src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJ_GeomP1P1_avx_256_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 avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_avx_256_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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 100     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 101     __m256i          ewitab;
 102     __m128i          ewitab_lo,ewitab_hi;
 103     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 105     real             *ewtab;
 106     __m256           dummy_mask,cutoff_mask;
 107     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 108     __m256           one     = _mm256_set1_ps(1.0);
 109     __m256           two     = _mm256_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            = _mm256_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 128     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 129     beta2            = _mm256_mul_ps(beta,beta);
 130     beta3            = _mm256_mul_ps(beta,beta2);
 131
 132     ewtab            = fr->ic->tabq_coul_FDV0;
 133     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 134     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142     j_coord_offsetE = 0;
 143     j_coord_offsetF = 0;
 144     j_coord_offsetG = 0;
 145     j_coord_offsetH = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 171
 172         fix0             = _mm256_setzero_ps();
 173         fiy0             = _mm256_setzero_ps();
 174         fiz0             = _mm256_setzero_ps();
 175
 176         /* Load parameters for i particles */
 177         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 178         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm256_setzero_ps();
 182         vvdwsum          = _mm256_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             jnrE             = jjnr[jidx+4];
 194             jnrF             = jjnr[jidx+5];
 195             jnrG             = jjnr[jidx+6];
 196             jnrH             = jjnr[jidx+7];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201             j_coord_offsetE  = DIM*jnrE;
 202             j_coord_offsetF  = DIM*jnrF;
 203             j_coord_offsetG  = DIM*jnrG;
 204             j_coord_offsetH  = DIM*jnrH;
 205
 206             /* load j atom coordinates */
 207             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 209                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 210                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_ps(ix0,jx0);
 215             dy00             = _mm256_sub_ps(iy0,jy0);
 216             dz00             = _mm256_sub_ps(iz0,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 220
 221             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 222
 223             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                                  charge+jnrC+0,charge+jnrD+0,
 228                                                                  charge+jnrE+0,charge+jnrF+0,
 229                                                                  charge+jnrG+0,charge+jnrH+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234             vdwjidx0E        = 2*vdwtype[jnrE+0];
 235             vdwjidx0F        = 2*vdwtype[jnrF+0];
 236             vdwjidx0G        = 2*vdwtype[jnrG+0];
 237             vdwjidx0H        = 2*vdwtype[jnrH+0];
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm256_mul_ps(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm256_mul_ps(iq0,jq0);
 247             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 248                                             vdwioffsetptr0+vdwjidx0B,
 249                                             vdwioffsetptr0+vdwjidx0C,
 250                                             vdwioffsetptr0+vdwjidx0D,
 251                                             vdwioffsetptr0+vdwjidx0E,
 252                                             vdwioffsetptr0+vdwjidx0F,
 253                                             vdwioffsetptr0+vdwjidx0G,
 254                                             vdwioffsetptr0+vdwjidx0H,
 255                                             &c6_00,&c12_00);
 256
 257             /* EWALD ELECTROSTATICS */
 258
 259             /* Analytical PME correction */
 260             zeta2            = _mm256_mul_ps(beta2,rsq00);
 261             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 262             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 263             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 264             felec            = _mm256_mul_ps(qq00,felec);
 265             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 266             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 267             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 268             velec            = _mm256_mul_ps(qq00,velec);
 269
 270             /* LENNARD-JONES DISPERSION/REPULSION */
 271
 272             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 273             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 274             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 275             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 276             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             velecsum         = _mm256_add_ps(velecsum,velec);
 280             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 281
 282             fscal            = _mm256_add_ps(felec,fvdw);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm256_mul_ps(fscal,dx00);
 286             ty               = _mm256_mul_ps(fscal,dy00);
 287             tz               = _mm256_mul_ps(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm256_add_ps(fix0,tx);
 291             fiy0             = _mm256_add_ps(fiy0,ty);
 292             fiz0             = _mm256_add_ps(fiz0,tz);
 293
 294             fjptrA             = f+j_coord_offsetA;
 295             fjptrB             = f+j_coord_offsetB;
 296             fjptrC             = f+j_coord_offsetC;
 297             fjptrD             = f+j_coord_offsetD;
 298             fjptrE             = f+j_coord_offsetE;
 299             fjptrF             = f+j_coord_offsetF;
 300             fjptrG             = f+j_coord_offsetG;
 301             fjptrH             = f+j_coord_offsetH;
 302             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 303
 304             /* Inner loop uses 96 flops */
 305         }
 306
 307         if(jidx<j_index_end)
 308         {
 309
 310             /* Get j neighbor index, and coordinate index */
 311             jnrlistA         = jjnr[jidx];
 312             jnrlistB         = jjnr[jidx+1];
 313             jnrlistC         = jjnr[jidx+2];
 314             jnrlistD         = jjnr[jidx+3];
 315             jnrlistE         = jjnr[jidx+4];
 316             jnrlistF         = jjnr[jidx+5];
 317             jnrlistG         = jjnr[jidx+6];
 318             jnrlistH         = jjnr[jidx+7];
 319             /* Sign of each element will be negative for non-real atoms.
 320              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 321              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 322              */
 323             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 324                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 325
 326             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 327             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 328             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 329             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 330             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 331             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 332             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 333             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 334             j_coord_offsetA  = DIM*jnrA;
 335             j_coord_offsetB  = DIM*jnrB;
 336             j_coord_offsetC  = DIM*jnrC;
 337             j_coord_offsetD  = DIM*jnrD;
 338             j_coord_offsetE  = DIM*jnrE;
 339             j_coord_offsetF  = DIM*jnrF;
 340             j_coord_offsetG  = DIM*jnrG;
 341             j_coord_offsetH  = DIM*jnrH;
 342
 343             /* load j atom coordinates */
 344             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 345                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 346                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 347                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 348                                                  &jx0,&jy0,&jz0);
 349
 350             /* Calculate displacement vector */
 351             dx00             = _mm256_sub_ps(ix0,jx0);
 352             dy00             = _mm256_sub_ps(iy0,jy0);
 353             dz00             = _mm256_sub_ps(iz0,jz0);
 354
 355             /* Calculate squared distance and things based on it */
 356             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 357
 358             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 359
 360             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 361
 362             /* Load parameters for j particles */
 363             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 364                                                                  charge+jnrC+0,charge+jnrD+0,
 365                                                                  charge+jnrE+0,charge+jnrF+0,
 366                                                                  charge+jnrG+0,charge+jnrH+0);
 367             vdwjidx0A        = 2*vdwtype[jnrA+0];
 368             vdwjidx0B        = 2*vdwtype[jnrB+0];
 369             vdwjidx0C        = 2*vdwtype[jnrC+0];
 370             vdwjidx0D        = 2*vdwtype[jnrD+0];
 371             vdwjidx0E        = 2*vdwtype[jnrE+0];
 372             vdwjidx0F        = 2*vdwtype[jnrF+0];
 373             vdwjidx0G        = 2*vdwtype[jnrG+0];
 374             vdwjidx0H        = 2*vdwtype[jnrH+0];
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             r00              = _mm256_mul_ps(rsq00,rinv00);
 381             r00              = _mm256_andnot_ps(dummy_mask,r00);
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq00             = _mm256_mul_ps(iq0,jq0);
 385             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 386                                             vdwioffsetptr0+vdwjidx0B,
 387                                             vdwioffsetptr0+vdwjidx0C,
 388                                             vdwioffsetptr0+vdwjidx0D,
 389                                             vdwioffsetptr0+vdwjidx0E,
 390                                             vdwioffsetptr0+vdwjidx0F,
 391                                             vdwioffsetptr0+vdwjidx0G,
 392                                             vdwioffsetptr0+vdwjidx0H,
 393                                             &c6_00,&c12_00);
 394
 395             /* EWALD ELECTROSTATICS */
 396
 397             /* Analytical PME correction */
 398             zeta2            = _mm256_mul_ps(beta2,rsq00);
 399             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 400             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 401             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 402             felec            = _mm256_mul_ps(qq00,felec);
 403             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 404             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 405             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 406             velec            = _mm256_mul_ps(qq00,velec);
 407
 408             /* LENNARD-JONES DISPERSION/REPULSION */
 409
 410             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 411             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 412             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 413             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 414             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 415
 416             /* Update potential sum for this i atom from the interaction with this j atom. */
 417             velec            = _mm256_andnot_ps(dummy_mask,velec);
 418             velecsum         = _mm256_add_ps(velecsum,velec);
 419             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 420             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 421
 422             fscal            = _mm256_add_ps(felec,fvdw);
 423
 424             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 425
 426             /* Calculate temporary vectorial force */
 427             tx               = _mm256_mul_ps(fscal,dx00);
 428             ty               = _mm256_mul_ps(fscal,dy00);
 429             tz               = _mm256_mul_ps(fscal,dz00);
 430
 431             /* Update vectorial force */
 432             fix0             = _mm256_add_ps(fix0,tx);
 433             fiy0             = _mm256_add_ps(fiy0,ty);
 434             fiz0             = _mm256_add_ps(fiz0,tz);
 435
 436             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 437             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 438             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 439             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 440             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 441             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 442             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 443             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 444             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 445
 446             /* Inner loop uses 97 flops */
 447         }
 448
 449         /* End of innermost loop */
 450
 451         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 452                                                  f+i_coord_offset,fshift+i_shift_offset);
 453
 454         ggid                        = gid[iidx];
 455         /* Update potential energies */
 456         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 457         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 458
 459         /* Increment number of inner iterations */
 460         inneriter                  += j_index_end - j_index_start;
 461
 462         /* Outer loop uses 9 flops */
 463     }
 464
 465     /* Increment number of outer iterations */
 466     outeriter        += nri;
 467
 468     /* Update outer/inner flops */
 469
 470     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*97);
 471 }
 472 /*
 473  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_256_single
 474  * Electrostatics interaction: Ewald
 475  * VdW interaction:            LennardJones
 476  * Geometry:                   Particle-Particle
 477  * Calculate force/pot:        Force
 478  */
 479 void
 480 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_256_single
 481                     (t_nblist                    * gmx_restrict       nlist,
 482                      rvec                        * gmx_restrict          xx,
 483                      rvec                        * gmx_restrict          ff,
 484                      t_forcerec                  * gmx_restrict          fr,
 485                      t_mdatoms                   * gmx_restrict     mdatoms,
 486                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 487                      t_nrnb                      * gmx_restrict        nrnb)
 488 {
 489     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 490      * just 0 for non-waters.
 491      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 492      * jnr indices corresponding to data put in the four positions in the SIMD register.
 493      */
 494     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 495     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 496     int              jnrA,jnrB,jnrC,jnrD;
 497     int              jnrE,jnrF,jnrG,jnrH;
 498     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 499     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 500     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 501     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 502     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 503     real             rcutoff_scalar;
 504     real             *shiftvec,*fshift,*x,*f;
 505     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 506     real             scratch[4*DIM];
 507     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 508     real *           vdwioffsetptr0;
 509     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 510     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 511     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 512     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 513     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 514     real             *charge;
 515     int              nvdwtype;
 516     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 517     int              *vdwtype;
 518     real             *vdwparam;
 519     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 520     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 521     __m256i          ewitab;
 522     __m128i          ewitab_lo,ewitab_hi;
 523     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 524     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 525     real             *ewtab;
 526     __m256           dummy_mask,cutoff_mask;
 527     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 528     __m256           one     = _mm256_set1_ps(1.0);
 529     __m256           two     = _mm256_set1_ps(2.0);
 530     x                = xx[0];
 531     f                = ff[0];
 532
 533     nri              = nlist->nri;
 534     iinr             = nlist->iinr;
 535     jindex           = nlist->jindex;
 536     jjnr             = nlist->jjnr;
 537     shiftidx         = nlist->shift;
 538     gid              = nlist->gid;
 539     shiftvec         = fr->shift_vec[0];
 540     fshift           = fr->fshift[0];
 541     facel            = _mm256_set1_ps(fr->epsfac);
 542     charge           = mdatoms->chargeA;
 543     nvdwtype         = fr->ntype;
 544     vdwparam         = fr->nbfp;
 545     vdwtype          = mdatoms->typeA;
 546
 547     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 548     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 549     beta2            = _mm256_mul_ps(beta,beta);
 550     beta3            = _mm256_mul_ps(beta,beta2);
 551
 552     ewtab            = fr->ic->tabq_coul_F;
 553     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 554     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 555
 556     /* Avoid stupid compiler warnings */
 557     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 558     j_coord_offsetA = 0;
 559     j_coord_offsetB = 0;
 560     j_coord_offsetC = 0;
 561     j_coord_offsetD = 0;
 562     j_coord_offsetE = 0;
 563     j_coord_offsetF = 0;
 564     j_coord_offsetG = 0;
 565     j_coord_offsetH = 0;
 566
 567     outeriter        = 0;
 568     inneriter        = 0;
 569
 570     for(iidx=0;iidx<4*DIM;iidx++)
 571     {
 572         scratch[iidx] = 0.0;
 573     }
 574
 575     /* Start outer loop over neighborlists */
 576     for(iidx=0; iidx<nri; iidx++)
 577     {
 578         /* Load shift vector for this list */
 579         i_shift_offset   = DIM*shiftidx[iidx];
 580
 581         /* Load limits for loop over neighbors */
 582         j_index_start    = jindex[iidx];
 583         j_index_end      = jindex[iidx+1];
 584
 585         /* Get outer coordinate index */
 586         inr              = iinr[iidx];
 587         i_coord_offset   = DIM*inr;
 588
 589         /* Load i particle coords and add shift vector */
 590         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 591
 592         fix0             = _mm256_setzero_ps();
 593         fiy0             = _mm256_setzero_ps();
 594         fiz0             = _mm256_setzero_ps();
 595
 596         /* Load parameters for i particles */
 597         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 598         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 599
 600         /* Start inner kernel loop */
 601         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 602         {
 603
 604             /* Get j neighbor index, and coordinate index */
 605             jnrA             = jjnr[jidx];
 606             jnrB             = jjnr[jidx+1];
 607             jnrC             = jjnr[jidx+2];
 608             jnrD             = jjnr[jidx+3];
 609             jnrE             = jjnr[jidx+4];
 610             jnrF             = jjnr[jidx+5];
 611             jnrG             = jjnr[jidx+6];
 612             jnrH             = jjnr[jidx+7];
 613             j_coord_offsetA  = DIM*jnrA;
 614             j_coord_offsetB  = DIM*jnrB;
 615             j_coord_offsetC  = DIM*jnrC;
 616             j_coord_offsetD  = DIM*jnrD;
 617             j_coord_offsetE  = DIM*jnrE;
 618             j_coord_offsetF  = DIM*jnrF;
 619             j_coord_offsetG  = DIM*jnrG;
 620             j_coord_offsetH  = DIM*jnrH;
 621
 622             /* load j atom coordinates */
 623             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 624                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 625                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 626                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 627                                                  &jx0,&jy0,&jz0);
 628
 629             /* Calculate displacement vector */
 630             dx00             = _mm256_sub_ps(ix0,jx0);
 631             dy00             = _mm256_sub_ps(iy0,jy0);
 632             dz00             = _mm256_sub_ps(iz0,jz0);
 633
 634             /* Calculate squared distance and things based on it */
 635             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 636
 637             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 638
 639             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 640
 641             /* Load parameters for j particles */
 642             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 643                                                                  charge+jnrC+0,charge+jnrD+0,
 644                                                                  charge+jnrE+0,charge+jnrF+0,
 645                                                                  charge+jnrG+0,charge+jnrH+0);
 646             vdwjidx0A        = 2*vdwtype[jnrA+0];
 647             vdwjidx0B        = 2*vdwtype[jnrB+0];
 648             vdwjidx0C        = 2*vdwtype[jnrC+0];
 649             vdwjidx0D        = 2*vdwtype[jnrD+0];
 650             vdwjidx0E        = 2*vdwtype[jnrE+0];
 651             vdwjidx0F        = 2*vdwtype[jnrF+0];
 652             vdwjidx0G        = 2*vdwtype[jnrG+0];
 653             vdwjidx0H        = 2*vdwtype[jnrH+0];
 654
 655             /**************************
 656              * CALCULATE INTERACTIONS *
 657              **************************/
 658
 659             r00              = _mm256_mul_ps(rsq00,rinv00);
 660
 661             /* Compute parameters for interactions between i and j atoms */
 662             qq00             = _mm256_mul_ps(iq0,jq0);
 663             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 664                                             vdwioffsetptr0+vdwjidx0B,
 665                                             vdwioffsetptr0+vdwjidx0C,
 666                                             vdwioffsetptr0+vdwjidx0D,
 667                                             vdwioffsetptr0+vdwjidx0E,
 668                                             vdwioffsetptr0+vdwjidx0F,
 669                                             vdwioffsetptr0+vdwjidx0G,
 670                                             vdwioffsetptr0+vdwjidx0H,
 671                                             &c6_00,&c12_00);
 672
 673             /* EWALD ELECTROSTATICS */
 674
 675             /* Analytical PME correction */
 676             zeta2            = _mm256_mul_ps(beta2,rsq00);
 677             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 678             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 679             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 680             felec            = _mm256_mul_ps(qq00,felec);
 681
 682             /* LENNARD-JONES DISPERSION/REPULSION */
 683
 684             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 685             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 686
 687             fscal            = _mm256_add_ps(felec,fvdw);
 688
 689             /* Calculate temporary vectorial force */
 690             tx               = _mm256_mul_ps(fscal,dx00);
 691             ty               = _mm256_mul_ps(fscal,dy00);
 692             tz               = _mm256_mul_ps(fscal,dz00);
 693
 694             /* Update vectorial force */
 695             fix0             = _mm256_add_ps(fix0,tx);
 696             fiy0             = _mm256_add_ps(fiy0,ty);
 697             fiz0             = _mm256_add_ps(fiz0,tz);
 698
 699             fjptrA             = f+j_coord_offsetA;
 700             fjptrB             = f+j_coord_offsetB;
 701             fjptrC             = f+j_coord_offsetC;
 702             fjptrD             = f+j_coord_offsetD;
 703             fjptrE             = f+j_coord_offsetE;
 704             fjptrF             = f+j_coord_offsetF;
 705             fjptrG             = f+j_coord_offsetG;
 706             fjptrH             = f+j_coord_offsetH;
 707             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 708
 709             /* Inner loop uses 63 flops */
 710         }
 711
 712         if(jidx<j_index_end)
 713         {
 714
 715             /* Get j neighbor index, and coordinate index */
 716             jnrlistA         = jjnr[jidx];
 717             jnrlistB         = jjnr[jidx+1];
 718             jnrlistC         = jjnr[jidx+2];
 719             jnrlistD         = jjnr[jidx+3];
 720             jnrlistE         = jjnr[jidx+4];
 721             jnrlistF         = jjnr[jidx+5];
 722             jnrlistG         = jjnr[jidx+6];
 723             jnrlistH         = jjnr[jidx+7];
 724             /* Sign of each element will be negative for non-real atoms.
 725              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 726              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 727              */
 728             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 729                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 730
 731             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 732             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 733             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 734             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 735             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 736             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 737             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 738             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 739             j_coord_offsetA  = DIM*jnrA;
 740             j_coord_offsetB  = DIM*jnrB;
 741             j_coord_offsetC  = DIM*jnrC;
 742             j_coord_offsetD  = DIM*jnrD;
 743             j_coord_offsetE  = DIM*jnrE;
 744             j_coord_offsetF  = DIM*jnrF;
 745             j_coord_offsetG  = DIM*jnrG;
 746             j_coord_offsetH  = DIM*jnrH;
 747
 748             /* load j atom coordinates */
 749             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 750                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 751                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 752                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 753                                                  &jx0,&jy0,&jz0);
 754
 755             /* Calculate displacement vector */
 756             dx00             = _mm256_sub_ps(ix0,jx0);
 757             dy00             = _mm256_sub_ps(iy0,jy0);
 758             dz00             = _mm256_sub_ps(iz0,jz0);
 759
 760             /* Calculate squared distance and things based on it */
 761             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 762
 763             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 764
 765             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 766
 767             /* Load parameters for j particles */
 768             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 769                                                                  charge+jnrC+0,charge+jnrD+0,
 770                                                                  charge+jnrE+0,charge+jnrF+0,
 771                                                                  charge+jnrG+0,charge+jnrH+0);
 772             vdwjidx0A        = 2*vdwtype[jnrA+0];
 773             vdwjidx0B        = 2*vdwtype[jnrB+0];
 774             vdwjidx0C        = 2*vdwtype[jnrC+0];
 775             vdwjidx0D        = 2*vdwtype[jnrD+0];
 776             vdwjidx0E        = 2*vdwtype[jnrE+0];
 777             vdwjidx0F        = 2*vdwtype[jnrF+0];
 778             vdwjidx0G        = 2*vdwtype[jnrG+0];
 779             vdwjidx0H        = 2*vdwtype[jnrH+0];
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             r00              = _mm256_mul_ps(rsq00,rinv00);
 786             r00              = _mm256_andnot_ps(dummy_mask,r00);
 787
 788             /* Compute parameters for interactions between i and j atoms */
 789             qq00             = _mm256_mul_ps(iq0,jq0);
 790             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 791                                             vdwioffsetptr0+vdwjidx0B,
 792                                             vdwioffsetptr0+vdwjidx0C,
 793                                             vdwioffsetptr0+vdwjidx0D,
 794                                             vdwioffsetptr0+vdwjidx0E,
 795                                             vdwioffsetptr0+vdwjidx0F,
 796                                             vdwioffsetptr0+vdwjidx0G,
 797                                             vdwioffsetptr0+vdwjidx0H,
 798                                             &c6_00,&c12_00);
 799
 800             /* EWALD ELECTROSTATICS */
 801
 802             /* Analytical PME correction */
 803             zeta2            = _mm256_mul_ps(beta2,rsq00);
 804             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 805             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 806             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 807             felec            = _mm256_mul_ps(qq00,felec);
 808
 809             /* LENNARD-JONES DISPERSION/REPULSION */
 810
 811             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 812             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 813
 814             fscal            = _mm256_add_ps(felec,fvdw);
 815
 816             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 817
 818             /* Calculate temporary vectorial force */
 819             tx               = _mm256_mul_ps(fscal,dx00);
 820             ty               = _mm256_mul_ps(fscal,dy00);
 821             tz               = _mm256_mul_ps(fscal,dz00);
 822
 823             /* Update vectorial force */
 824             fix0             = _mm256_add_ps(fix0,tx);
 825             fiy0             = _mm256_add_ps(fiy0,ty);
 826             fiz0             = _mm256_add_ps(fiz0,tz);
 827
 828             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 829             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 830             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 831             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 832             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 833             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 834             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 835             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 836             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 837
 838             /* Inner loop uses 64 flops */
 839         }
 840
 841         /* End of innermost loop */
 842
 843         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 844                                                  f+i_coord_offset,fshift+i_shift_offset);
 845
 846         /* Increment number of inner iterations */
 847         inneriter                  += j_index_end - j_index_start;
 848
 849         /* Outer loop uses 7 flops */
 850     }
 851
 852     /* Increment number of outer iterations */
 853     outeriter        += nri;
 854
 855     /* Update outer/inner flops */
 856
 857     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 858 }