src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJEw_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_ElecEw_VdwLJEw_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_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     __m128           c6grid_00;
  99     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 100     real             *vdwgridparam;
 101     __m128           one_half  = _mm_set1_ps(0.5);
 102     __m128           minus_one = _mm_set1_ps(-1.0);
 103     __m128i          ewitab;
 104     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     real             *ewtab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126     vdwgridparam     = fr->ljpme_c6grid;
 127     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 128     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 129     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 130
 131     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 132     ewtab            = fr->ic->tabq_coul_FDV0;
 133     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 134     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 167
 168         fix0             = _mm_setzero_ps();
 169         fiy0             = _mm_setzero_ps();
 170         fiz0             = _mm_setzero_ps();
 171
 172         /* Load parameters for i particles */
 173         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 174         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_ps();
 178         vvdwsum          = _mm_setzero_ps();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               x+j_coord_offsetC,x+j_coord_offsetD,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_ps(ix0,jx0);
 201             dy00             = _mm_sub_ps(iy0,jy0);
 202             dz00             = _mm_sub_ps(iz0,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 206
 207             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 208
 209             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                               charge+jnrC+0,charge+jnrD+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216             vdwjidx0C        = 2*vdwtype[jnrC+0];
 217             vdwjidx0D        = 2*vdwtype[jnrD+0];
 218
 219             /**************************
 220              * CALCULATE INTERACTIONS *
 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             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 234                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 235                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 236                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 237
 238             /* EWALD ELECTROSTATICS */
 239
 240             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 241             ewrt             = _mm_mul_ps(r00,ewtabscale);
 242             ewitab           = _mm_cvttps_epi32(ewrt);
 243             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 244             ewitab           = _mm_slli_epi32(ewitab,2);
 245             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 246             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 247             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 248             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 249             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 250             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 251             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 252             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 253             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 254
 255             /* Analytical LJ-PME */
 256             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 257             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 258             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 259             exponent         = gmx_simd_exp_r(ewcljrsq);
 260             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 261             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 262             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 263             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 264             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 265             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 266             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 267             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);
 268
 269             /* Update potential sum for this i atom from the interaction with this j atom. */
 270             velecsum         = _mm_add_ps(velecsum,velec);
 271             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 272
 273             fscal            = _mm_add_ps(felec,fvdw);
 274
 275             /* Calculate temporary vectorial force */
 276             tx               = _mm_mul_ps(fscal,dx00);
 277             ty               = _mm_mul_ps(fscal,dy00);
 278             tz               = _mm_mul_ps(fscal,dz00);
 279
 280             /* Update vectorial force */
 281             fix0             = _mm_add_ps(fix0,tx);
 282             fiy0             = _mm_add_ps(fiy0,ty);
 283             fiz0             = _mm_add_ps(fiz0,tz);
 284
 285             fjptrA             = f+j_coord_offsetA;
 286             fjptrB             = f+j_coord_offsetB;
 287             fjptrC             = f+j_coord_offsetC;
 288             fjptrD             = f+j_coord_offsetD;
 289             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 290
 291             /* Inner loop uses 69 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             r00              = _mm_mul_ps(rsq00,rinv00);
 346             r00              = _mm_andnot_ps(dummy_mask,r00);
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq00             = _mm_mul_ps(iq0,jq0);
 350             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 351                                          vdwparam+vdwioffset0+vdwjidx0B,
 352                                          vdwparam+vdwioffset0+vdwjidx0C,
 353                                          vdwparam+vdwioffset0+vdwjidx0D,
 354                                          &c6_00,&c12_00);
 355
 356             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 357                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 358                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 359                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 360
 361             /* EWALD ELECTROSTATICS */
 362
 363             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 364             ewrt             = _mm_mul_ps(r00,ewtabscale);
 365             ewitab           = _mm_cvttps_epi32(ewrt);
 366             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 367             ewitab           = _mm_slli_epi32(ewitab,2);
 368             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 369             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 370             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 371             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 372             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 373             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 374             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 375             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 376             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 377
 378             /* Analytical LJ-PME */
 379             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 380             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 381             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 382             exponent         = gmx_simd_exp_r(ewcljrsq);
 383             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 384             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 385             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 386             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 387             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 388             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 389             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 390             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);
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velec            = _mm_andnot_ps(dummy_mask,velec);
 394             velecsum         = _mm_add_ps(velecsum,velec);
 395             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 396             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 397
 398             fscal            = _mm_add_ps(felec,fvdw);
 399
 400             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 401
 402             /* Calculate temporary vectorial force */
 403             tx               = _mm_mul_ps(fscal,dx00);
 404             ty               = _mm_mul_ps(fscal,dy00);
 405             tz               = _mm_mul_ps(fscal,dz00);
 406
 407             /* Update vectorial force */
 408             fix0             = _mm_add_ps(fix0,tx);
 409             fiy0             = _mm_add_ps(fiy0,ty);
 410             fiz0             = _mm_add_ps(fiz0,tz);
 411
 412             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 413             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 414             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 415             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 416             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 417
 418             /* Inner loop uses 70 flops */
 419         }
 420
 421         /* End of innermost loop */
 422
 423         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 424                                               f+i_coord_offset,fshift+i_shift_offset);
 425
 426         ggid                        = gid[iidx];
 427         /* Update potential energies */
 428         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 429         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 430
 431         /* Increment number of inner iterations */
 432         inneriter                  += j_index_end - j_index_start;
 433
 434         /* Outer loop uses 9 flops */
 435     }
 436
 437     /* Increment number of outer iterations */
 438     outeriter        += nri;
 439
 440     /* Update outer/inner flops */
 441
 442     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 443 }
 444 /*
 445  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse4_1_single
 446  * Electrostatics interaction: Ewald
 447  * VdW interaction:            LJEwald
 448  * Geometry:                   Particle-Particle
 449  * Calculate force/pot:        Force
 450  */
 451 void
 452 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_sse4_1_single
 453                     (t_nblist                    * gmx_restrict       nlist,
 454                      rvec                        * gmx_restrict          xx,
 455                      rvec                        * gmx_restrict          ff,
 456                      t_forcerec                  * gmx_restrict          fr,
 457                      t_mdatoms                   * gmx_restrict     mdatoms,
 458                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 459                      t_nrnb                      * gmx_restrict        nrnb)
 460 {
 461     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 462      * just 0 for non-waters.
 463      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 464      * jnr indices corresponding to data put in the four positions in the SIMD register.
 465      */
 466     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 467     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 468     int              jnrA,jnrB,jnrC,jnrD;
 469     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 470     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 471     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 472     real             rcutoff_scalar;
 473     real             *shiftvec,*fshift,*x,*f;
 474     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 475     real             scratch[4*DIM];
 476     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 477     int              vdwioffset0;
 478     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 479     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 480     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 481     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 482     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 483     real             *charge;
 484     int              nvdwtype;
 485     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 486     int              *vdwtype;
 487     real             *vdwparam;
 488     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 489     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 490     __m128           c6grid_00;
 491     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 492     real             *vdwgridparam;
 493     __m128           one_half  = _mm_set1_ps(0.5);
 494     __m128           minus_one = _mm_set1_ps(-1.0);
 495     __m128i          ewitab;
 496     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 497     real             *ewtab;
 498     __m128           dummy_mask,cutoff_mask;
 499     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 500     __m128           one     = _mm_set1_ps(1.0);
 501     __m128           two     = _mm_set1_ps(2.0);
 502     x                = xx[0];
 503     f                = ff[0];
 504
 505     nri              = nlist->nri;
 506     iinr             = nlist->iinr;
 507     jindex           = nlist->jindex;
 508     jjnr             = nlist->jjnr;
 509     shiftidx         = nlist->shift;
 510     gid              = nlist->gid;
 511     shiftvec         = fr->shift_vec[0];
 512     fshift           = fr->fshift[0];
 513     facel            = _mm_set1_ps(fr->epsfac);
 514     charge           = mdatoms->chargeA;
 515     nvdwtype         = fr->ntype;
 516     vdwparam         = fr->nbfp;
 517     vdwtype          = mdatoms->typeA;
 518     vdwgridparam     = fr->ljpme_c6grid;
 519     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 520     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 521     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 522
 523     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 524     ewtab            = fr->ic->tabq_coul_F;
 525     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 526     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 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             r00              = _mm_mul_ps(rsq00,rinv00);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq00             = _mm_mul_ps(iq0,jq0);
 615             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 616                                          vdwparam+vdwioffset0+vdwjidx0B,
 617                                          vdwparam+vdwioffset0+vdwjidx0C,
 618                                          vdwparam+vdwioffset0+vdwjidx0D,
 619                                          &c6_00,&c12_00);
 620
 621             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 622                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 623                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 624                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 625
 626             /* EWALD ELECTROSTATICS */
 627
 628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 629             ewrt             = _mm_mul_ps(r00,ewtabscale);
 630             ewitab           = _mm_cvttps_epi32(ewrt);
 631             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 632             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 633                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 634                                          &ewtabF,&ewtabFn);
 635             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 636             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 637
 638             /* Analytical LJ-PME */
 639             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 640             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 641             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 642             exponent         = gmx_simd_exp_r(ewcljrsq);
 643             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 644             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 645             /* f6A = 6 * C6grid * (1 - poly) */
 646             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 647             /* f6B = C6grid * exponent * beta^6 */
 648             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 649             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 650             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);
 651
 652             fscal            = _mm_add_ps(felec,fvdw);
 653
 654             /* Calculate temporary vectorial force */
 655             tx               = _mm_mul_ps(fscal,dx00);
 656             ty               = _mm_mul_ps(fscal,dy00);
 657             tz               = _mm_mul_ps(fscal,dz00);
 658
 659             /* Update vectorial force */
 660             fix0             = _mm_add_ps(fix0,tx);
 661             fiy0             = _mm_add_ps(fiy0,ty);
 662             fiz0             = _mm_add_ps(fiz0,tz);
 663
 664             fjptrA             = f+j_coord_offsetA;
 665             fjptrB             = f+j_coord_offsetB;
 666             fjptrC             = f+j_coord_offsetC;
 667             fjptrD             = f+j_coord_offsetD;
 668             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 669
 670             /* Inner loop uses 59 flops */
 671         }
 672
 673         if(jidx<j_index_end)
 674         {
 675
 676             /* Get j neighbor index, and coordinate index */
 677             jnrlistA         = jjnr[jidx];
 678             jnrlistB         = jjnr[jidx+1];
 679             jnrlistC         = jjnr[jidx+2];
 680             jnrlistD         = jjnr[jidx+3];
 681             /* Sign of each element will be negative for non-real atoms.
 682              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 683              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 684              */
 685             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 686             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 687             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 688             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 689             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 690             j_coord_offsetA  = DIM*jnrA;
 691             j_coord_offsetB  = DIM*jnrB;
 692             j_coord_offsetC  = DIM*jnrC;
 693             j_coord_offsetD  = DIM*jnrD;
 694
 695             /* load j atom coordinates */
 696             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 697                                               x+j_coord_offsetC,x+j_coord_offsetD,
 698                                               &jx0,&jy0,&jz0);
 699
 700             /* Calculate displacement vector */
 701             dx00             = _mm_sub_ps(ix0,jx0);
 702             dy00             = _mm_sub_ps(iy0,jy0);
 703             dz00             = _mm_sub_ps(iz0,jz0);
 704
 705             /* Calculate squared distance and things based on it */
 706             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 707
 708             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 709
 710             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 711
 712             /* Load parameters for j particles */
 713             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 714                                                               charge+jnrC+0,charge+jnrD+0);
 715             vdwjidx0A        = 2*vdwtype[jnrA+0];
 716             vdwjidx0B        = 2*vdwtype[jnrB+0];
 717             vdwjidx0C        = 2*vdwtype[jnrC+0];
 718             vdwjidx0D        = 2*vdwtype[jnrD+0];
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             r00              = _mm_mul_ps(rsq00,rinv00);
 725             r00              = _mm_andnot_ps(dummy_mask,r00);
 726
 727             /* Compute parameters for interactions between i and j atoms */
 728             qq00             = _mm_mul_ps(iq0,jq0);
 729             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 730                                          vdwparam+vdwioffset0+vdwjidx0B,
 731                                          vdwparam+vdwioffset0+vdwjidx0C,
 732                                          vdwparam+vdwioffset0+vdwjidx0D,
 733                                          &c6_00,&c12_00);
 734
 735             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 736                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 737                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 738                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 739
 740             /* EWALD ELECTROSTATICS */
 741
 742             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 743             ewrt             = _mm_mul_ps(r00,ewtabscale);
 744             ewitab           = _mm_cvttps_epi32(ewrt);
 745             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 746             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 747                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 748                                          &ewtabF,&ewtabFn);
 749             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 750             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 751
 752             /* Analytical LJ-PME */
 753             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 754             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 755             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 756             exponent         = gmx_simd_exp_r(ewcljrsq);
 757             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 758             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 759             /* f6A = 6 * C6grid * (1 - poly) */
 760             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 761             /* f6B = C6grid * exponent * beta^6 */
 762             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 763             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 764             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);
 765
 766             fscal            = _mm_add_ps(felec,fvdw);
 767
 768             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 769
 770             /* Calculate temporary vectorial force */
 771             tx               = _mm_mul_ps(fscal,dx00);
 772             ty               = _mm_mul_ps(fscal,dy00);
 773             tz               = _mm_mul_ps(fscal,dz00);
 774
 775             /* Update vectorial force */
 776             fix0             = _mm_add_ps(fix0,tx);
 777             fiy0             = _mm_add_ps(fiy0,ty);
 778             fiz0             = _mm_add_ps(fiz0,tz);
 779
 780             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 781             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 782             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 783             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 784             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 785
 786             /* Inner loop uses 60 flops */
 787         }
 788
 789         /* End of innermost loop */
 790
 791         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 792                                               f+i_coord_offset,fshift+i_shift_offset);
 793
 794         /* Increment number of inner iterations */
 795         inneriter                  += j_index_end - j_index_start;
 796
 797         /* Outer loop uses 7 flops */
 798     }
 799
 800     /* Increment number of outer iterations */
 801     outeriter        += nri;
 802
 803     /* Update outer/inner flops */
 804
 805     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 806 }