src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_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_ElecEw_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_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              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128i          vfitab;
 108     __m128i          ifour       = _mm_set1_epi32(4);
 109     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 110     real             *vftab;
 111     __m128i          ewitab;
 112     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 113     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 114     real             *ewtab;
 115     __m128           dummy_mask,cutoff_mask;
 116     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 117     __m128           one     = _mm_set1_ps(1.0);
 118     __m128           two     = _mm_set1_ps(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_ps(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     vftab            = kernel_data->table_vdw->data;
 137     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 138
 139     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 140     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 141     beta2            = _mm_mul_ps(beta,beta);
 142     beta3            = _mm_mul_ps(beta,beta2);
 143     ewtab            = fr->ic->tabq_coul_FDV0;
 144     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 145     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 146
 147     /* Setup water-specific parameters */
 148     inr              = nlist->iinr[0];
 149     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 150     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 151     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 152     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     for(iidx=0;iidx<4*DIM;iidx++)
 165     {
 166         scratch[iidx] = 0.0;
 167     }
 168
 169     /* Start outer loop over neighborlists */
 170     for(iidx=0; iidx<nri; iidx++)
 171     {
 172         /* Load shift vector for this list */
 173         i_shift_offset   = DIM*shiftidx[iidx];
 174
 175         /* Load limits for loop over neighbors */
 176         j_index_start    = jindex[iidx];
 177         j_index_end      = jindex[iidx+1];
 178
 179         /* Get outer coordinate index */
 180         inr              = iinr[iidx];
 181         i_coord_offset   = DIM*inr;
 182
 183         /* Load i particle coords and add shift vector */
 184         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 185                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 186
 187         fix0             = _mm_setzero_ps();
 188         fiy0             = _mm_setzero_ps();
 189         fiz0             = _mm_setzero_ps();
 190         fix1             = _mm_setzero_ps();
 191         fiy1             = _mm_setzero_ps();
 192         fiz1             = _mm_setzero_ps();
 193         fix2             = _mm_setzero_ps();
 194         fiy2             = _mm_setzero_ps();
 195         fiz2             = _mm_setzero_ps();
 196         fix3             = _mm_setzero_ps();
 197         fiy3             = _mm_setzero_ps();
 198         fiz3             = _mm_setzero_ps();
 199
 200         /* Reset potential sums */
 201         velecsum         = _mm_setzero_ps();
 202         vvdwsum          = _mm_setzero_ps();
 203
 204         /* Start inner kernel loop */
 205         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 206         {
 207
 208             /* Get j neighbor index, and coordinate index */
 209             jnrA             = jjnr[jidx];
 210             jnrB             = jjnr[jidx+1];
 211             jnrC             = jjnr[jidx+2];
 212             jnrD             = jjnr[jidx+3];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215             j_coord_offsetC  = DIM*jnrC;
 216             j_coord_offsetD  = DIM*jnrD;
 217
 218             /* load j atom coordinates */
 219             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 220                                               x+j_coord_offsetC,x+j_coord_offsetD,
 221                                               &jx0,&jy0,&jz0);
 222
 223             /* Calculate displacement vector */
 224             dx00             = _mm_sub_ps(ix0,jx0);
 225             dy00             = _mm_sub_ps(iy0,jy0);
 226             dz00             = _mm_sub_ps(iz0,jz0);
 227             dx10             = _mm_sub_ps(ix1,jx0);
 228             dy10             = _mm_sub_ps(iy1,jy0);
 229             dz10             = _mm_sub_ps(iz1,jz0);
 230             dx20             = _mm_sub_ps(ix2,jx0);
 231             dy20             = _mm_sub_ps(iy2,jy0);
 232             dz20             = _mm_sub_ps(iz2,jz0);
 233             dx30             = _mm_sub_ps(ix3,jx0);
 234             dy30             = _mm_sub_ps(iy3,jy0);
 235             dz30             = _mm_sub_ps(iz3,jz0);
 236
 237             /* Calculate squared distance and things based on it */
 238             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 239             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 240             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 241             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 242
 243             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 244             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 245             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 246             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 247
 248             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 249             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 250             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 251
 252             /* Load parameters for j particles */
 253             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 254                                                               charge+jnrC+0,charge+jnrD+0);
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256             vdwjidx0B        = 2*vdwtype[jnrB+0];
 257             vdwjidx0C        = 2*vdwtype[jnrC+0];
 258             vdwjidx0D        = 2*vdwtype[jnrD+0];
 259
 260             fjx0             = _mm_setzero_ps();
 261             fjy0             = _mm_setzero_ps();
 262             fjz0             = _mm_setzero_ps();
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             r00              = _mm_mul_ps(rsq00,rinv00);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 272                                          vdwparam+vdwioffset0+vdwjidx0B,
 273                                          vdwparam+vdwioffset0+vdwjidx0C,
 274                                          vdwparam+vdwioffset0+vdwjidx0D,
 275                                          &c6_00,&c12_00);
 276
 277             /* Calculate table index by multiplying r with table scale and truncate to integer */
 278             rt               = _mm_mul_ps(r00,vftabscale);
 279             vfitab           = _mm_cvttps_epi32(rt);
 280 #ifdef __XOP__
 281             vfeps            = _mm_frcz_ps(rt);
 282 #else
 283             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 284 #endif
 285             twovfeps         = _mm_add_ps(vfeps,vfeps);
 286             vfitab           = _mm_slli_epi32(vfitab,3);
 287
 288             /* CUBIC SPLINE TABLE DISPERSION */
 289             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 290             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 291             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 292             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 293             _MM_TRANSPOSE4_PS(Y,F,G,H);
 294             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 295             VV               = _mm_macc_ps(vfeps,Fp,Y);
 296             vvdw6            = _mm_mul_ps(c6_00,VV);
 297             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 298             fvdw6            = _mm_mul_ps(c6_00,FF);
 299
 300             /* CUBIC SPLINE TABLE REPULSION */
 301             vfitab           = _mm_add_epi32(vfitab,ifour);
 302             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 303             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 304             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 305             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 306             _MM_TRANSPOSE4_PS(Y,F,G,H);
 307             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 308             VV               = _mm_macc_ps(vfeps,Fp,Y);
 309             vvdw12           = _mm_mul_ps(c12_00,VV);
 310             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 311             fvdw12           = _mm_mul_ps(c12_00,FF);
 312             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 313             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 317
 318             fscal            = fvdw;
 319
 320              /* Update vectorial force */
 321             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 322             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 323             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 324
 325             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 326             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 327             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             r10              = _mm_mul_ps(rsq10,rinv10);
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq10             = _mm_mul_ps(iq1,jq0);
 337
 338             /* EWALD ELECTROSTATICS */
 339
 340             /* Analytical PME correction */
 341             zeta2            = _mm_mul_ps(beta2,rsq10);
 342             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 343             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 344             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 345             felec            = _mm_mul_ps(qq10,felec);
 346             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 347             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 348             velec            = _mm_mul_ps(qq10,velec);
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velecsum         = _mm_add_ps(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355              /* Update vectorial force */
 356             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 357             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 358             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 359
 360             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 361             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 362             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             r20              = _mm_mul_ps(rsq20,rinv20);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq20             = _mm_mul_ps(iq2,jq0);
 372
 373             /* EWALD ELECTROSTATICS */
 374
 375             /* Analytical PME correction */
 376             zeta2            = _mm_mul_ps(beta2,rsq20);
 377             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 378             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 379             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 380             felec            = _mm_mul_ps(qq20,felec);
 381             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 382             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 383             velec            = _mm_mul_ps(qq20,velec);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velecsum         = _mm_add_ps(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390              /* Update vectorial force */
 391             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 392             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 393             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 394
 395             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 396             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 397             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             r30              = _mm_mul_ps(rsq30,rinv30);
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq30             = _mm_mul_ps(iq3,jq0);
 407
 408             /* EWALD ELECTROSTATICS */
 409
 410             /* Analytical PME correction */
 411             zeta2            = _mm_mul_ps(beta2,rsq30);
 412             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 413             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 414             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 415             felec            = _mm_mul_ps(qq30,felec);
 416             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 417             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 418             velec            = _mm_mul_ps(qq30,velec);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velecsum         = _mm_add_ps(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425              /* Update vectorial force */
 426             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 427             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 428             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 429
 430             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 431             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 432             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 433
 434             fjptrA             = f+j_coord_offsetA;
 435             fjptrB             = f+j_coord_offsetB;
 436             fjptrC             = f+j_coord_offsetC;
 437             fjptrD             = f+j_coord_offsetD;
 438
 439             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 440
 441             /* Inner loop uses 146 flops */
 442         }
 443
 444         if(jidx<j_index_end)
 445         {
 446
 447             /* Get j neighbor index, and coordinate index */
 448             jnrlistA         = jjnr[jidx];
 449             jnrlistB         = jjnr[jidx+1];
 450             jnrlistC         = jjnr[jidx+2];
 451             jnrlistD         = jjnr[jidx+3];
 452             /* Sign of each element will be negative for non-real atoms.
 453              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 454              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 455              */
 456             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 457             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 458             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 459             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 460             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 461             j_coord_offsetA  = DIM*jnrA;
 462             j_coord_offsetB  = DIM*jnrB;
 463             j_coord_offsetC  = DIM*jnrC;
 464             j_coord_offsetD  = DIM*jnrD;
 465
 466             /* load j atom coordinates */
 467             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 468                                               x+j_coord_offsetC,x+j_coord_offsetD,
 469                                               &jx0,&jy0,&jz0);
 470
 471             /* Calculate displacement vector */
 472             dx00             = _mm_sub_ps(ix0,jx0);
 473             dy00             = _mm_sub_ps(iy0,jy0);
 474             dz00             = _mm_sub_ps(iz0,jz0);
 475             dx10             = _mm_sub_ps(ix1,jx0);
 476             dy10             = _mm_sub_ps(iy1,jy0);
 477             dz10             = _mm_sub_ps(iz1,jz0);
 478             dx20             = _mm_sub_ps(ix2,jx0);
 479             dy20             = _mm_sub_ps(iy2,jy0);
 480             dz20             = _mm_sub_ps(iz2,jz0);
 481             dx30             = _mm_sub_ps(ix3,jx0);
 482             dy30             = _mm_sub_ps(iy3,jy0);
 483             dz30             = _mm_sub_ps(iz3,jz0);
 484
 485             /* Calculate squared distance and things based on it */
 486             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 487             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 488             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 489             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 490
 491             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 492             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 493             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 494             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 495
 496             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 497             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 498             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 499
 500             /* Load parameters for j particles */
 501             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 502                                                               charge+jnrC+0,charge+jnrD+0);
 503             vdwjidx0A        = 2*vdwtype[jnrA+0];
 504             vdwjidx0B        = 2*vdwtype[jnrB+0];
 505             vdwjidx0C        = 2*vdwtype[jnrC+0];
 506             vdwjidx0D        = 2*vdwtype[jnrD+0];
 507
 508             fjx0             = _mm_setzero_ps();
 509             fjy0             = _mm_setzero_ps();
 510             fjz0             = _mm_setzero_ps();
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             r00              = _mm_mul_ps(rsq00,rinv00);
 517             r00              = _mm_andnot_ps(dummy_mask,r00);
 518
 519             /* Compute parameters for interactions between i and j atoms */
 520             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 521                                          vdwparam+vdwioffset0+vdwjidx0B,
 522                                          vdwparam+vdwioffset0+vdwjidx0C,
 523                                          vdwparam+vdwioffset0+vdwjidx0D,
 524                                          &c6_00,&c12_00);
 525
 526             /* Calculate table index by multiplying r with table scale and truncate to integer */
 527             rt               = _mm_mul_ps(r00,vftabscale);
 528             vfitab           = _mm_cvttps_epi32(rt);
 529 #ifdef __XOP__
 530             vfeps            = _mm_frcz_ps(rt);
 531 #else
 532             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 533 #endif
 534             twovfeps         = _mm_add_ps(vfeps,vfeps);
 535             vfitab           = _mm_slli_epi32(vfitab,3);
 536
 537             /* CUBIC SPLINE TABLE DISPERSION */
 538             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 539             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 540             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 541             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 542             _MM_TRANSPOSE4_PS(Y,F,G,H);
 543             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 544             VV               = _mm_macc_ps(vfeps,Fp,Y);
 545             vvdw6            = _mm_mul_ps(c6_00,VV);
 546             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 547             fvdw6            = _mm_mul_ps(c6_00,FF);
 548
 549             /* CUBIC SPLINE TABLE REPULSION */
 550             vfitab           = _mm_add_epi32(vfitab,ifour);
 551             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 552             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 553             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 554             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 555             _MM_TRANSPOSE4_PS(Y,F,G,H);
 556             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 557             VV               = _mm_macc_ps(vfeps,Fp,Y);
 558             vvdw12           = _mm_mul_ps(c12_00,VV);
 559             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 560             fvdw12           = _mm_mul_ps(c12_00,FF);
 561             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 562             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 566             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 567
 568             fscal            = fvdw;
 569
 570             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 571
 572              /* Update vectorial force */
 573             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 574             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 575             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 576
 577             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 578             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 579             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             r10              = _mm_mul_ps(rsq10,rinv10);
 586             r10              = _mm_andnot_ps(dummy_mask,r10);
 587
 588             /* Compute parameters for interactions between i and j atoms */
 589             qq10             = _mm_mul_ps(iq1,jq0);
 590
 591             /* EWALD ELECTROSTATICS */
 592
 593             /* Analytical PME correction */
 594             zeta2            = _mm_mul_ps(beta2,rsq10);
 595             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 596             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 597             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 598             felec            = _mm_mul_ps(qq10,felec);
 599             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 600             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 601             velec            = _mm_mul_ps(qq10,velec);
 602
 603             /* Update potential sum for this i atom from the interaction with this j atom. */
 604             velec            = _mm_andnot_ps(dummy_mask,velec);
 605             velecsum         = _mm_add_ps(velecsum,velec);
 606
 607             fscal            = felec;
 608
 609             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 610
 611              /* Update vectorial force */
 612             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 613             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 614             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 615
 616             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 617             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 618             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             r20              = _mm_mul_ps(rsq20,rinv20);
 625             r20              = _mm_andnot_ps(dummy_mask,r20);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq20             = _mm_mul_ps(iq2,jq0);
 629
 630             /* EWALD ELECTROSTATICS */
 631
 632             /* Analytical PME correction */
 633             zeta2            = _mm_mul_ps(beta2,rsq20);
 634             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 635             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 636             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 637             felec            = _mm_mul_ps(qq20,felec);
 638             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 639             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 640             velec            = _mm_mul_ps(qq20,velec);
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velec            = _mm_andnot_ps(dummy_mask,velec);
 644             velecsum         = _mm_add_ps(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 649
 650              /* Update vectorial force */
 651             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 652             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 653             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 654
 655             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 656             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 657             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             r30              = _mm_mul_ps(rsq30,rinv30);
 664             r30              = _mm_andnot_ps(dummy_mask,r30);
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq30             = _mm_mul_ps(iq3,jq0);
 668
 669             /* EWALD ELECTROSTATICS */
 670
 671             /* Analytical PME correction */
 672             zeta2            = _mm_mul_ps(beta2,rsq30);
 673             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 674             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 675             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 676             felec            = _mm_mul_ps(qq30,felec);
 677             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 678             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 679             velec            = _mm_mul_ps(qq30,velec);
 680
 681             /* Update potential sum for this i atom from the interaction with this j atom. */
 682             velec            = _mm_andnot_ps(dummy_mask,velec);
 683             velecsum         = _mm_add_ps(velecsum,velec);
 684
 685             fscal            = felec;
 686
 687             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 688
 689              /* Update vectorial force */
 690             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 691             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 692             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 693
 694             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 695             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 696             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 697
 698             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 699             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 700             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 701             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 702
 703             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 704
 705             /* Inner loop uses 150 flops */
 706         }
 707
 708         /* End of innermost loop */
 709
 710         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 711                                               f+i_coord_offset,fshift+i_shift_offset);
 712
 713         ggid                        = gid[iidx];
 714         /* Update potential energies */
 715         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 716         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 717
 718         /* Increment number of inner iterations */
 719         inneriter                  += j_index_end - j_index_start;
 720
 721         /* Outer loop uses 26 flops */
 722     }
 723
 724     /* Increment number of outer iterations */
 725     outeriter        += nri;
 726
 727     /* Update outer/inner flops */
 728
 729     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*150);
 730 }
 731 /*
 732  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 733  * Electrostatics interaction: Ewald
 734  * VdW interaction:            CubicSplineTable
 735  * Geometry:                   Water4-Particle
 736  * Calculate force/pot:        Force
 737  */
 738 void
 739 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 740                     (t_nblist                    * gmx_restrict       nlist,
 741                      rvec                        * gmx_restrict          xx,
 742                      rvec                        * gmx_restrict          ff,
 743                      t_forcerec                  * gmx_restrict          fr,
 744                      t_mdatoms                   * gmx_restrict     mdatoms,
 745                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 746                      t_nrnb                      * gmx_restrict        nrnb)
 747 {
 748     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 749      * just 0 for non-waters.
 750      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 751      * jnr indices corresponding to data put in the four positions in the SIMD register.
 752      */
 753     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 754     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 755     int              jnrA,jnrB,jnrC,jnrD;
 756     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 757     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 758     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 759     real             rcutoff_scalar;
 760     real             *shiftvec,*fshift,*x,*f;
 761     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 762     real             scratch[4*DIM];
 763     __m128           fscal,rcutoff,rcutoff2,jidxall;
 764     int              vdwioffset0;
 765     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 766     int              vdwioffset1;
 767     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 768     int              vdwioffset2;
 769     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 770     int              vdwioffset3;
 771     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 772     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 773     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 774     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 775     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 776     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 777     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 778     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 779     real             *charge;
 780     int              nvdwtype;
 781     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 782     int              *vdwtype;
 783     real             *vdwparam;
 784     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 785     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 786     __m128i          vfitab;
 787     __m128i          ifour       = _mm_set1_epi32(4);
 788     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 789     real             *vftab;
 790     __m128i          ewitab;
 791     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 792     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 793     real             *ewtab;
 794     __m128           dummy_mask,cutoff_mask;
 795     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 796     __m128           one     = _mm_set1_ps(1.0);
 797     __m128           two     = _mm_set1_ps(2.0);
 798     x                = xx[0];
 799     f                = ff[0];
 800
 801     nri              = nlist->nri;
 802     iinr             = nlist->iinr;
 803     jindex           = nlist->jindex;
 804     jjnr             = nlist->jjnr;
 805     shiftidx         = nlist->shift;
 806     gid              = nlist->gid;
 807     shiftvec         = fr->shift_vec[0];
 808     fshift           = fr->fshift[0];
 809     facel            = _mm_set1_ps(fr->epsfac);
 810     charge           = mdatoms->chargeA;
 811     nvdwtype         = fr->ntype;
 812     vdwparam         = fr->nbfp;
 813     vdwtype          = mdatoms->typeA;
 814
 815     vftab            = kernel_data->table_vdw->data;
 816     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 817
 818     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 819     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 820     beta2            = _mm_mul_ps(beta,beta);
 821     beta3            = _mm_mul_ps(beta,beta2);
 822     ewtab            = fr->ic->tabq_coul_F;
 823     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 824     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 825
 826     /* Setup water-specific parameters */
 827     inr              = nlist->iinr[0];
 828     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 829     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 830     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 831     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 832
 833     /* Avoid stupid compiler warnings */
 834     jnrA = jnrB = jnrC = jnrD = 0;
 835     j_coord_offsetA = 0;
 836     j_coord_offsetB = 0;
 837     j_coord_offsetC = 0;
 838     j_coord_offsetD = 0;
 839
 840     outeriter        = 0;
 841     inneriter        = 0;
 842
 843     for(iidx=0;iidx<4*DIM;iidx++)
 844     {
 845         scratch[iidx] = 0.0;
 846     }
 847
 848     /* Start outer loop over neighborlists */
 849     for(iidx=0; iidx<nri; iidx++)
 850     {
 851         /* Load shift vector for this list */
 852         i_shift_offset   = DIM*shiftidx[iidx];
 853
 854         /* Load limits for loop over neighbors */
 855         j_index_start    = jindex[iidx];
 856         j_index_end      = jindex[iidx+1];
 857
 858         /* Get outer coordinate index */
 859         inr              = iinr[iidx];
 860         i_coord_offset   = DIM*inr;
 861
 862         /* Load i particle coords and add shift vector */
 863         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 864                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 865
 866         fix0             = _mm_setzero_ps();
 867         fiy0             = _mm_setzero_ps();
 868         fiz0             = _mm_setzero_ps();
 869         fix1             = _mm_setzero_ps();
 870         fiy1             = _mm_setzero_ps();
 871         fiz1             = _mm_setzero_ps();
 872         fix2             = _mm_setzero_ps();
 873         fiy2             = _mm_setzero_ps();
 874         fiz2             = _mm_setzero_ps();
 875         fix3             = _mm_setzero_ps();
 876         fiy3             = _mm_setzero_ps();
 877         fiz3             = _mm_setzero_ps();
 878
 879         /* Start inner kernel loop */
 880         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 881         {
 882
 883             /* Get j neighbor index, and coordinate index */
 884             jnrA             = jjnr[jidx];
 885             jnrB             = jjnr[jidx+1];
 886             jnrC             = jjnr[jidx+2];
 887             jnrD             = jjnr[jidx+3];
 888             j_coord_offsetA  = DIM*jnrA;
 889             j_coord_offsetB  = DIM*jnrB;
 890             j_coord_offsetC  = DIM*jnrC;
 891             j_coord_offsetD  = DIM*jnrD;
 892
 893             /* load j atom coordinates */
 894             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 895                                               x+j_coord_offsetC,x+j_coord_offsetD,
 896                                               &jx0,&jy0,&jz0);
 897
 898             /* Calculate displacement vector */
 899             dx00             = _mm_sub_ps(ix0,jx0);
 900             dy00             = _mm_sub_ps(iy0,jy0);
 901             dz00             = _mm_sub_ps(iz0,jz0);
 902             dx10             = _mm_sub_ps(ix1,jx0);
 903             dy10             = _mm_sub_ps(iy1,jy0);
 904             dz10             = _mm_sub_ps(iz1,jz0);
 905             dx20             = _mm_sub_ps(ix2,jx0);
 906             dy20             = _mm_sub_ps(iy2,jy0);
 907             dz20             = _mm_sub_ps(iz2,jz0);
 908             dx30             = _mm_sub_ps(ix3,jx0);
 909             dy30             = _mm_sub_ps(iy3,jy0);
 910             dz30             = _mm_sub_ps(iz3,jz0);
 911
 912             /* Calculate squared distance and things based on it */
 913             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 914             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 915             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 916             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 917
 918             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 919             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 920             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 921             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 922
 923             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 924             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 925             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 926
 927             /* Load parameters for j particles */
 928             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 929                                                               charge+jnrC+0,charge+jnrD+0);
 930             vdwjidx0A        = 2*vdwtype[jnrA+0];
 931             vdwjidx0B        = 2*vdwtype[jnrB+0];
 932             vdwjidx0C        = 2*vdwtype[jnrC+0];
 933             vdwjidx0D        = 2*vdwtype[jnrD+0];
 934
 935             fjx0             = _mm_setzero_ps();
 936             fjy0             = _mm_setzero_ps();
 937             fjz0             = _mm_setzero_ps();
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             r00              = _mm_mul_ps(rsq00,rinv00);
 944
 945             /* Compute parameters for interactions between i and j atoms */
 946             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 947                                          vdwparam+vdwioffset0+vdwjidx0B,
 948                                          vdwparam+vdwioffset0+vdwjidx0C,
 949                                          vdwparam+vdwioffset0+vdwjidx0D,
 950                                          &c6_00,&c12_00);
 951
 952             /* Calculate table index by multiplying r with table scale and truncate to integer */
 953             rt               = _mm_mul_ps(r00,vftabscale);
 954             vfitab           = _mm_cvttps_epi32(rt);
 955 #ifdef __XOP__
 956             vfeps            = _mm_frcz_ps(rt);
 957 #else
 958             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 959 #endif
 960             twovfeps         = _mm_add_ps(vfeps,vfeps);
 961             vfitab           = _mm_slli_epi32(vfitab,3);
 962
 963             /* CUBIC SPLINE TABLE DISPERSION */
 964             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 965             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 966             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 967             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 968             _MM_TRANSPOSE4_PS(Y,F,G,H);
 969             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 970             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 971             fvdw6            = _mm_mul_ps(c6_00,FF);
 972
 973             /* CUBIC SPLINE TABLE REPULSION */
 974             vfitab           = _mm_add_epi32(vfitab,ifour);
 975             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 976             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 977             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 978             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 979             _MM_TRANSPOSE4_PS(Y,F,G,H);
 980             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 981             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 982             fvdw12           = _mm_mul_ps(c12_00,FF);
 983             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 984
 985             fscal            = fvdw;
 986
 987              /* Update vectorial force */
 988             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 989             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 990             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 991
 992             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 993             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 994             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             r10              = _mm_mul_ps(rsq10,rinv10);
1001
1002             /* Compute parameters for interactions between i and j atoms */
1003             qq10             = _mm_mul_ps(iq1,jq0);
1004
1005             /* EWALD ELECTROSTATICS */
1006
1007             /* Analytical PME correction */
1008             zeta2            = _mm_mul_ps(beta2,rsq10);
1009             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1010             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1011             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1012             felec            = _mm_mul_ps(qq10,felec);
1013
1014             fscal            = felec;
1015
1016              /* Update vectorial force */
1017             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1018             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1019             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1020
1021             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1022             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1023             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             r20              = _mm_mul_ps(rsq20,rinv20);
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq20             = _mm_mul_ps(iq2,jq0);
1033
1034             /* EWALD ELECTROSTATICS */
1035
1036             /* Analytical PME correction */
1037             zeta2            = _mm_mul_ps(beta2,rsq20);
1038             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1039             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1040             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1041             felec            = _mm_mul_ps(qq20,felec);
1042
1043             fscal            = felec;
1044
1045              /* Update vectorial force */
1046             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1047             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1048             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1049
1050             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1051             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1052             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             r30              = _mm_mul_ps(rsq30,rinv30);
1059
1060             /* Compute parameters for interactions between i and j atoms */
1061             qq30             = _mm_mul_ps(iq3,jq0);
1062
1063             /* EWALD ELECTROSTATICS */
1064
1065             /* Analytical PME correction */
1066             zeta2            = _mm_mul_ps(beta2,rsq30);
1067             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1068             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1069             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1070             felec            = _mm_mul_ps(qq30,felec);
1071
1072             fscal            = felec;
1073
1074              /* Update vectorial force */
1075             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1076             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1077             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1078
1079             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1080             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1081             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1082
1083             fjptrA             = f+j_coord_offsetA;
1084             fjptrB             = f+j_coord_offsetB;
1085             fjptrC             = f+j_coord_offsetC;
1086             fjptrD             = f+j_coord_offsetD;
1087
1088             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1089
1090             /* Inner loop uses 135 flops */
1091         }
1092
1093         if(jidx<j_index_end)
1094         {
1095
1096             /* Get j neighbor index, and coordinate index */
1097             jnrlistA         = jjnr[jidx];
1098             jnrlistB         = jjnr[jidx+1];
1099             jnrlistC         = jjnr[jidx+2];
1100             jnrlistD         = jjnr[jidx+3];
1101             /* Sign of each element will be negative for non-real atoms.
1102              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1103              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1104              */
1105             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1106             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1107             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1108             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1109             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1110             j_coord_offsetA  = DIM*jnrA;
1111             j_coord_offsetB  = DIM*jnrB;
1112             j_coord_offsetC  = DIM*jnrC;
1113             j_coord_offsetD  = DIM*jnrD;
1114
1115             /* load j atom coordinates */
1116             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1117                                               x+j_coord_offsetC,x+j_coord_offsetD,
1118                                               &jx0,&jy0,&jz0);
1119
1120             /* Calculate displacement vector */
1121             dx00             = _mm_sub_ps(ix0,jx0);
1122             dy00             = _mm_sub_ps(iy0,jy0);
1123             dz00             = _mm_sub_ps(iz0,jz0);
1124             dx10             = _mm_sub_ps(ix1,jx0);
1125             dy10             = _mm_sub_ps(iy1,jy0);
1126             dz10             = _mm_sub_ps(iz1,jz0);
1127             dx20             = _mm_sub_ps(ix2,jx0);
1128             dy20             = _mm_sub_ps(iy2,jy0);
1129             dz20             = _mm_sub_ps(iz2,jz0);
1130             dx30             = _mm_sub_ps(ix3,jx0);
1131             dy30             = _mm_sub_ps(iy3,jy0);
1132             dz30             = _mm_sub_ps(iz3,jz0);
1133
1134             /* Calculate squared distance and things based on it */
1135             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1136             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1137             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1138             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1139
1140             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1141             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1142             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1143             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1144
1145             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1146             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1147             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1148
1149             /* Load parameters for j particles */
1150             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1151                                                               charge+jnrC+0,charge+jnrD+0);
1152             vdwjidx0A        = 2*vdwtype[jnrA+0];
1153             vdwjidx0B        = 2*vdwtype[jnrB+0];
1154             vdwjidx0C        = 2*vdwtype[jnrC+0];
1155             vdwjidx0D        = 2*vdwtype[jnrD+0];
1156
1157             fjx0             = _mm_setzero_ps();
1158             fjy0             = _mm_setzero_ps();
1159             fjz0             = _mm_setzero_ps();
1160
1161             /**************************
1162              * CALCULATE INTERACTIONS *
1163              **************************/
1164
1165             r00              = _mm_mul_ps(rsq00,rinv00);
1166             r00              = _mm_andnot_ps(dummy_mask,r00);
1167
1168             /* Compute parameters for interactions between i and j atoms */
1169             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1170                                          vdwparam+vdwioffset0+vdwjidx0B,
1171                                          vdwparam+vdwioffset0+vdwjidx0C,
1172                                          vdwparam+vdwioffset0+vdwjidx0D,
1173                                          &c6_00,&c12_00);
1174
1175             /* Calculate table index by multiplying r with table scale and truncate to integer */
1176             rt               = _mm_mul_ps(r00,vftabscale);
1177             vfitab           = _mm_cvttps_epi32(rt);
1178 #ifdef __XOP__
1179             vfeps            = _mm_frcz_ps(rt);
1180 #else
1181             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1182 #endif
1183             twovfeps         = _mm_add_ps(vfeps,vfeps);
1184             vfitab           = _mm_slli_epi32(vfitab,3);
1185
1186             /* CUBIC SPLINE TABLE DISPERSION */
1187             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1188             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1189             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1190             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1191             _MM_TRANSPOSE4_PS(Y,F,G,H);
1192             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1193             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1194             fvdw6            = _mm_mul_ps(c6_00,FF);
1195
1196             /* CUBIC SPLINE TABLE REPULSION */
1197             vfitab           = _mm_add_epi32(vfitab,ifour);
1198             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1199             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1200             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1201             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1202             _MM_TRANSPOSE4_PS(Y,F,G,H);
1203             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1204             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1205             fvdw12           = _mm_mul_ps(c12_00,FF);
1206             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1207
1208             fscal            = fvdw;
1209
1210             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1211
1212              /* Update vectorial force */
1213             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1214             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1215             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1216
1217             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1218             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1219             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1220
1221             /**************************
1222              * CALCULATE INTERACTIONS *
1223              **************************/
1224
1225             r10              = _mm_mul_ps(rsq10,rinv10);
1226             r10              = _mm_andnot_ps(dummy_mask,r10);
1227
1228             /* Compute parameters for interactions between i and j atoms */
1229             qq10             = _mm_mul_ps(iq1,jq0);
1230
1231             /* EWALD ELECTROSTATICS */
1232
1233             /* Analytical PME correction */
1234             zeta2            = _mm_mul_ps(beta2,rsq10);
1235             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1236             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1237             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1238             felec            = _mm_mul_ps(qq10,felec);
1239
1240             fscal            = felec;
1241
1242             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1243
1244              /* Update vectorial force */
1245             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1246             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1247             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1248
1249             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1250             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1251             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1252
1253             /**************************
1254              * CALCULATE INTERACTIONS *
1255              **************************/
1256
1257             r20              = _mm_mul_ps(rsq20,rinv20);
1258             r20              = _mm_andnot_ps(dummy_mask,r20);
1259
1260             /* Compute parameters for interactions between i and j atoms */
1261             qq20             = _mm_mul_ps(iq2,jq0);
1262
1263             /* EWALD ELECTROSTATICS */
1264
1265             /* Analytical PME correction */
1266             zeta2            = _mm_mul_ps(beta2,rsq20);
1267             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1268             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1269             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1270             felec            = _mm_mul_ps(qq20,felec);
1271
1272             fscal            = felec;
1273
1274             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1275
1276              /* Update vectorial force */
1277             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1278             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1279             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1280
1281             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1282             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1283             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1284
1285             /**************************
1286              * CALCULATE INTERACTIONS *
1287              **************************/
1288
1289             r30              = _mm_mul_ps(rsq30,rinv30);
1290             r30              = _mm_andnot_ps(dummy_mask,r30);
1291
1292             /* Compute parameters for interactions between i and j atoms */
1293             qq30             = _mm_mul_ps(iq3,jq0);
1294
1295             /* EWALD ELECTROSTATICS */
1296
1297             /* Analytical PME correction */
1298             zeta2            = _mm_mul_ps(beta2,rsq30);
1299             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1300             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1301             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1302             felec            = _mm_mul_ps(qq30,felec);
1303
1304             fscal            = felec;
1305
1306             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1307
1308              /* Update vectorial force */
1309             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1310             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1311             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1312
1313             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1314             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1315             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1316
1317             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1318             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1319             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1320             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1321
1322             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1323
1324             /* Inner loop uses 139 flops */
1325         }
1326
1327         /* End of innermost loop */
1328
1329         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1330                                               f+i_coord_offset,fshift+i_shift_offset);
1331
1332         /* Increment number of inner iterations */
1333         inneriter                  += j_index_end - j_index_start;
1334
1335         /* Outer loop uses 24 flops */
1336     }
1337
1338     /* Increment number of outer iterations */
1339     outeriter        += nri;
1340
1341     /* Update outer/inner flops */
1342
1343     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*139);
1344 }