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 257             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 258             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 259             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 260             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 261             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 262             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 263             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             velecsum         = _mm_add_ps(velecsum,velec);
 267             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = _mm_add_ps(felec,fvdw);
 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             /* Inner loop uses 69 flops */
 288         }
 289
 290         if(jidx<j_index_end)
 291         {
 292
 293             /* Get j neighbor index, and coordinate index */
 294             jnrlistA         = jjnr[jidx];
 295             jnrlistB         = jjnr[jidx+1];
 296             jnrlistC         = jjnr[jidx+2];
 297             jnrlistD         = jjnr[jidx+3];
 298             /* Sign of each element will be negative for non-real atoms.
 299              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 300              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 301              */
 302             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 303             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 304             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 305             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 306             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 307             j_coord_offsetA  = DIM*jnrA;
 308             j_coord_offsetB  = DIM*jnrB;
 309             j_coord_offsetC  = DIM*jnrC;
 310             j_coord_offsetD  = DIM*jnrD;
 311
 312             /* load j atom coordinates */
 313             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 314                                               x+j_coord_offsetC,x+j_coord_offsetD,
 315                                               &jx0,&jy0,&jz0);
 316
 317             /* Calculate displacement vector */
 318             dx00             = _mm_sub_ps(ix0,jx0);
 319             dy00             = _mm_sub_ps(iy0,jy0);
 320             dz00             = _mm_sub_ps(iz0,jz0);
 321
 322             /* Calculate squared distance and things based on it */
 323             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 324
 325             rinv00           = sse2_invsqrt_f(rsq00);
 326
 327             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 328
 329             /* Load parameters for j particles */
 330             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 331                                                               charge+jnrC+0,charge+jnrD+0);
 332             vdwjidx0A        = 2*vdwtype[jnrA+0];
 333             vdwjidx0B        = 2*vdwtype[jnrB+0];
 334             vdwjidx0C        = 2*vdwtype[jnrC+0];
 335             vdwjidx0D        = 2*vdwtype[jnrD+0];
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             r00              = _mm_mul_ps(rsq00,rinv00);
 342             r00              = _mm_andnot_ps(dummy_mask,r00);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq00             = _mm_mul_ps(iq0,jq0);
 346             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 347                                          vdwparam+vdwioffset0+vdwjidx0B,
 348                                          vdwparam+vdwioffset0+vdwjidx0C,
 349                                          vdwparam+vdwioffset0+vdwjidx0D,
 350                                          &c6_00,&c12_00);
 351             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 352                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 353                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 354                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 355
 356             /* EWALD ELECTROSTATICS */
 357
 358             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 359             ewrt             = _mm_mul_ps(r00,ewtabscale);
 360             ewitab           = _mm_cvttps_epi32(ewrt);
 361             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 362             ewitab           = _mm_slli_epi32(ewitab,2);
 363             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 364             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 365             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 366             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 367             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 368             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 369             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 370             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 371             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 372
 373             /* Analytical LJ-PME */
 374             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 375             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 376             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 377             exponent         = sse2_exp_f(ewcljrsq);
 378             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 379             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 380             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 381             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 382             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 383             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 384             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 385             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm_andnot_ps(dummy_mask,velec);
 389             velecsum         = _mm_add_ps(velecsum,velec);
 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_andnot_ps(dummy_mask,fscal);
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm_mul_ps(fscal,dx00);
 399             ty               = _mm_mul_ps(fscal,dy00);
 400             tz               = _mm_mul_ps(fscal,dz00);
 401
 402             /* Update vectorial force */
 403             fix0             = _mm_add_ps(fix0,tx);
 404             fiy0             = _mm_add_ps(fiy0,ty);
 405             fiz0             = _mm_add_ps(fiz0,tz);
 406
 407             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 408             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 409             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 410             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 411             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 412
 413             /* Inner loop uses 70 flops */
 414         }
 415
 416         /* End of innermost loop */
 417
 418         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 419                                               f+i_coord_offset,fshift+i_shift_offset);
 420
 421         ggid                        = gid[iidx];
 422         /* Update potential energies */
 423         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 424         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 425
 426         /* Increment number of inner iterations */
 427         inneriter                  += j_index_end - j_index_start;
 428
 429         /* Outer loop uses 9 flops */
 430     }
 431
 432     /* Increment number of outer iterations */
 433     outeriter        += nri;
 434
 435     /* Update outer/inner flops */
 436
 437     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 438 }
 439 /*
 440  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_single
 441  * Electrostatics interaction: Ewald
 442  * VdW interaction:            LJEwald
 443  * Geometry:                   Particle-Particle
 444  * Calculate force/pot:        Force
 445  */
 446 void
 447 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse2_single
 448                     (t_nblist                    * gmx_restrict       nlist,
 449                      rvec                        * gmx_restrict          xx,
 450                      rvec                        * gmx_restrict          ff,
 451                      struct t_forcerec           * gmx_restrict          fr,
 452                      t_mdatoms                   * gmx_restrict     mdatoms,
 453                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 454                      t_nrnb                      * gmx_restrict        nrnb)
 455 {
 456     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 457      * just 0 for non-waters.
 458      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 459      * jnr indices corresponding to data put in the four positions in the SIMD register.
 460      */
 461     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 462     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 463     int              jnrA,jnrB,jnrC,jnrD;
 464     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 465     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 466     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 467     real             rcutoff_scalar;
 468     real             *shiftvec,*fshift,*x,*f;
 469     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 470     real             scratch[4*DIM];
 471     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 472     int              vdwioffset0;
 473     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 474     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 475     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 476     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 477     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 478     real             *charge;
 479     int              nvdwtype;
 480     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 481     int              *vdwtype;
 482     real             *vdwparam;
 483     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 484     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 485     __m128           c6grid_00;
 486     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 487     real             *vdwgridparam;
 488     __m128           one_half = _mm_set1_ps(0.5);
 489     __m128           minus_one = _mm_set1_ps(-1.0);
 490     __m128i          ewitab;
 491     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 492     real             *ewtab;
 493     __m128           dummy_mask,cutoff_mask;
 494     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 495     __m128           one     = _mm_set1_ps(1.0);
 496     __m128           two     = _mm_set1_ps(2.0);
 497     x                = xx[0];
 498     f                = ff[0];
 499
 500     nri              = nlist->nri;
 501     iinr             = nlist->iinr;
 502     jindex           = nlist->jindex;
 503     jjnr             = nlist->jjnr;
 504     shiftidx         = nlist->shift;
 505     gid              = nlist->gid;
 506     shiftvec         = fr->shift_vec[0];
 507     fshift           = fr->fshift[0];
 508     facel            = _mm_set1_ps(fr->ic->epsfac);
 509     charge           = mdatoms->chargeA;
 510     nvdwtype         = fr->ntype;
 511     vdwparam         = fr->nbfp;
 512     vdwtype          = mdatoms->typeA;
 513     vdwgridparam     = fr->ljpme_c6grid;
 514     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 515     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 516     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 517
 518     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 519     ewtab            = fr->ic->tabq_coul_F;
 520     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 521     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 522
 523     /* Avoid stupid compiler warnings */
 524     jnrA = jnrB = jnrC = jnrD = 0;
 525     j_coord_offsetA = 0;
 526     j_coord_offsetB = 0;
 527     j_coord_offsetC = 0;
 528     j_coord_offsetD = 0;
 529
 530     outeriter        = 0;
 531     inneriter        = 0;
 532
 533     for(iidx=0;iidx<4*DIM;iidx++)
 534     {
 535         scratch[iidx] = 0.0;
 536     }
 537
 538     /* Start outer loop over neighborlists */
 539     for(iidx=0; iidx<nri; iidx++)
 540     {
 541         /* Load shift vector for this list */
 542         i_shift_offset   = DIM*shiftidx[iidx];
 543
 544         /* Load limits for loop over neighbors */
 545         j_index_start    = jindex[iidx];
 546         j_index_end      = jindex[iidx+1];
 547
 548         /* Get outer coordinate index */
 549         inr              = iinr[iidx];
 550         i_coord_offset   = DIM*inr;
 551
 552         /* Load i particle coords and add shift vector */
 553         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 554
 555         fix0             = _mm_setzero_ps();
 556         fiy0             = _mm_setzero_ps();
 557         fiz0             = _mm_setzero_ps();
 558
 559         /* Load parameters for i particles */
 560         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 561         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 562
 563         /* Start inner kernel loop */
 564         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 565         {
 566
 567             /* Get j neighbor index, and coordinate index */
 568             jnrA             = jjnr[jidx];
 569             jnrB             = jjnr[jidx+1];
 570             jnrC             = jjnr[jidx+2];
 571             jnrD             = jjnr[jidx+3];
 572             j_coord_offsetA  = DIM*jnrA;
 573             j_coord_offsetB  = DIM*jnrB;
 574             j_coord_offsetC  = DIM*jnrC;
 575             j_coord_offsetD  = DIM*jnrD;
 576
 577             /* load j atom coordinates */
 578             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 579                                               x+j_coord_offsetC,x+j_coord_offsetD,
 580                                               &jx0,&jy0,&jz0);
 581
 582             /* Calculate displacement vector */
 583             dx00             = _mm_sub_ps(ix0,jx0);
 584             dy00             = _mm_sub_ps(iy0,jy0);
 585             dz00             = _mm_sub_ps(iz0,jz0);
 586
 587             /* Calculate squared distance and things based on it */
 588             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 589
 590             rinv00           = sse2_invsqrt_f(rsq00);
 591
 592             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 593
 594             /* Load parameters for j particles */
 595             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 596                                                               charge+jnrC+0,charge+jnrD+0);
 597             vdwjidx0A        = 2*vdwtype[jnrA+0];
 598             vdwjidx0B        = 2*vdwtype[jnrB+0];
 599             vdwjidx0C        = 2*vdwtype[jnrC+0];
 600             vdwjidx0D        = 2*vdwtype[jnrD+0];
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             r00              = _mm_mul_ps(rsq00,rinv00);
 607
 608             /* Compute parameters for interactions between i and j atoms */
 609             qq00             = _mm_mul_ps(iq0,jq0);
 610             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 611                                          vdwparam+vdwioffset0+vdwjidx0B,
 612                                          vdwparam+vdwioffset0+vdwjidx0C,
 613                                          vdwparam+vdwioffset0+vdwjidx0D,
 614                                          &c6_00,&c12_00);
 615             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 616                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 617                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 618                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 619
 620             /* EWALD ELECTROSTATICS */
 621
 622             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 623             ewrt             = _mm_mul_ps(r00,ewtabscale);
 624             ewitab           = _mm_cvttps_epi32(ewrt);
 625             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 626             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 627                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 628                                          &ewtabF,&ewtabFn);
 629             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 630             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 631
 632             /* Analytical LJ-PME */
 633             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 634             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 635             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 636             exponent         = sse2_exp_f(ewcljrsq);
 637             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 638             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 639             /* f6A = 6 * C6grid * (1 - poly) */
 640             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 641             /* f6B = C6grid * exponent * beta^6 */
 642             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 643             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 644             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 645
 646             fscal            = _mm_add_ps(felec,fvdw);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm_mul_ps(fscal,dx00);
 650             ty               = _mm_mul_ps(fscal,dy00);
 651             tz               = _mm_mul_ps(fscal,dz00);
 652
 653             /* Update vectorial force */
 654             fix0             = _mm_add_ps(fix0,tx);
 655             fiy0             = _mm_add_ps(fiy0,ty);
 656             fiz0             = _mm_add_ps(fiz0,tz);
 657
 658             fjptrA             = f+j_coord_offsetA;
 659             fjptrB             = f+j_coord_offsetB;
 660             fjptrC             = f+j_coord_offsetC;
 661             fjptrD             = f+j_coord_offsetD;
 662             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 663
 664             /* Inner loop uses 59 flops */
 665         }
 666
 667         if(jidx<j_index_end)
 668         {
 669
 670             /* Get j neighbor index, and coordinate index */
 671             jnrlistA         = jjnr[jidx];
 672             jnrlistB         = jjnr[jidx+1];
 673             jnrlistC         = jjnr[jidx+2];
 674             jnrlistD         = jjnr[jidx+3];
 675             /* Sign of each element will be negative for non-real atoms.
 676              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 677              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 678              */
 679             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 680             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 681             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 682             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 683             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 684             j_coord_offsetA  = DIM*jnrA;
 685             j_coord_offsetB  = DIM*jnrB;
 686             j_coord_offsetC  = DIM*jnrC;
 687             j_coord_offsetD  = DIM*jnrD;
 688
 689             /* load j atom coordinates */
 690             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 691                                               x+j_coord_offsetC,x+j_coord_offsetD,
 692                                               &jx0,&jy0,&jz0);
 693
 694             /* Calculate displacement vector */
 695             dx00             = _mm_sub_ps(ix0,jx0);
 696             dy00             = _mm_sub_ps(iy0,jy0);
 697             dz00             = _mm_sub_ps(iz0,jz0);
 698
 699             /* Calculate squared distance and things based on it */
 700             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 701
 702             rinv00           = sse2_invsqrt_f(rsq00);
 703
 704             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 705
 706             /* Load parameters for j particles */
 707             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 708                                                               charge+jnrC+0,charge+jnrD+0);
 709             vdwjidx0A        = 2*vdwtype[jnrA+0];
 710             vdwjidx0B        = 2*vdwtype[jnrB+0];
 711             vdwjidx0C        = 2*vdwtype[jnrC+0];
 712             vdwjidx0D        = 2*vdwtype[jnrD+0];
 713
 714             /**************************
 715              * CALCULATE INTERACTIONS *
 716              **************************/
 717
 718             r00              = _mm_mul_ps(rsq00,rinv00);
 719             r00              = _mm_andnot_ps(dummy_mask,r00);
 720
 721             /* Compute parameters for interactions between i and j atoms */
 722             qq00             = _mm_mul_ps(iq0,jq0);
 723             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 724                                          vdwparam+vdwioffset0+vdwjidx0B,
 725                                          vdwparam+vdwioffset0+vdwjidx0C,
 726                                          vdwparam+vdwioffset0+vdwjidx0D,
 727                                          &c6_00,&c12_00);
 728             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 729                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 730                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 731                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 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_cvtepi32_ps(ewitab));
 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             /* Analytical LJ-PME */
 746             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 747             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 748             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 749             exponent         = sse2_exp_f(ewcljrsq);
 750             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 751             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 752             /* f6A = 6 * C6grid * (1 - poly) */
 753             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 754             /* f6B = C6grid * exponent * beta^6 */
 755             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 756             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 757             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 758
 759             fscal            = _mm_add_ps(felec,fvdw);
 760
 761             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 762
 763             /* Calculate temporary vectorial force */
 764             tx               = _mm_mul_ps(fscal,dx00);
 765             ty               = _mm_mul_ps(fscal,dy00);
 766             tz               = _mm_mul_ps(fscal,dz00);
 767
 768             /* Update vectorial force */
 769             fix0             = _mm_add_ps(fix0,tx);
 770             fiy0             = _mm_add_ps(fiy0,ty);
 771             fiz0             = _mm_add_ps(fiz0,tz);
 772
 773             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 774             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 775             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 776             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 777             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 778
 779             /* Inner loop uses 60 flops */
 780         }
 781
 782         /* End of innermost loop */
 783
 784         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 785                                               f+i_coord_offset,fshift+i_shift_offset);
 786
 787         /* Increment number of inner iterations */
 788         inneriter                  += j_index_end - j_index_start;
 789
 790         /* Outer loop uses 7 flops */
 791     }
 792
 793     /* Increment number of outer iterations */
 794     outeriter        += nri;
 795
 796     /* Update outer/inner flops */
 797
 798     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 799 }