src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSw_VdwNone_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  89     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m256i          ewitab;
  94     __m128i          ewitab_lo,ewitab_hi;
  95     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  96     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  97     real             *ewtab;
  98     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  99     real             rswitch_scalar,d_scalar;
 100     __m256           dummy_mask,cutoff_mask;
 101     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 102     __m256           one     = _mm256_set1_ps(1.0);
 103     __m256           two     = _mm256_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm256_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117
 118     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 119     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 120     beta2            = _mm256_mul_ps(beta,beta);
 121     beta3            = _mm256_mul_ps(beta,beta2);
 122
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 126
 127     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 130     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 131
 132     rswitch_scalar   = fr->rcoulomb_switch;
 133     rswitch          = _mm256_set1_ps(rswitch_scalar);
 134     /* Setup switch parameters */
 135     d_scalar         = rcutoff_scalar-rswitch_scalar;
 136     d                = _mm256_set1_ps(d_scalar);
 137     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 138     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 141     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 142     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150     j_coord_offsetE = 0;
 151     j_coord_offsetF = 0;
 152     j_coord_offsetG = 0;
 153     j_coord_offsetH = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 179
 180         fix0             = _mm256_setzero_ps();
 181         fiy0             = _mm256_setzero_ps();
 182         fiz0             = _mm256_setzero_ps();
 183
 184         /* Load parameters for i particles */
 185         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm256_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             jnrE             = jjnr[jidx+4];
 200             jnrF             = jjnr[jidx+5];
 201             jnrG             = jjnr[jidx+6];
 202             jnrH             = jjnr[jidx+7];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207             j_coord_offsetE  = DIM*jnrE;
 208             j_coord_offsetF  = DIM*jnrF;
 209             j_coord_offsetG  = DIM*jnrG;
 210             j_coord_offsetH  = DIM*jnrH;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 216                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 217                                                  &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm256_sub_ps(ix0,jx0);
 221             dy00             = _mm256_sub_ps(iy0,jy0);
 222             dz00             = _mm256_sub_ps(iz0,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 226
 227             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 228
 229             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                                  charge+jnrC+0,charge+jnrD+0,
 234                                                                  charge+jnrE+0,charge+jnrF+0,
 235                                                                  charge+jnrG+0,charge+jnrH+0);
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 242             {
 243
 244             r00              = _mm256_mul_ps(rsq00,rinv00);
 245
 246             /* Compute parameters for interactions between i and j atoms */
 247             qq00             = _mm256_mul_ps(iq0,jq0);
 248
 249             /* EWALD ELECTROSTATICS */
 250
 251             /* Analytical PME correction */
 252             zeta2            = _mm256_mul_ps(beta2,rsq00);
 253             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 254             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 255             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 256             felec            = _mm256_mul_ps(qq00,felec);
 257             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 258             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 259             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 260             velec            = _mm256_mul_ps(qq00,velec);
 261
 262             d                = _mm256_sub_ps(r00,rswitch);
 263             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 264             d2               = _mm256_mul_ps(d,d);
 265             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 266
 267             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 268
 269             /* Evaluate switch function */
 270             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 271             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 272             velec            = _mm256_mul_ps(velec,sw);
 273             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm256_and_ps(velec,cutoff_mask);
 277             velecsum         = _mm256_add_ps(velecsum,velec);
 278
 279             fscal            = felec;
 280
 281             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 282
 283             /* Calculate temporary vectorial force */
 284             tx               = _mm256_mul_ps(fscal,dx00);
 285             ty               = _mm256_mul_ps(fscal,dy00);
 286             tz               = _mm256_mul_ps(fscal,dz00);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm256_add_ps(fix0,tx);
 290             fiy0             = _mm256_add_ps(fiy0,ty);
 291             fiz0             = _mm256_add_ps(fiz0,tz);
 292
 293             fjptrA             = f+j_coord_offsetA;
 294             fjptrB             = f+j_coord_offsetB;
 295             fjptrC             = f+j_coord_offsetC;
 296             fjptrD             = f+j_coord_offsetD;
 297             fjptrE             = f+j_coord_offsetE;
 298             fjptrF             = f+j_coord_offsetF;
 299             fjptrG             = f+j_coord_offsetG;
 300             fjptrH             = f+j_coord_offsetH;
 301             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 302
 303             }
 304
 305             /* Inner loop uses 108 flops */
 306         }
 307
 308         if(jidx<j_index_end)
 309         {
 310
 311             /* Get j neighbor index, and coordinate index */
 312             jnrlistA         = jjnr[jidx];
 313             jnrlistB         = jjnr[jidx+1];
 314             jnrlistC         = jjnr[jidx+2];
 315             jnrlistD         = jjnr[jidx+3];
 316             jnrlistE         = jjnr[jidx+4];
 317             jnrlistF         = jjnr[jidx+5];
 318             jnrlistG         = jjnr[jidx+6];
 319             jnrlistH         = jjnr[jidx+7];
 320             /* Sign of each element will be negative for non-real atoms.
 321              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 322              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 323              */
 324             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 325                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 326
 327             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 328             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 329             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 330             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 331             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 332             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 333             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 334             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 335             j_coord_offsetA  = DIM*jnrA;
 336             j_coord_offsetB  = DIM*jnrB;
 337             j_coord_offsetC  = DIM*jnrC;
 338             j_coord_offsetD  = DIM*jnrD;
 339             j_coord_offsetE  = DIM*jnrE;
 340             j_coord_offsetF  = DIM*jnrF;
 341             j_coord_offsetG  = DIM*jnrG;
 342             j_coord_offsetH  = DIM*jnrH;
 343
 344             /* load j atom coordinates */
 345             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 346                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 347                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 348                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 349                                                  &jx0,&jy0,&jz0);
 350
 351             /* Calculate displacement vector */
 352             dx00             = _mm256_sub_ps(ix0,jx0);
 353             dy00             = _mm256_sub_ps(iy0,jy0);
 354             dz00             = _mm256_sub_ps(iz0,jz0);
 355
 356             /* Calculate squared distance and things based on it */
 357             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 358
 359             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 360
 361             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 362
 363             /* Load parameters for j particles */
 364             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 365                                                                  charge+jnrC+0,charge+jnrD+0,
 366                                                                  charge+jnrE+0,charge+jnrF+0,
 367                                                                  charge+jnrG+0,charge+jnrH+0);
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 374             {
 375
 376             r00              = _mm256_mul_ps(rsq00,rinv00);
 377             r00              = _mm256_andnot_ps(dummy_mask,r00);
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq00             = _mm256_mul_ps(iq0,jq0);
 381
 382             /* EWALD ELECTROSTATICS */
 383
 384             /* Analytical PME correction */
 385             zeta2            = _mm256_mul_ps(beta2,rsq00);
 386             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 387             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 388             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 389             felec            = _mm256_mul_ps(qq00,felec);
 390             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 391             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 392             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 393             velec            = _mm256_mul_ps(qq00,velec);
 394
 395             d                = _mm256_sub_ps(r00,rswitch);
 396             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 397             d2               = _mm256_mul_ps(d,d);
 398             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 399
 400             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 401
 402             /* Evaluate switch function */
 403             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 404             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 405             velec            = _mm256_mul_ps(velec,sw);
 406             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velec            = _mm256_and_ps(velec,cutoff_mask);
 410             velec            = _mm256_andnot_ps(dummy_mask,velec);
 411             velecsum         = _mm256_add_ps(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 416
 417             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 418
 419             /* Calculate temporary vectorial force */
 420             tx               = _mm256_mul_ps(fscal,dx00);
 421             ty               = _mm256_mul_ps(fscal,dy00);
 422             tz               = _mm256_mul_ps(fscal,dz00);
 423
 424             /* Update vectorial force */
 425             fix0             = _mm256_add_ps(fix0,tx);
 426             fiy0             = _mm256_add_ps(fiy0,ty);
 427             fiz0             = _mm256_add_ps(fiz0,tz);
 428
 429             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 430             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 431             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 432             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 433             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 434             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 435             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 436             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 437             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 438
 439             }
 440
 441             /* Inner loop uses 109 flops */
 442         }
 443
 444         /* End of innermost loop */
 445
 446         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 447                                                  f+i_coord_offset,fshift+i_shift_offset);
 448
 449         ggid                        = gid[iidx];
 450         /* Update potential energies */
 451         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 452
 453         /* Increment number of inner iterations */
 454         inneriter                  += j_index_end - j_index_start;
 455
 456         /* Outer loop uses 8 flops */
 457     }
 458
 459     /* Increment number of outer iterations */
 460     outeriter        += nri;
 461
 462     /* Update outer/inner flops */
 463
 464     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*109);
 465 }
 466 /*
 467  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_256_single
 468  * Electrostatics interaction: Ewald
 469  * VdW interaction:            None
 470  * Geometry:                   Particle-Particle
 471  * Calculate force/pot:        Force
 472  */
 473 void
 474 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_256_single
 475                     (t_nblist                    * gmx_restrict       nlist,
 476                      rvec                        * gmx_restrict          xx,
 477                      rvec                        * gmx_restrict          ff,
 478                      t_forcerec                  * gmx_restrict          fr,
 479                      t_mdatoms                   * gmx_restrict     mdatoms,
 480                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 481                      t_nrnb                      * gmx_restrict        nrnb)
 482 {
 483     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 484      * just 0 for non-waters.
 485      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 486      * jnr indices corresponding to data put in the four positions in the SIMD register.
 487      */
 488     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 489     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 490     int              jnrA,jnrB,jnrC,jnrD;
 491     int              jnrE,jnrF,jnrG,jnrH;
 492     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 493     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 494     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 495     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 496     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 497     real             rcutoff_scalar;
 498     real             *shiftvec,*fshift,*x,*f;
 499     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 500     real             scratch[4*DIM];
 501     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 502     real *           vdwioffsetptr0;
 503     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 504     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 505     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 506     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 507     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 508     real             *charge;
 509     __m256i          ewitab;
 510     __m128i          ewitab_lo,ewitab_hi;
 511     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 512     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 513     real             *ewtab;
 514     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 515     real             rswitch_scalar,d_scalar;
 516     __m256           dummy_mask,cutoff_mask;
 517     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 518     __m256           one     = _mm256_set1_ps(1.0);
 519     __m256           two     = _mm256_set1_ps(2.0);
 520     x                = xx[0];
 521     f                = ff[0];
 522
 523     nri              = nlist->nri;
 524     iinr             = nlist->iinr;
 525     jindex           = nlist->jindex;
 526     jjnr             = nlist->jjnr;
 527     shiftidx         = nlist->shift;
 528     gid              = nlist->gid;
 529     shiftvec         = fr->shift_vec[0];
 530     fshift           = fr->fshift[0];
 531     facel            = _mm256_set1_ps(fr->epsfac);
 532     charge           = mdatoms->chargeA;
 533
 534     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 535     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 536     beta2            = _mm256_mul_ps(beta,beta);
 537     beta3            = _mm256_mul_ps(beta,beta2);
 538
 539     ewtab            = fr->ic->tabq_coul_FDV0;
 540     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 541     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 542
 543     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 544     rcutoff_scalar   = fr->rcoulomb;
 545     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 546     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 547
 548     rswitch_scalar   = fr->rcoulomb_switch;
 549     rswitch          = _mm256_set1_ps(rswitch_scalar);
 550     /* Setup switch parameters */
 551     d_scalar         = rcutoff_scalar-rswitch_scalar;
 552     d                = _mm256_set1_ps(d_scalar);
 553     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 554     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 555     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 556     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 557     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 558     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 559
 560     /* Avoid stupid compiler warnings */
 561     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 562     j_coord_offsetA = 0;
 563     j_coord_offsetB = 0;
 564     j_coord_offsetC = 0;
 565     j_coord_offsetD = 0;
 566     j_coord_offsetE = 0;
 567     j_coord_offsetF = 0;
 568     j_coord_offsetG = 0;
 569     j_coord_offsetH = 0;
 570
 571     outeriter        = 0;
 572     inneriter        = 0;
 573
 574     for(iidx=0;iidx<4*DIM;iidx++)
 575     {
 576         scratch[iidx] = 0.0;
 577     }
 578
 579     /* Start outer loop over neighborlists */
 580     for(iidx=0; iidx<nri; iidx++)
 581     {
 582         /* Load shift vector for this list */
 583         i_shift_offset   = DIM*shiftidx[iidx];
 584
 585         /* Load limits for loop over neighbors */
 586         j_index_start    = jindex[iidx];
 587         j_index_end      = jindex[iidx+1];
 588
 589         /* Get outer coordinate index */
 590         inr              = iinr[iidx];
 591         i_coord_offset   = DIM*inr;
 592
 593         /* Load i particle coords and add shift vector */
 594         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 595
 596         fix0             = _mm256_setzero_ps();
 597         fiy0             = _mm256_setzero_ps();
 598         fiz0             = _mm256_setzero_ps();
 599
 600         /* Load parameters for i particles */
 601         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 602
 603         /* Start inner kernel loop */
 604         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 605         {
 606
 607             /* Get j neighbor index, and coordinate index */
 608             jnrA             = jjnr[jidx];
 609             jnrB             = jjnr[jidx+1];
 610             jnrC             = jjnr[jidx+2];
 611             jnrD             = jjnr[jidx+3];
 612             jnrE             = jjnr[jidx+4];
 613             jnrF             = jjnr[jidx+5];
 614             jnrG             = jjnr[jidx+6];
 615             jnrH             = jjnr[jidx+7];
 616             j_coord_offsetA  = DIM*jnrA;
 617             j_coord_offsetB  = DIM*jnrB;
 618             j_coord_offsetC  = DIM*jnrC;
 619             j_coord_offsetD  = DIM*jnrD;
 620             j_coord_offsetE  = DIM*jnrE;
 621             j_coord_offsetF  = DIM*jnrF;
 622             j_coord_offsetG  = DIM*jnrG;
 623             j_coord_offsetH  = DIM*jnrH;
 624
 625             /* load j atom coordinates */
 626             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 627                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 628                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 629                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 630                                                  &jx0,&jy0,&jz0);
 631
 632             /* Calculate displacement vector */
 633             dx00             = _mm256_sub_ps(ix0,jx0);
 634             dy00             = _mm256_sub_ps(iy0,jy0);
 635             dz00             = _mm256_sub_ps(iz0,jz0);
 636
 637             /* Calculate squared distance and things based on it */
 638             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 639
 640             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 641
 642             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 643
 644             /* Load parameters for j particles */
 645             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 646                                                                  charge+jnrC+0,charge+jnrD+0,
 647                                                                  charge+jnrE+0,charge+jnrF+0,
 648                                                                  charge+jnrG+0,charge+jnrH+0);
 649
 650             /**************************
 651              * CALCULATE INTERACTIONS *
 652              **************************/
 653
 654             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 655             {
 656
 657             r00              = _mm256_mul_ps(rsq00,rinv00);
 658
 659             /* Compute parameters for interactions between i and j atoms */
 660             qq00             = _mm256_mul_ps(iq0,jq0);
 661
 662             /* EWALD ELECTROSTATICS */
 663
 664             /* Analytical PME correction */
 665             zeta2            = _mm256_mul_ps(beta2,rsq00);
 666             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 667             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 668             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 669             felec            = _mm256_mul_ps(qq00,felec);
 670             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 671             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 672             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 673             velec            = _mm256_mul_ps(qq00,velec);
 674
 675             d                = _mm256_sub_ps(r00,rswitch);
 676             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 677             d2               = _mm256_mul_ps(d,d);
 678             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 679
 680             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 681
 682             /* Evaluate switch function */
 683             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 684             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 685             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 686
 687             fscal            = felec;
 688
 689             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 690
 691             /* Calculate temporary vectorial force */
 692             tx               = _mm256_mul_ps(fscal,dx00);
 693             ty               = _mm256_mul_ps(fscal,dy00);
 694             tz               = _mm256_mul_ps(fscal,dz00);
 695
 696             /* Update vectorial force */
 697             fix0             = _mm256_add_ps(fix0,tx);
 698             fiy0             = _mm256_add_ps(fiy0,ty);
 699             fiz0             = _mm256_add_ps(fiz0,tz);
 700
 701             fjptrA             = f+j_coord_offsetA;
 702             fjptrB             = f+j_coord_offsetB;
 703             fjptrC             = f+j_coord_offsetC;
 704             fjptrD             = f+j_coord_offsetD;
 705             fjptrE             = f+j_coord_offsetE;
 706             fjptrF             = f+j_coord_offsetF;
 707             fjptrG             = f+j_coord_offsetG;
 708             fjptrH             = f+j_coord_offsetH;
 709             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 710
 711             }
 712
 713             /* Inner loop uses 105 flops */
 714         }
 715
 716         if(jidx<j_index_end)
 717         {
 718
 719             /* Get j neighbor index, and coordinate index */
 720             jnrlistA         = jjnr[jidx];
 721             jnrlistB         = jjnr[jidx+1];
 722             jnrlistC         = jjnr[jidx+2];
 723             jnrlistD         = jjnr[jidx+3];
 724             jnrlistE         = jjnr[jidx+4];
 725             jnrlistF         = jjnr[jidx+5];
 726             jnrlistG         = jjnr[jidx+6];
 727             jnrlistH         = jjnr[jidx+7];
 728             /* Sign of each element will be negative for non-real atoms.
 729              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 730              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 731              */
 732             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 733                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 734
 735             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 736             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 737             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 738             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 739             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 740             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 741             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 742             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 743             j_coord_offsetA  = DIM*jnrA;
 744             j_coord_offsetB  = DIM*jnrB;
 745             j_coord_offsetC  = DIM*jnrC;
 746             j_coord_offsetD  = DIM*jnrD;
 747             j_coord_offsetE  = DIM*jnrE;
 748             j_coord_offsetF  = DIM*jnrF;
 749             j_coord_offsetG  = DIM*jnrG;
 750             j_coord_offsetH  = DIM*jnrH;
 751
 752             /* load j atom coordinates */
 753             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 754                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 755                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 756                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 757                                                  &jx0,&jy0,&jz0);
 758
 759             /* Calculate displacement vector */
 760             dx00             = _mm256_sub_ps(ix0,jx0);
 761             dy00             = _mm256_sub_ps(iy0,jy0);
 762             dz00             = _mm256_sub_ps(iz0,jz0);
 763
 764             /* Calculate squared distance and things based on it */
 765             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 766
 767             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 768
 769             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 770
 771             /* Load parameters for j particles */
 772             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 773                                                                  charge+jnrC+0,charge+jnrD+0,
 774                                                                  charge+jnrE+0,charge+jnrF+0,
 775                                                                  charge+jnrG+0,charge+jnrH+0);
 776
 777             /**************************
 778              * CALCULATE INTERACTIONS *
 779              **************************/
 780
 781             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 782             {
 783
 784             r00              = _mm256_mul_ps(rsq00,rinv00);
 785             r00              = _mm256_andnot_ps(dummy_mask,r00);
 786
 787             /* Compute parameters for interactions between i and j atoms */
 788             qq00             = _mm256_mul_ps(iq0,jq0);
 789
 790             /* EWALD ELECTROSTATICS */
 791
 792             /* Analytical PME correction */
 793             zeta2            = _mm256_mul_ps(beta2,rsq00);
 794             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 795             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 796             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 797             felec            = _mm256_mul_ps(qq00,felec);
 798             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 799             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 800             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 801             velec            = _mm256_mul_ps(qq00,velec);
 802
 803             d                = _mm256_sub_ps(r00,rswitch);
 804             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 805             d2               = _mm256_mul_ps(d,d);
 806             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 807
 808             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 809
 810             /* Evaluate switch function */
 811             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 812             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 813             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 814
 815             fscal            = felec;
 816
 817             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 818
 819             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 820
 821             /* Calculate temporary vectorial force */
 822             tx               = _mm256_mul_ps(fscal,dx00);
 823             ty               = _mm256_mul_ps(fscal,dy00);
 824             tz               = _mm256_mul_ps(fscal,dz00);
 825
 826             /* Update vectorial force */
 827             fix0             = _mm256_add_ps(fix0,tx);
 828             fiy0             = _mm256_add_ps(fiy0,ty);
 829             fiz0             = _mm256_add_ps(fiz0,tz);
 830
 831             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 832             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 833             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 834             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 835             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 836             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 837             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 838             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 839             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 840
 841             }
 842
 843             /* Inner loop uses 106 flops */
 844         }
 845
 846         /* End of innermost loop */
 847
 848         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 849                                                  f+i_coord_offset,fshift+i_shift_offset);
 850
 851         /* Increment number of inner iterations */
 852         inneriter                  += j_index_end - j_index_start;
 853
 854         /* Outer loop uses 7 flops */
 855     }
 856
 857     /* Increment number of outer iterations */
 858     outeriter        += nri;
 859
 860     /* Update outer/inner flops */
 861
 862     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*106);
 863 }