src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwNone_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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_128_fma_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  50 #include "kernelutil_x86_avx_128_fma_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_avx_128_fma_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     __m128i          ewitab;
  99     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 101     real             *ewtab;
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119
 120     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 121     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 122     beta2            = _mm_mul_ps(beta,beta);
 123     beta3            = _mm_mul_ps(beta,beta2);
 124     ewtab            = fr->ic->tabq_coul_FDV0;
 125     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 126     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 131     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 132     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->rcoulomb;
 136     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 137     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_ps();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               x+j_coord_offsetC,x+j_coord_offsetD,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_ps(ix0,jx0);
 206             dy00             = _mm_sub_ps(iy0,jy0);
 207             dz00             = _mm_sub_ps(iz0,jz0);
 208             dx10             = _mm_sub_ps(ix1,jx0);
 209             dy10             = _mm_sub_ps(iy1,jy0);
 210             dz10             = _mm_sub_ps(iz1,jz0);
 211             dx20             = _mm_sub_ps(ix2,jx0);
 212             dy20             = _mm_sub_ps(iy2,jy0);
 213             dz20             = _mm_sub_ps(iz2,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 217             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 218             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 219
 220             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 221             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 222             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 223
 224             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 225             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 226             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 230                                                               charge+jnrC+0,charge+jnrD+0);
 231
 232             fjx0             = _mm_setzero_ps();
 233             fjy0             = _mm_setzero_ps();
 234             fjz0             = _mm_setzero_ps();
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             if (gmx_mm_any_lt(rsq00,rcutoff2))
 241             {
 242
 243             r00              = _mm_mul_ps(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm_mul_ps(iq0,jq0);
 247
 248             /* EWALD ELECTROSTATICS */
 249
 250             /* Analytical PME correction */
 251             zeta2            = _mm_mul_ps(beta2,rsq00);
 252             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 253             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 254             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 255             felec            = _mm_mul_ps(qq00,felec);
 256             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 257             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 258             velec            = _mm_mul_ps(qq00,velec);
 259
 260             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 261
 262             /* Update potential sum for this i atom from the interaction with this j atom. */
 263             velec            = _mm_and_ps(velec,cutoff_mask);
 264             velecsum         = _mm_add_ps(velecsum,velec);
 265
 266             fscal            = felec;
 267
 268             fscal            = _mm_and_ps(fscal,cutoff_mask);
 269
 270              /* Update vectorial force */
 271             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 272             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 273             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 274
 275             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 276             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 277             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 278
 279             }
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             if (gmx_mm_any_lt(rsq10,rcutoff2))
 286             {
 287
 288             r10              = _mm_mul_ps(rsq10,rinv10);
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq10             = _mm_mul_ps(iq1,jq0);
 292
 293             /* EWALD ELECTROSTATICS */
 294
 295             /* Analytical PME correction */
 296             zeta2            = _mm_mul_ps(beta2,rsq10);
 297             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 298             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 299             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 300             felec            = _mm_mul_ps(qq10,felec);
 301             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 302             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 303             velec            = _mm_mul_ps(qq10,velec);
 304
 305             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _mm_and_ps(velec,cutoff_mask);
 309             velecsum         = _mm_add_ps(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             fscal            = _mm_and_ps(fscal,cutoff_mask);
 314
 315              /* Update vectorial force */
 316             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 317             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 318             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 319
 320             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 321             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 322             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 323
 324             }
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             if (gmx_mm_any_lt(rsq20,rcutoff2))
 331             {
 332
 333             r20              = _mm_mul_ps(rsq20,rinv20);
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq20             = _mm_mul_ps(iq2,jq0);
 337
 338             /* EWALD ELECTROSTATICS */
 339
 340             /* Analytical PME correction */
 341             zeta2            = _mm_mul_ps(beta2,rsq20);
 342             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 343             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 344             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 345             felec            = _mm_mul_ps(qq20,felec);
 346             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 347             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 348             velec            = _mm_mul_ps(qq20,velec);
 349
 350             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velec            = _mm_and_ps(velec,cutoff_mask);
 354             velecsum         = _mm_add_ps(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358             fscal            = _mm_and_ps(fscal,cutoff_mask);
 359
 360              /* Update vectorial force */
 361             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 362             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 363             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 364
 365             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 366             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 367             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 368
 369             }
 370
 371             fjptrA             = f+j_coord_offsetA;
 372             fjptrB             = f+j_coord_offsetB;
 373             fjptrC             = f+j_coord_offsetC;
 374             fjptrD             = f+j_coord_offsetD;
 375
 376             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 377
 378             /* Inner loop uses 99 flops */
 379         }
 380
 381         if(jidx<j_index_end)
 382         {
 383
 384             /* Get j neighbor index, and coordinate index */
 385             jnrlistA         = jjnr[jidx];
 386             jnrlistB         = jjnr[jidx+1];
 387             jnrlistC         = jjnr[jidx+2];
 388             jnrlistD         = jjnr[jidx+3];
 389             /* Sign of each element will be negative for non-real atoms.
 390              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 391              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 392              */
 393             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 394             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 395             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 396             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 397             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 398             j_coord_offsetA  = DIM*jnrA;
 399             j_coord_offsetB  = DIM*jnrB;
 400             j_coord_offsetC  = DIM*jnrC;
 401             j_coord_offsetD  = DIM*jnrD;
 402
 403             /* load j atom coordinates */
 404             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 405                                               x+j_coord_offsetC,x+j_coord_offsetD,
 406                                               &jx0,&jy0,&jz0);
 407
 408             /* Calculate displacement vector */
 409             dx00             = _mm_sub_ps(ix0,jx0);
 410             dy00             = _mm_sub_ps(iy0,jy0);
 411             dz00             = _mm_sub_ps(iz0,jz0);
 412             dx10             = _mm_sub_ps(ix1,jx0);
 413             dy10             = _mm_sub_ps(iy1,jy0);
 414             dz10             = _mm_sub_ps(iz1,jz0);
 415             dx20             = _mm_sub_ps(ix2,jx0);
 416             dy20             = _mm_sub_ps(iy2,jy0);
 417             dz20             = _mm_sub_ps(iz2,jz0);
 418
 419             /* Calculate squared distance and things based on it */
 420             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 421             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 422             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 423
 424             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 425             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 426             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 427
 428             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 429             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 430             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 431
 432             /* Load parameters for j particles */
 433             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 434                                                               charge+jnrC+0,charge+jnrD+0);
 435
 436             fjx0             = _mm_setzero_ps();
 437             fjy0             = _mm_setzero_ps();
 438             fjz0             = _mm_setzero_ps();
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             if (gmx_mm_any_lt(rsq00,rcutoff2))
 445             {
 446
 447             r00              = _mm_mul_ps(rsq00,rinv00);
 448             r00              = _mm_andnot_ps(dummy_mask,r00);
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq00             = _mm_mul_ps(iq0,jq0);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Analytical PME correction */
 456             zeta2            = _mm_mul_ps(beta2,rsq00);
 457             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 458             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 459             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 460             felec            = _mm_mul_ps(qq00,felec);
 461             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 462             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 463             velec            = _mm_mul_ps(qq00,velec);
 464
 465             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 466
 467             /* Update potential sum for this i atom from the interaction with this j atom. */
 468             velec            = _mm_and_ps(velec,cutoff_mask);
 469             velec            = _mm_andnot_ps(dummy_mask,velec);
 470             velecsum         = _mm_add_ps(velecsum,velec);
 471
 472             fscal            = felec;
 473
 474             fscal            = _mm_and_ps(fscal,cutoff_mask);
 475
 476             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 477
 478              /* Update vectorial force */
 479             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 480             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 481             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 482
 483             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 484             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 485             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 486
 487             }
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             if (gmx_mm_any_lt(rsq10,rcutoff2))
 494             {
 495
 496             r10              = _mm_mul_ps(rsq10,rinv10);
 497             r10              = _mm_andnot_ps(dummy_mask,r10);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             qq10             = _mm_mul_ps(iq1,jq0);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Analytical PME correction */
 505             zeta2            = _mm_mul_ps(beta2,rsq10);
 506             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 507             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 508             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 509             felec            = _mm_mul_ps(qq10,felec);
 510             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 511             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 512             velec            = _mm_mul_ps(qq10,velec);
 513
 514             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm_and_ps(velec,cutoff_mask);
 518             velec            = _mm_andnot_ps(dummy_mask,velec);
 519             velecsum         = _mm_add_ps(velecsum,velec);
 520
 521             fscal            = felec;
 522
 523             fscal            = _mm_and_ps(fscal,cutoff_mask);
 524
 525             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 526
 527              /* Update vectorial force */
 528             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 529             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 530             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 531
 532             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 533             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 534             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 535
 536             }
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             if (gmx_mm_any_lt(rsq20,rcutoff2))
 543             {
 544
 545             r20              = _mm_mul_ps(rsq20,rinv20);
 546             r20              = _mm_andnot_ps(dummy_mask,r20);
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq20             = _mm_mul_ps(iq2,jq0);
 550
 551             /* EWALD ELECTROSTATICS */
 552
 553             /* Analytical PME correction */
 554             zeta2            = _mm_mul_ps(beta2,rsq20);
 555             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 556             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 557             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 558             felec            = _mm_mul_ps(qq20,felec);
 559             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 560             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 561             velec            = _mm_mul_ps(qq20,velec);
 562
 563             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velec            = _mm_and_ps(velec,cutoff_mask);
 567             velec            = _mm_andnot_ps(dummy_mask,velec);
 568             velecsum         = _mm_add_ps(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm_and_ps(fscal,cutoff_mask);
 573
 574             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 575
 576              /* Update vectorial force */
 577             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 578             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 579             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 580
 581             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 582             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 583             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 584
 585             }
 586
 587             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 588             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 589             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 590             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 591
 592             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 593
 594             /* Inner loop uses 102 flops */
 595         }
 596
 597         /* End of innermost loop */
 598
 599         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 600                                               f+i_coord_offset,fshift+i_shift_offset);
 601
 602         ggid                        = gid[iidx];
 603         /* Update potential energies */
 604         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 605
 606         /* Increment number of inner iterations */
 607         inneriter                  += j_index_end - j_index_start;
 608
 609         /* Outer loop uses 19 flops */
 610     }
 611
 612     /* Increment number of outer iterations */
 613     outeriter        += nri;
 614
 615     /* Update outer/inner flops */
 616
 617     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*102);
 618 }
 619 /*
 620  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_128_fma_single
 621  * Electrostatics interaction: Ewald
 622  * VdW interaction:            None
 623  * Geometry:                   Water3-Particle
 624  * Calculate force/pot:        Force
 625  */
 626 void
 627 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_avx_128_fma_single
 628                     (t_nblist                    * gmx_restrict       nlist,
 629                      rvec                        * gmx_restrict          xx,
 630                      rvec                        * gmx_restrict          ff,
 631                      t_forcerec                  * gmx_restrict          fr,
 632                      t_mdatoms                   * gmx_restrict     mdatoms,
 633                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 634                      t_nrnb                      * gmx_restrict        nrnb)
 635 {
 636     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 637      * just 0 for non-waters.
 638      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 639      * jnr indices corresponding to data put in the four positions in the SIMD register.
 640      */
 641     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 642     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 643     int              jnrA,jnrB,jnrC,jnrD;
 644     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 645     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 646     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 647     real             rcutoff_scalar;
 648     real             *shiftvec,*fshift,*x,*f;
 649     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 650     real             scratch[4*DIM];
 651     __m128           fscal,rcutoff,rcutoff2,jidxall;
 652     int              vdwioffset0;
 653     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 654     int              vdwioffset1;
 655     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 656     int              vdwioffset2;
 657     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 658     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 659     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 660     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 661     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 662     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 663     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 664     real             *charge;
 665     __m128i          ewitab;
 666     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 667     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 668     real             *ewtab;
 669     __m128           dummy_mask,cutoff_mask;
 670     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 671     __m128           one     = _mm_set1_ps(1.0);
 672     __m128           two     = _mm_set1_ps(2.0);
 673     x                = xx[0];
 674     f                = ff[0];
 675
 676     nri              = nlist->nri;
 677     iinr             = nlist->iinr;
 678     jindex           = nlist->jindex;
 679     jjnr             = nlist->jjnr;
 680     shiftidx         = nlist->shift;
 681     gid              = nlist->gid;
 682     shiftvec         = fr->shift_vec[0];
 683     fshift           = fr->fshift[0];
 684     facel            = _mm_set1_ps(fr->epsfac);
 685     charge           = mdatoms->chargeA;
 686
 687     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 688     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 689     beta2            = _mm_mul_ps(beta,beta);
 690     beta3            = _mm_mul_ps(beta,beta2);
 691     ewtab            = fr->ic->tabq_coul_F;
 692     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 693     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 694
 695     /* Setup water-specific parameters */
 696     inr              = nlist->iinr[0];
 697     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 698     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 699     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 700
 701     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 702     rcutoff_scalar   = fr->rcoulomb;
 703     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 704     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 705
 706     /* Avoid stupid compiler warnings */
 707     jnrA = jnrB = jnrC = jnrD = 0;
 708     j_coord_offsetA = 0;
 709     j_coord_offsetB = 0;
 710     j_coord_offsetC = 0;
 711     j_coord_offsetD = 0;
 712
 713     outeriter        = 0;
 714     inneriter        = 0;
 715
 716     for(iidx=0;iidx<4*DIM;iidx++)
 717     {
 718         scratch[iidx] = 0.0;
 719     }
 720
 721     /* Start outer loop over neighborlists */
 722     for(iidx=0; iidx<nri; iidx++)
 723     {
 724         /* Load shift vector for this list */
 725         i_shift_offset   = DIM*shiftidx[iidx];
 726
 727         /* Load limits for loop over neighbors */
 728         j_index_start    = jindex[iidx];
 729         j_index_end      = jindex[iidx+1];
 730
 731         /* Get outer coordinate index */
 732         inr              = iinr[iidx];
 733         i_coord_offset   = DIM*inr;
 734
 735         /* Load i particle coords and add shift vector */
 736         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 737                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 738
 739         fix0             = _mm_setzero_ps();
 740         fiy0             = _mm_setzero_ps();
 741         fiz0             = _mm_setzero_ps();
 742         fix1             = _mm_setzero_ps();
 743         fiy1             = _mm_setzero_ps();
 744         fiz1             = _mm_setzero_ps();
 745         fix2             = _mm_setzero_ps();
 746         fiy2             = _mm_setzero_ps();
 747         fiz2             = _mm_setzero_ps();
 748
 749         /* Start inner kernel loop */
 750         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 751         {
 752
 753             /* Get j neighbor index, and coordinate index */
 754             jnrA             = jjnr[jidx];
 755             jnrB             = jjnr[jidx+1];
 756             jnrC             = jjnr[jidx+2];
 757             jnrD             = jjnr[jidx+3];
 758             j_coord_offsetA  = DIM*jnrA;
 759             j_coord_offsetB  = DIM*jnrB;
 760             j_coord_offsetC  = DIM*jnrC;
 761             j_coord_offsetD  = DIM*jnrD;
 762
 763             /* load j atom coordinates */
 764             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 765                                               x+j_coord_offsetC,x+j_coord_offsetD,
 766                                               &jx0,&jy0,&jz0);
 767
 768             /* Calculate displacement vector */
 769             dx00             = _mm_sub_ps(ix0,jx0);
 770             dy00             = _mm_sub_ps(iy0,jy0);
 771             dz00             = _mm_sub_ps(iz0,jz0);
 772             dx10             = _mm_sub_ps(ix1,jx0);
 773             dy10             = _mm_sub_ps(iy1,jy0);
 774             dz10             = _mm_sub_ps(iz1,jz0);
 775             dx20             = _mm_sub_ps(ix2,jx0);
 776             dy20             = _mm_sub_ps(iy2,jy0);
 777             dz20             = _mm_sub_ps(iz2,jz0);
 778
 779             /* Calculate squared distance and things based on it */
 780             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 781             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 782             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 783
 784             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 785             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 786             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 787
 788             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 789             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 790             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 791
 792             /* Load parameters for j particles */
 793             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 794                                                               charge+jnrC+0,charge+jnrD+0);
 795
 796             fjx0             = _mm_setzero_ps();
 797             fjy0             = _mm_setzero_ps();
 798             fjz0             = _mm_setzero_ps();
 799
 800             /**************************
 801              * CALCULATE INTERACTIONS *
 802              **************************/
 803
 804             if (gmx_mm_any_lt(rsq00,rcutoff2))
 805             {
 806
 807             r00              = _mm_mul_ps(rsq00,rinv00);
 808
 809             /* Compute parameters for interactions between i and j atoms */
 810             qq00             = _mm_mul_ps(iq0,jq0);
 811
 812             /* EWALD ELECTROSTATICS */
 813
 814             /* Analytical PME correction */
 815             zeta2            = _mm_mul_ps(beta2,rsq00);
 816             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 817             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 818             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 819             felec            = _mm_mul_ps(qq00,felec);
 820
 821             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 822
 823             fscal            = felec;
 824
 825             fscal            = _mm_and_ps(fscal,cutoff_mask);
 826
 827              /* Update vectorial force */
 828             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 829             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 830             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 831
 832             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 833             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 834             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 835
 836             }
 837
 838             /**************************
 839              * CALCULATE INTERACTIONS *
 840              **************************/
 841
 842             if (gmx_mm_any_lt(rsq10,rcutoff2))
 843             {
 844
 845             r10              = _mm_mul_ps(rsq10,rinv10);
 846
 847             /* Compute parameters for interactions between i and j atoms */
 848             qq10             = _mm_mul_ps(iq1,jq0);
 849
 850             /* EWALD ELECTROSTATICS */
 851
 852             /* Analytical PME correction */
 853             zeta2            = _mm_mul_ps(beta2,rsq10);
 854             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 855             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 856             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 857             felec            = _mm_mul_ps(qq10,felec);
 858
 859             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 860
 861             fscal            = felec;
 862
 863             fscal            = _mm_and_ps(fscal,cutoff_mask);
 864
 865              /* Update vectorial force */
 866             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 867             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 868             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 869
 870             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 871             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 872             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 873
 874             }
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             if (gmx_mm_any_lt(rsq20,rcutoff2))
 881             {
 882
 883             r20              = _mm_mul_ps(rsq20,rinv20);
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq20             = _mm_mul_ps(iq2,jq0);
 887
 888             /* EWALD ELECTROSTATICS */
 889
 890             /* Analytical PME correction */
 891             zeta2            = _mm_mul_ps(beta2,rsq20);
 892             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 893             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 894             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 895             felec            = _mm_mul_ps(qq20,felec);
 896
 897             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 898
 899             fscal            = felec;
 900
 901             fscal            = _mm_and_ps(fscal,cutoff_mask);
 902
 903              /* Update vectorial force */
 904             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 905             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 906             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 907
 908             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 909             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 910             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 911
 912             }
 913
 914             fjptrA             = f+j_coord_offsetA;
 915             fjptrB             = f+j_coord_offsetB;
 916             fjptrC             = f+j_coord_offsetC;
 917             fjptrD             = f+j_coord_offsetD;
 918
 919             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 920
 921             /* Inner loop uses 93 flops */
 922         }
 923
 924         if(jidx<j_index_end)
 925         {
 926
 927             /* Get j neighbor index, and coordinate index */
 928             jnrlistA         = jjnr[jidx];
 929             jnrlistB         = jjnr[jidx+1];
 930             jnrlistC         = jjnr[jidx+2];
 931             jnrlistD         = jjnr[jidx+3];
 932             /* Sign of each element will be negative for non-real atoms.
 933              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 934              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 935              */
 936             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 937             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 938             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 939             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 940             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 941             j_coord_offsetA  = DIM*jnrA;
 942             j_coord_offsetB  = DIM*jnrB;
 943             j_coord_offsetC  = DIM*jnrC;
 944             j_coord_offsetD  = DIM*jnrD;
 945
 946             /* load j atom coordinates */
 947             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 948                                               x+j_coord_offsetC,x+j_coord_offsetD,
 949                                               &jx0,&jy0,&jz0);
 950
 951             /* Calculate displacement vector */
 952             dx00             = _mm_sub_ps(ix0,jx0);
 953             dy00             = _mm_sub_ps(iy0,jy0);
 954             dz00             = _mm_sub_ps(iz0,jz0);
 955             dx10             = _mm_sub_ps(ix1,jx0);
 956             dy10             = _mm_sub_ps(iy1,jy0);
 957             dz10             = _mm_sub_ps(iz1,jz0);
 958             dx20             = _mm_sub_ps(ix2,jx0);
 959             dy20             = _mm_sub_ps(iy2,jy0);
 960             dz20             = _mm_sub_ps(iz2,jz0);
 961
 962             /* Calculate squared distance and things based on it */
 963             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 964             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 965             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 966
 967             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 968             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 969             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 970
 971             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 972             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 973             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 974
 975             /* Load parameters for j particles */
 976             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 977                                                               charge+jnrC+0,charge+jnrD+0);
 978
 979             fjx0             = _mm_setzero_ps();
 980             fjy0             = _mm_setzero_ps();
 981             fjz0             = _mm_setzero_ps();
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             if (gmx_mm_any_lt(rsq00,rcutoff2))
 988             {
 989
 990             r00              = _mm_mul_ps(rsq00,rinv00);
 991             r00              = _mm_andnot_ps(dummy_mask,r00);
 992
 993             /* Compute parameters for interactions between i and j atoms */
 994             qq00             = _mm_mul_ps(iq0,jq0);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Analytical PME correction */
 999             zeta2            = _mm_mul_ps(beta2,rsq00);
1000             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1001             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1002             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1003             felec            = _mm_mul_ps(qq00,felec);
1004
1005             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1006
1007             fscal            = felec;
1008
1009             fscal            = _mm_and_ps(fscal,cutoff_mask);
1010
1011             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1012
1013              /* Update vectorial force */
1014             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1015             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1016             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1017
1018             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1019             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1020             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1021
1022             }
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             if (gmx_mm_any_lt(rsq10,rcutoff2))
1029             {
1030
1031             r10              = _mm_mul_ps(rsq10,rinv10);
1032             r10              = _mm_andnot_ps(dummy_mask,r10);
1033
1034             /* Compute parameters for interactions between i and j atoms */
1035             qq10             = _mm_mul_ps(iq1,jq0);
1036
1037             /* EWALD ELECTROSTATICS */
1038
1039             /* Analytical PME correction */
1040             zeta2            = _mm_mul_ps(beta2,rsq10);
1041             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1042             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1043             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1044             felec            = _mm_mul_ps(qq10,felec);
1045
1046             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm_and_ps(fscal,cutoff_mask);
1051
1052             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1053
1054              /* Update vectorial force */
1055             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1056             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1057             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1058
1059             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1060             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1061             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1062
1063             }
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             if (gmx_mm_any_lt(rsq20,rcutoff2))
1070             {
1071
1072             r20              = _mm_mul_ps(rsq20,rinv20);
1073             r20              = _mm_andnot_ps(dummy_mask,r20);
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq20             = _mm_mul_ps(iq2,jq0);
1077
1078             /* EWALD ELECTROSTATICS */
1079
1080             /* Analytical PME correction */
1081             zeta2            = _mm_mul_ps(beta2,rsq20);
1082             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1083             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1084             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1085             felec            = _mm_mul_ps(qq20,felec);
1086
1087             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm_and_ps(fscal,cutoff_mask);
1092
1093             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1094
1095              /* Update vectorial force */
1096             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1097             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1098             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1099
1100             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1101             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1102             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1103
1104             }
1105
1106             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1107             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1108             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1109             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1110
1111             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1112
1113             /* Inner loop uses 96 flops */
1114         }
1115
1116         /* End of innermost loop */
1117
1118         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1119                                               f+i_coord_offset,fshift+i_shift_offset);
1120
1121         /* Increment number of inner iterations */
1122         inneriter                  += j_index_end - j_index_start;
1123
1124         /* Outer loop uses 18 flops */
1125     }
1126
1127     /* Increment number of outer iterations */
1128     outeriter        += nri;
1129
1130     /* Update outer/inner flops */
1131
1132     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*96);
1133 }