src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwNone_GeomW3P1_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_VdwNone_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwNone_GeomW3P1_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     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 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
 124     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 125     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 126     beta2            = _mm256_mul_ps(beta,beta);
 127     beta3            = _mm256_mul_ps(beta,beta2);
 128
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 136     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 137     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145     j_coord_offsetE = 0;
 146     j_coord_offsetF = 0;
 147     j_coord_offsetG = 0;
 148     j_coord_offsetH = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     for(iidx=0;iidx<4*DIM;iidx++)
 154     {
 155         scratch[iidx] = 0.0;
 156     }
 157
 158     /* Start outer loop over neighborlists */
 159     for(iidx=0; iidx<nri; iidx++)
 160     {
 161         /* Load shift vector for this list */
 162         i_shift_offset   = DIM*shiftidx[iidx];
 163
 164         /* Load limits for loop over neighbors */
 165         j_index_start    = jindex[iidx];
 166         j_index_end      = jindex[iidx+1];
 167
 168         /* Get outer coordinate index */
 169         inr              = iinr[iidx];
 170         i_coord_offset   = DIM*inr;
 171
 172         /* Load i particle coords and add shift vector */
 173         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 174                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 175
 176         fix0             = _mm256_setzero_ps();
 177         fiy0             = _mm256_setzero_ps();
 178         fiz0             = _mm256_setzero_ps();
 179         fix1             = _mm256_setzero_ps();
 180         fiy1             = _mm256_setzero_ps();
 181         fiz1             = _mm256_setzero_ps();
 182         fix2             = _mm256_setzero_ps();
 183         fiy2             = _mm256_setzero_ps();
 184         fiz2             = _mm256_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm256_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             jnrE             = jjnr[jidx+4];
 199             jnrF             = jjnr[jidx+5];
 200             jnrG             = jjnr[jidx+6];
 201             jnrH             = jjnr[jidx+7];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204             j_coord_offsetC  = DIM*jnrC;
 205             j_coord_offsetD  = DIM*jnrD;
 206             j_coord_offsetE  = DIM*jnrE;
 207             j_coord_offsetF  = DIM*jnrF;
 208             j_coord_offsetG  = DIM*jnrG;
 209             j_coord_offsetH  = DIM*jnrH;
 210
 211             /* load j atom coordinates */
 212             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 214                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 215                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 216                                                  &jx0,&jy0,&jz0);
 217
 218             /* Calculate displacement vector */
 219             dx00             = _mm256_sub_ps(ix0,jx0);
 220             dy00             = _mm256_sub_ps(iy0,jy0);
 221             dz00             = _mm256_sub_ps(iz0,jz0);
 222             dx10             = _mm256_sub_ps(ix1,jx0);
 223             dy10             = _mm256_sub_ps(iy1,jy0);
 224             dz10             = _mm256_sub_ps(iz1,jz0);
 225             dx20             = _mm256_sub_ps(ix2,jx0);
 226             dy20             = _mm256_sub_ps(iy2,jy0);
 227             dz20             = _mm256_sub_ps(iz2,jz0);
 228
 229             /* Calculate squared distance and things based on it */
 230             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 231             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 232             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 233
 234             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 235             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 236             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 237
 238             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 239             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 240             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                                  charge+jnrC+0,charge+jnrD+0,
 245                                                                  charge+jnrE+0,charge+jnrF+0,
 246                                                                  charge+jnrG+0,charge+jnrH+0);
 247
 248             fjx0             = _mm256_setzero_ps();
 249             fjy0             = _mm256_setzero_ps();
 250             fjz0             = _mm256_setzero_ps();
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             r00              = _mm256_mul_ps(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             qq00             = _mm256_mul_ps(iq0,jq0);
 260
 261             /* EWALD ELECTROSTATICS */
 262
 263             /* Analytical PME correction */
 264             zeta2            = _mm256_mul_ps(beta2,rsq00);
 265             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 266             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 267             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 268             felec            = _mm256_mul_ps(qq00,felec);
 269             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 270             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 271             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 272             velec            = _mm256_mul_ps(qq00,velec);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velecsum         = _mm256_add_ps(velecsum,velec);
 276
 277             fscal            = felec;
 278
 279             /* Calculate temporary vectorial force */
 280             tx               = _mm256_mul_ps(fscal,dx00);
 281             ty               = _mm256_mul_ps(fscal,dy00);
 282             tz               = _mm256_mul_ps(fscal,dz00);
 283
 284             /* Update vectorial force */
 285             fix0             = _mm256_add_ps(fix0,tx);
 286             fiy0             = _mm256_add_ps(fiy0,ty);
 287             fiz0             = _mm256_add_ps(fiz0,tz);
 288
 289             fjx0             = _mm256_add_ps(fjx0,tx);
 290             fjy0             = _mm256_add_ps(fjy0,ty);
 291             fjz0             = _mm256_add_ps(fjz0,tz);
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             r10              = _mm256_mul_ps(rsq10,rinv10);
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq10             = _mm256_mul_ps(iq1,jq0);
 301
 302             /* EWALD ELECTROSTATICS */
 303
 304             /* Analytical PME correction */
 305             zeta2            = _mm256_mul_ps(beta2,rsq10);
 306             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 307             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 308             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 309             felec            = _mm256_mul_ps(qq10,felec);
 310             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 311             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 312             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 313             velec            = _mm256_mul_ps(qq10,velec);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velecsum         = _mm256_add_ps(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm256_mul_ps(fscal,dx10);
 322             ty               = _mm256_mul_ps(fscal,dy10);
 323             tz               = _mm256_mul_ps(fscal,dz10);
 324
 325             /* Update vectorial force */
 326             fix1             = _mm256_add_ps(fix1,tx);
 327             fiy1             = _mm256_add_ps(fiy1,ty);
 328             fiz1             = _mm256_add_ps(fiz1,tz);
 329
 330             fjx0             = _mm256_add_ps(fjx0,tx);
 331             fjy0             = _mm256_add_ps(fjy0,ty);
 332             fjz0             = _mm256_add_ps(fjz0,tz);
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             r20              = _mm256_mul_ps(rsq20,rinv20);
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq20             = _mm256_mul_ps(iq2,jq0);
 342
 343             /* EWALD ELECTROSTATICS */
 344
 345             /* Analytical PME correction */
 346             zeta2            = _mm256_mul_ps(beta2,rsq20);
 347             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 348             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 349             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 350             felec            = _mm256_mul_ps(qq20,felec);
 351             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 352             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 353             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 354             velec            = _mm256_mul_ps(qq20,velec);
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm256_add_ps(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm256_mul_ps(fscal,dx20);
 363             ty               = _mm256_mul_ps(fscal,dy20);
 364             tz               = _mm256_mul_ps(fscal,dz20);
 365
 366             /* Update vectorial force */
 367             fix2             = _mm256_add_ps(fix2,tx);
 368             fiy2             = _mm256_add_ps(fiy2,ty);
 369             fiz2             = _mm256_add_ps(fiz2,tz);
 370
 371             fjx0             = _mm256_add_ps(fjx0,tx);
 372             fjy0             = _mm256_add_ps(fjy0,ty);
 373             fjz0             = _mm256_add_ps(fjz0,tz);
 374
 375             fjptrA             = f+j_coord_offsetA;
 376             fjptrB             = f+j_coord_offsetB;
 377             fjptrC             = f+j_coord_offsetC;
 378             fjptrD             = f+j_coord_offsetD;
 379             fjptrE             = f+j_coord_offsetE;
 380             fjptrF             = f+j_coord_offsetF;
 381             fjptrG             = f+j_coord_offsetG;
 382             fjptrH             = f+j_coord_offsetH;
 383
 384             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 385
 386             /* Inner loop uses 255 flops */
 387         }
 388
 389         if(jidx<j_index_end)
 390         {
 391
 392             /* Get j neighbor index, and coordinate index */
 393             jnrlistA         = jjnr[jidx];
 394             jnrlistB         = jjnr[jidx+1];
 395             jnrlistC         = jjnr[jidx+2];
 396             jnrlistD         = jjnr[jidx+3];
 397             jnrlistE         = jjnr[jidx+4];
 398             jnrlistF         = jjnr[jidx+5];
 399             jnrlistG         = jjnr[jidx+6];
 400             jnrlistH         = jjnr[jidx+7];
 401             /* Sign of each element will be negative for non-real atoms.
 402              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 403              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 404              */
 405             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 406                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 407
 408             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 409             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 410             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 411             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 412             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 413             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 414             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 415             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 416             j_coord_offsetA  = DIM*jnrA;
 417             j_coord_offsetB  = DIM*jnrB;
 418             j_coord_offsetC  = DIM*jnrC;
 419             j_coord_offsetD  = DIM*jnrD;
 420             j_coord_offsetE  = DIM*jnrE;
 421             j_coord_offsetF  = DIM*jnrF;
 422             j_coord_offsetG  = DIM*jnrG;
 423             j_coord_offsetH  = DIM*jnrH;
 424
 425             /* load j atom coordinates */
 426             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 428                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 429                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 430                                                  &jx0,&jy0,&jz0);
 431
 432             /* Calculate displacement vector */
 433             dx00             = _mm256_sub_ps(ix0,jx0);
 434             dy00             = _mm256_sub_ps(iy0,jy0);
 435             dz00             = _mm256_sub_ps(iz0,jz0);
 436             dx10             = _mm256_sub_ps(ix1,jx0);
 437             dy10             = _mm256_sub_ps(iy1,jy0);
 438             dz10             = _mm256_sub_ps(iz1,jz0);
 439             dx20             = _mm256_sub_ps(ix2,jx0);
 440             dy20             = _mm256_sub_ps(iy2,jy0);
 441             dz20             = _mm256_sub_ps(iz2,jz0);
 442
 443             /* Calculate squared distance and things based on it */
 444             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 445             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 446             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 447
 448             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 449             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 450             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 451
 452             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 453             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 454             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 455
 456             /* Load parameters for j particles */
 457             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 458                                                                  charge+jnrC+0,charge+jnrD+0,
 459                                                                  charge+jnrE+0,charge+jnrF+0,
 460                                                                  charge+jnrG+0,charge+jnrH+0);
 461
 462             fjx0             = _mm256_setzero_ps();
 463             fjy0             = _mm256_setzero_ps();
 464             fjz0             = _mm256_setzero_ps();
 465
 466             /**************************
 467              * CALCULATE INTERACTIONS *
 468              **************************/
 469
 470             r00              = _mm256_mul_ps(rsq00,rinv00);
 471             r00              = _mm256_andnot_ps(dummy_mask,r00);
 472
 473             /* Compute parameters for interactions between i and j atoms */
 474             qq00             = _mm256_mul_ps(iq0,jq0);
 475
 476             /* EWALD ELECTROSTATICS */
 477
 478             /* Analytical PME correction */
 479             zeta2            = _mm256_mul_ps(beta2,rsq00);
 480             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 481             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 482             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 483             felec            = _mm256_mul_ps(qq00,felec);
 484             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 485             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 486             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 487             velec            = _mm256_mul_ps(qq00,velec);
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velec            = _mm256_andnot_ps(dummy_mask,velec);
 491             velecsum         = _mm256_add_ps(velecsum,velec);
 492
 493             fscal            = felec;
 494
 495             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 496
 497             /* Calculate temporary vectorial force */
 498             tx               = _mm256_mul_ps(fscal,dx00);
 499             ty               = _mm256_mul_ps(fscal,dy00);
 500             tz               = _mm256_mul_ps(fscal,dz00);
 501
 502             /* Update vectorial force */
 503             fix0             = _mm256_add_ps(fix0,tx);
 504             fiy0             = _mm256_add_ps(fiy0,ty);
 505             fiz0             = _mm256_add_ps(fiz0,tz);
 506
 507             fjx0             = _mm256_add_ps(fjx0,tx);
 508             fjy0             = _mm256_add_ps(fjy0,ty);
 509             fjz0             = _mm256_add_ps(fjz0,tz);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             r10              = _mm256_mul_ps(rsq10,rinv10);
 516             r10              = _mm256_andnot_ps(dummy_mask,r10);
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             qq10             = _mm256_mul_ps(iq1,jq0);
 520
 521             /* EWALD ELECTROSTATICS */
 522
 523             /* Analytical PME correction */
 524             zeta2            = _mm256_mul_ps(beta2,rsq10);
 525             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 526             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 527             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 528             felec            = _mm256_mul_ps(qq10,felec);
 529             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 530             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 531             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 532             velec            = _mm256_mul_ps(qq10,velec);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm256_andnot_ps(dummy_mask,velec);
 536             velecsum         = _mm256_add_ps(velecsum,velec);
 537
 538             fscal            = felec;
 539
 540             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm256_mul_ps(fscal,dx10);
 544             ty               = _mm256_mul_ps(fscal,dy10);
 545             tz               = _mm256_mul_ps(fscal,dz10);
 546
 547             /* Update vectorial force */
 548             fix1             = _mm256_add_ps(fix1,tx);
 549             fiy1             = _mm256_add_ps(fiy1,ty);
 550             fiz1             = _mm256_add_ps(fiz1,tz);
 551
 552             fjx0             = _mm256_add_ps(fjx0,tx);
 553             fjy0             = _mm256_add_ps(fjy0,ty);
 554             fjz0             = _mm256_add_ps(fjz0,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r20              = _mm256_mul_ps(rsq20,rinv20);
 561             r20              = _mm256_andnot_ps(dummy_mask,r20);
 562
 563             /* Compute parameters for interactions between i and j atoms */
 564             qq20             = _mm256_mul_ps(iq2,jq0);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Analytical PME correction */
 569             zeta2            = _mm256_mul_ps(beta2,rsq20);
 570             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 571             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 572             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 573             felec            = _mm256_mul_ps(qq20,felec);
 574             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 575             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 576             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 577             velec            = _mm256_mul_ps(qq20,velec);
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velec            = _mm256_andnot_ps(dummy_mask,velec);
 581             velecsum         = _mm256_add_ps(velecsum,velec);
 582
 583             fscal            = felec;
 584
 585             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm256_mul_ps(fscal,dx20);
 589             ty               = _mm256_mul_ps(fscal,dy20);
 590             tz               = _mm256_mul_ps(fscal,dz20);
 591
 592             /* Update vectorial force */
 593             fix2             = _mm256_add_ps(fix2,tx);
 594             fiy2             = _mm256_add_ps(fiy2,ty);
 595             fiz2             = _mm256_add_ps(fiz2,tz);
 596
 597             fjx0             = _mm256_add_ps(fjx0,tx);
 598             fjy0             = _mm256_add_ps(fjy0,ty);
 599             fjz0             = _mm256_add_ps(fjz0,tz);
 600
 601             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 602             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 603             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 604             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 605             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 606             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 607             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 608             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 609
 610             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 611
 612             /* Inner loop uses 258 flops */
 613         }
 614
 615         /* End of innermost loop */
 616
 617         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 618                                                  f+i_coord_offset,fshift+i_shift_offset);
 619
 620         ggid                        = gid[iidx];
 621         /* Update potential energies */
 622         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 623
 624         /* Increment number of inner iterations */
 625         inneriter                  += j_index_end - j_index_start;
 626
 627         /* Outer loop uses 19 flops */
 628     }
 629
 630     /* Increment number of outer iterations */
 631     outeriter        += nri;
 632
 633     /* Update outer/inner flops */
 634
 635     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*258);
 636 }
 637 /*
 638  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_256_single
 639  * Electrostatics interaction: Ewald
 640  * VdW interaction:            None
 641  * Geometry:                   Water3-Particle
 642  * Calculate force/pot:        Force
 643  */
 644 void
 645 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_avx_256_single
 646                     (t_nblist                    * gmx_restrict       nlist,
 647                      rvec                        * gmx_restrict          xx,
 648                      rvec                        * gmx_restrict          ff,
 649                      t_forcerec                  * gmx_restrict          fr,
 650                      t_mdatoms                   * gmx_restrict     mdatoms,
 651                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 652                      t_nrnb                      * gmx_restrict        nrnb)
 653 {
 654     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 655      * just 0 for non-waters.
 656      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 657      * jnr indices corresponding to data put in the four positions in the SIMD register.
 658      */
 659     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 660     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 661     int              jnrA,jnrB,jnrC,jnrD;
 662     int              jnrE,jnrF,jnrG,jnrH;
 663     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 664     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 665     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 666     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 667     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 668     real             rcutoff_scalar;
 669     real             *shiftvec,*fshift,*x,*f;
 670     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 671     real             scratch[4*DIM];
 672     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 673     real *           vdwioffsetptr0;
 674     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 675     real *           vdwioffsetptr1;
 676     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 677     real *           vdwioffsetptr2;
 678     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 679     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 680     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 681     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 682     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 683     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 684     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 685     real             *charge;
 686     __m256i          ewitab;
 687     __m128i          ewitab_lo,ewitab_hi;
 688     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 689     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 690     real             *ewtab;
 691     __m256           dummy_mask,cutoff_mask;
 692     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 693     __m256           one     = _mm256_set1_ps(1.0);
 694     __m256           two     = _mm256_set1_ps(2.0);
 695     x                = xx[0];
 696     f                = ff[0];
 697
 698     nri              = nlist->nri;
 699     iinr             = nlist->iinr;
 700     jindex           = nlist->jindex;
 701     jjnr             = nlist->jjnr;
 702     shiftidx         = nlist->shift;
 703     gid              = nlist->gid;
 704     shiftvec         = fr->shift_vec[0];
 705     fshift           = fr->fshift[0];
 706     facel            = _mm256_set1_ps(fr->epsfac);
 707     charge           = mdatoms->chargeA;
 708
 709     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 710     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 711     beta2            = _mm256_mul_ps(beta,beta);
 712     beta3            = _mm256_mul_ps(beta,beta2);
 713
 714     ewtab            = fr->ic->tabq_coul_F;
 715     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 716     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 717
 718     /* Setup water-specific parameters */
 719     inr              = nlist->iinr[0];
 720     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 721     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 722     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 723
 724     /* Avoid stupid compiler warnings */
 725     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 726     j_coord_offsetA = 0;
 727     j_coord_offsetB = 0;
 728     j_coord_offsetC = 0;
 729     j_coord_offsetD = 0;
 730     j_coord_offsetE = 0;
 731     j_coord_offsetF = 0;
 732     j_coord_offsetG = 0;
 733     j_coord_offsetH = 0;
 734
 735     outeriter        = 0;
 736     inneriter        = 0;
 737
 738     for(iidx=0;iidx<4*DIM;iidx++)
 739     {
 740         scratch[iidx] = 0.0;
 741     }
 742
 743     /* Start outer loop over neighborlists */
 744     for(iidx=0; iidx<nri; iidx++)
 745     {
 746         /* Load shift vector for this list */
 747         i_shift_offset   = DIM*shiftidx[iidx];
 748
 749         /* Load limits for loop over neighbors */
 750         j_index_start    = jindex[iidx];
 751         j_index_end      = jindex[iidx+1];
 752
 753         /* Get outer coordinate index */
 754         inr              = iinr[iidx];
 755         i_coord_offset   = DIM*inr;
 756
 757         /* Load i particle coords and add shift vector */
 758         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 759                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 760
 761         fix0             = _mm256_setzero_ps();
 762         fiy0             = _mm256_setzero_ps();
 763         fiz0             = _mm256_setzero_ps();
 764         fix1             = _mm256_setzero_ps();
 765         fiy1             = _mm256_setzero_ps();
 766         fiz1             = _mm256_setzero_ps();
 767         fix2             = _mm256_setzero_ps();
 768         fiy2             = _mm256_setzero_ps();
 769         fiz2             = _mm256_setzero_ps();
 770
 771         /* Start inner kernel loop */
 772         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 773         {
 774
 775             /* Get j neighbor index, and coordinate index */
 776             jnrA             = jjnr[jidx];
 777             jnrB             = jjnr[jidx+1];
 778             jnrC             = jjnr[jidx+2];
 779             jnrD             = jjnr[jidx+3];
 780             jnrE             = jjnr[jidx+4];
 781             jnrF             = jjnr[jidx+5];
 782             jnrG             = jjnr[jidx+6];
 783             jnrH             = jjnr[jidx+7];
 784             j_coord_offsetA  = DIM*jnrA;
 785             j_coord_offsetB  = DIM*jnrB;
 786             j_coord_offsetC  = DIM*jnrC;
 787             j_coord_offsetD  = DIM*jnrD;
 788             j_coord_offsetE  = DIM*jnrE;
 789             j_coord_offsetF  = DIM*jnrF;
 790             j_coord_offsetG  = DIM*jnrG;
 791             j_coord_offsetH  = DIM*jnrH;
 792
 793             /* load j atom coordinates */
 794             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 795                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 796                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 797                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 798                                                  &jx0,&jy0,&jz0);
 799
 800             /* Calculate displacement vector */
 801             dx00             = _mm256_sub_ps(ix0,jx0);
 802             dy00             = _mm256_sub_ps(iy0,jy0);
 803             dz00             = _mm256_sub_ps(iz0,jz0);
 804             dx10             = _mm256_sub_ps(ix1,jx0);
 805             dy10             = _mm256_sub_ps(iy1,jy0);
 806             dz10             = _mm256_sub_ps(iz1,jz0);
 807             dx20             = _mm256_sub_ps(ix2,jx0);
 808             dy20             = _mm256_sub_ps(iy2,jy0);
 809             dz20             = _mm256_sub_ps(iz2,jz0);
 810
 811             /* Calculate squared distance and things based on it */
 812             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 813             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 814             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 815
 816             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 817             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 818             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 819
 820             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 821             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 822             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 823
 824             /* Load parameters for j particles */
 825             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 826                                                                  charge+jnrC+0,charge+jnrD+0,
 827                                                                  charge+jnrE+0,charge+jnrF+0,
 828                                                                  charge+jnrG+0,charge+jnrH+0);
 829
 830             fjx0             = _mm256_setzero_ps();
 831             fjy0             = _mm256_setzero_ps();
 832             fjz0             = _mm256_setzero_ps();
 833
 834             /**************************
 835              * CALCULATE INTERACTIONS *
 836              **************************/
 837
 838             r00              = _mm256_mul_ps(rsq00,rinv00);
 839
 840             /* Compute parameters for interactions between i and j atoms */
 841             qq00             = _mm256_mul_ps(iq0,jq0);
 842
 843             /* EWALD ELECTROSTATICS */
 844
 845             /* Analytical PME correction */
 846             zeta2            = _mm256_mul_ps(beta2,rsq00);
 847             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 848             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 849             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 850             felec            = _mm256_mul_ps(qq00,felec);
 851
 852             fscal            = felec;
 853
 854             /* Calculate temporary vectorial force */
 855             tx               = _mm256_mul_ps(fscal,dx00);
 856             ty               = _mm256_mul_ps(fscal,dy00);
 857             tz               = _mm256_mul_ps(fscal,dz00);
 858
 859             /* Update vectorial force */
 860             fix0             = _mm256_add_ps(fix0,tx);
 861             fiy0             = _mm256_add_ps(fiy0,ty);
 862             fiz0             = _mm256_add_ps(fiz0,tz);
 863
 864             fjx0             = _mm256_add_ps(fjx0,tx);
 865             fjy0             = _mm256_add_ps(fjy0,ty);
 866             fjz0             = _mm256_add_ps(fjz0,tz);
 867
 868             /**************************
 869              * CALCULATE INTERACTIONS *
 870              **************************/
 871
 872             r10              = _mm256_mul_ps(rsq10,rinv10);
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq10             = _mm256_mul_ps(iq1,jq0);
 876
 877             /* EWALD ELECTROSTATICS */
 878
 879             /* Analytical PME correction */
 880             zeta2            = _mm256_mul_ps(beta2,rsq10);
 881             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 882             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 883             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 884             felec            = _mm256_mul_ps(qq10,felec);
 885
 886             fscal            = felec;
 887
 888             /* Calculate temporary vectorial force */
 889             tx               = _mm256_mul_ps(fscal,dx10);
 890             ty               = _mm256_mul_ps(fscal,dy10);
 891             tz               = _mm256_mul_ps(fscal,dz10);
 892
 893             /* Update vectorial force */
 894             fix1             = _mm256_add_ps(fix1,tx);
 895             fiy1             = _mm256_add_ps(fiy1,ty);
 896             fiz1             = _mm256_add_ps(fiz1,tz);
 897
 898             fjx0             = _mm256_add_ps(fjx0,tx);
 899             fjy0             = _mm256_add_ps(fjy0,ty);
 900             fjz0             = _mm256_add_ps(fjz0,tz);
 901
 902             /**************************
 903              * CALCULATE INTERACTIONS *
 904              **************************/
 905
 906             r20              = _mm256_mul_ps(rsq20,rinv20);
 907
 908             /* Compute parameters for interactions between i and j atoms */
 909             qq20             = _mm256_mul_ps(iq2,jq0);
 910
 911             /* EWALD ELECTROSTATICS */
 912
 913             /* Analytical PME correction */
 914             zeta2            = _mm256_mul_ps(beta2,rsq20);
 915             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 916             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 917             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 918             felec            = _mm256_mul_ps(qq20,felec);
 919
 920             fscal            = felec;
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm256_mul_ps(fscal,dx20);
 924             ty               = _mm256_mul_ps(fscal,dy20);
 925             tz               = _mm256_mul_ps(fscal,dz20);
 926
 927             /* Update vectorial force */
 928             fix2             = _mm256_add_ps(fix2,tx);
 929             fiy2             = _mm256_add_ps(fiy2,ty);
 930             fiz2             = _mm256_add_ps(fiz2,tz);
 931
 932             fjx0             = _mm256_add_ps(fjx0,tx);
 933             fjy0             = _mm256_add_ps(fjy0,ty);
 934             fjz0             = _mm256_add_ps(fjz0,tz);
 935
 936             fjptrA             = f+j_coord_offsetA;
 937             fjptrB             = f+j_coord_offsetB;
 938             fjptrC             = f+j_coord_offsetC;
 939             fjptrD             = f+j_coord_offsetD;
 940             fjptrE             = f+j_coord_offsetE;
 941             fjptrF             = f+j_coord_offsetF;
 942             fjptrG             = f+j_coord_offsetG;
 943             fjptrH             = f+j_coord_offsetH;
 944
 945             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 946
 947             /* Inner loop uses 171 flops */
 948         }
 949
 950         if(jidx<j_index_end)
 951         {
 952
 953             /* Get j neighbor index, and coordinate index */
 954             jnrlistA         = jjnr[jidx];
 955             jnrlistB         = jjnr[jidx+1];
 956             jnrlistC         = jjnr[jidx+2];
 957             jnrlistD         = jjnr[jidx+3];
 958             jnrlistE         = jjnr[jidx+4];
 959             jnrlistF         = jjnr[jidx+5];
 960             jnrlistG         = jjnr[jidx+6];
 961             jnrlistH         = jjnr[jidx+7];
 962             /* Sign of each element will be negative for non-real atoms.
 963              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 964              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 965              */
 966             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 967                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 968
 969             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 970             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 971             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 972             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 973             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 974             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 975             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 976             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 977             j_coord_offsetA  = DIM*jnrA;
 978             j_coord_offsetB  = DIM*jnrB;
 979             j_coord_offsetC  = DIM*jnrC;
 980             j_coord_offsetD  = DIM*jnrD;
 981             j_coord_offsetE  = DIM*jnrE;
 982             j_coord_offsetF  = DIM*jnrF;
 983             j_coord_offsetG  = DIM*jnrG;
 984             j_coord_offsetH  = DIM*jnrH;
 985
 986             /* load j atom coordinates */
 987             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 988                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 989                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 990                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 991                                                  &jx0,&jy0,&jz0);
 992
 993             /* Calculate displacement vector */
 994             dx00             = _mm256_sub_ps(ix0,jx0);
 995             dy00             = _mm256_sub_ps(iy0,jy0);
 996             dz00             = _mm256_sub_ps(iz0,jz0);
 997             dx10             = _mm256_sub_ps(ix1,jx0);
 998             dy10             = _mm256_sub_ps(iy1,jy0);
 999             dz10             = _mm256_sub_ps(iz1,jz0);
1000             dx20             = _mm256_sub_ps(ix2,jx0);
1001             dy20             = _mm256_sub_ps(iy2,jy0);
1002             dz20             = _mm256_sub_ps(iz2,jz0);
1003
1004             /* Calculate squared distance and things based on it */
1005             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1006             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1007             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1008
1009             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1010             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1011             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1012
1013             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1014             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1015             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1016
1017             /* Load parameters for j particles */
1018             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1019                                                                  charge+jnrC+0,charge+jnrD+0,
1020                                                                  charge+jnrE+0,charge+jnrF+0,
1021                                                                  charge+jnrG+0,charge+jnrH+0);
1022
1023             fjx0             = _mm256_setzero_ps();
1024             fjy0             = _mm256_setzero_ps();
1025             fjz0             = _mm256_setzero_ps();
1026
1027             /**************************
1028              * CALCULATE INTERACTIONS *
1029              **************************/
1030
1031             r00              = _mm256_mul_ps(rsq00,rinv00);
1032             r00              = _mm256_andnot_ps(dummy_mask,r00);
1033
1034             /* Compute parameters for interactions between i and j atoms */
1035             qq00             = _mm256_mul_ps(iq0,jq0);
1036
1037             /* EWALD ELECTROSTATICS */
1038
1039             /* Analytical PME correction */
1040             zeta2            = _mm256_mul_ps(beta2,rsq00);
1041             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1042             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1043             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1044             felec            = _mm256_mul_ps(qq00,felec);
1045
1046             fscal            = felec;
1047
1048             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm256_mul_ps(fscal,dx00);
1052             ty               = _mm256_mul_ps(fscal,dy00);
1053             tz               = _mm256_mul_ps(fscal,dz00);
1054
1055             /* Update vectorial force */
1056             fix0             = _mm256_add_ps(fix0,tx);
1057             fiy0             = _mm256_add_ps(fiy0,ty);
1058             fiz0             = _mm256_add_ps(fiz0,tz);
1059
1060             fjx0             = _mm256_add_ps(fjx0,tx);
1061             fjy0             = _mm256_add_ps(fjy0,ty);
1062             fjz0             = _mm256_add_ps(fjz0,tz);
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             r10              = _mm256_mul_ps(rsq10,rinv10);
1069             r10              = _mm256_andnot_ps(dummy_mask,r10);
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq10             = _mm256_mul_ps(iq1,jq0);
1073
1074             /* EWALD ELECTROSTATICS */
1075
1076             /* Analytical PME correction */
1077             zeta2            = _mm256_mul_ps(beta2,rsq10);
1078             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1079             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1080             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1081             felec            = _mm256_mul_ps(qq10,felec);
1082
1083             fscal            = felec;
1084
1085             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1086
1087             /* Calculate temporary vectorial force */
1088             tx               = _mm256_mul_ps(fscal,dx10);
1089             ty               = _mm256_mul_ps(fscal,dy10);
1090             tz               = _mm256_mul_ps(fscal,dz10);
1091
1092             /* Update vectorial force */
1093             fix1             = _mm256_add_ps(fix1,tx);
1094             fiy1             = _mm256_add_ps(fiy1,ty);
1095             fiz1             = _mm256_add_ps(fiz1,tz);
1096
1097             fjx0             = _mm256_add_ps(fjx0,tx);
1098             fjy0             = _mm256_add_ps(fjy0,ty);
1099             fjz0             = _mm256_add_ps(fjz0,tz);
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             r20              = _mm256_mul_ps(rsq20,rinv20);
1106             r20              = _mm256_andnot_ps(dummy_mask,r20);
1107
1108             /* Compute parameters for interactions between i and j atoms */
1109             qq20             = _mm256_mul_ps(iq2,jq0);
1110
1111             /* EWALD ELECTROSTATICS */
1112
1113             /* Analytical PME correction */
1114             zeta2            = _mm256_mul_ps(beta2,rsq20);
1115             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1116             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1117             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1118             felec            = _mm256_mul_ps(qq20,felec);
1119
1120             fscal            = felec;
1121
1122             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1123
1124             /* Calculate temporary vectorial force */
1125             tx               = _mm256_mul_ps(fscal,dx20);
1126             ty               = _mm256_mul_ps(fscal,dy20);
1127             tz               = _mm256_mul_ps(fscal,dz20);
1128
1129             /* Update vectorial force */
1130             fix2             = _mm256_add_ps(fix2,tx);
1131             fiy2             = _mm256_add_ps(fiy2,ty);
1132             fiz2             = _mm256_add_ps(fiz2,tz);
1133
1134             fjx0             = _mm256_add_ps(fjx0,tx);
1135             fjy0             = _mm256_add_ps(fjy0,ty);
1136             fjz0             = _mm256_add_ps(fjz0,tz);
1137
1138             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1139             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1140             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1141             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1142             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1143             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1144             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1145             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1146
1147             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1148
1149             /* Inner loop uses 174 flops */
1150         }
1151
1152         /* End of innermost loop */
1153
1154         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1155                                                  f+i_coord_offset,fshift+i_shift_offset);
1156
1157         /* Increment number of inner iterations */
1158         inneriter                  += j_index_end - j_index_start;
1159
1160         /* Outer loop uses 18 flops */
1161     }
1162
1163     /* Increment number of outer iterations */
1164     outeriter        += nri;
1165
1166     /* Update outer/inner flops */
1167
1168     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*174);
1169 }