src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJ_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128i          ewitab;
  97     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 122     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 123     beta2            = _mm_mul_ps(beta,beta);
 124     beta3            = _mm_mul_ps(beta,beta2);
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 160
 161         fix0             = _mm_setzero_ps();
 162         fiy0             = _mm_setzero_ps();
 163         fiz0             = _mm_setzero_ps();
 164
 165         /* Load parameters for i particles */
 166         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 167         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_ps();
 171         vvdwsum          = _mm_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184             j_coord_offsetC  = DIM*jnrC;
 185             j_coord_offsetD  = DIM*jnrD;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               x+j_coord_offsetC,x+j_coord_offsetD,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_ps(ix0,jx0);
 194             dy00             = _mm_sub_ps(iy0,jy0);
 195             dz00             = _mm_sub_ps(iz0,jz0);
 196
 197             /* Calculate squared distance and things based on it */
 198             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 199
 200             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 201
 202             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 203
 204             /* Load parameters for j particles */
 205             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 206                                                               charge+jnrC+0,charge+jnrD+0);
 207             vdwjidx0A        = 2*vdwtype[jnrA+0];
 208             vdwjidx0B        = 2*vdwtype[jnrB+0];
 209             vdwjidx0C        = 2*vdwtype[jnrC+0];
 210             vdwjidx0D        = 2*vdwtype[jnrD+0];
 211
 212             /**************************
 213              * CALCULATE INTERACTIONS *
 214              **************************/
 215
 216             r00              = _mm_mul_ps(rsq00,rinv00);
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm_mul_ps(iq0,jq0);
 220             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 221                                          vdwparam+vdwioffset0+vdwjidx0B,
 222                                          vdwparam+vdwioffset0+vdwjidx0C,
 223                                          vdwparam+vdwioffset0+vdwjidx0D,
 224                                          &c6_00,&c12_00);
 225
 226             /* EWALD ELECTROSTATICS */
 227
 228             /* Analytical PME correction */
 229             zeta2            = _mm_mul_ps(beta2,rsq00);
 230             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 231             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 232             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 233             felec            = _mm_mul_ps(qq00,felec);
 234             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 235             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 236             velec            = _mm_mul_ps(qq00,velec);
 237
 238             /* LENNARD-JONES DISPERSION/REPULSION */
 239
 240             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 241             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 242             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 243             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 244             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velecsum         = _mm_add_ps(velecsum,velec);
 248             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 249
 250             fscal            = _mm_add_ps(felec,fvdw);
 251
 252              /* Update vectorial force */
 253             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 254             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 255             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 256
 257             fjptrA             = f+j_coord_offsetA;
 258             fjptrB             = f+j_coord_offsetB;
 259             fjptrC             = f+j_coord_offsetC;
 260             fjptrD             = f+j_coord_offsetD;
 261             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 262                                                    _mm_mul_ps(dx00,fscal),
 263                                                    _mm_mul_ps(dy00,fscal),
 264                                                    _mm_mul_ps(dz00,fscal));
 265
 266             /* Inner loop uses 41 flops */
 267         }
 268
 269         if(jidx<j_index_end)
 270         {
 271
 272             /* Get j neighbor index, and coordinate index */
 273             jnrlistA         = jjnr[jidx];
 274             jnrlistB         = jjnr[jidx+1];
 275             jnrlistC         = jjnr[jidx+2];
 276             jnrlistD         = jjnr[jidx+3];
 277             /* Sign of each element will be negative for non-real atoms.
 278              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 279              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 280              */
 281             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 282             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 283             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 284             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 285             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 286             j_coord_offsetA  = DIM*jnrA;
 287             j_coord_offsetB  = DIM*jnrB;
 288             j_coord_offsetC  = DIM*jnrC;
 289             j_coord_offsetD  = DIM*jnrD;
 290
 291             /* load j atom coordinates */
 292             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 293                                               x+j_coord_offsetC,x+j_coord_offsetD,
 294                                               &jx0,&jy0,&jz0);
 295
 296             /* Calculate displacement vector */
 297             dx00             = _mm_sub_ps(ix0,jx0);
 298             dy00             = _mm_sub_ps(iy0,jy0);
 299             dz00             = _mm_sub_ps(iz0,jz0);
 300
 301             /* Calculate squared distance and things based on it */
 302             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 303
 304             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 305
 306             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 307
 308             /* Load parameters for j particles */
 309             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 310                                                               charge+jnrC+0,charge+jnrD+0);
 311             vdwjidx0A        = 2*vdwtype[jnrA+0];
 312             vdwjidx0B        = 2*vdwtype[jnrB+0];
 313             vdwjidx0C        = 2*vdwtype[jnrC+0];
 314             vdwjidx0D        = 2*vdwtype[jnrD+0];
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r00              = _mm_mul_ps(rsq00,rinv00);
 321             r00              = _mm_andnot_ps(dummy_mask,r00);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq00             = _mm_mul_ps(iq0,jq0);
 325             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 326                                          vdwparam+vdwioffset0+vdwjidx0B,
 327                                          vdwparam+vdwioffset0+vdwjidx0C,
 328                                          vdwparam+vdwioffset0+vdwjidx0D,
 329                                          &c6_00,&c12_00);
 330
 331             /* EWALD ELECTROSTATICS */
 332
 333             /* Analytical PME correction */
 334             zeta2            = _mm_mul_ps(beta2,rsq00);
 335             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 336             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 337             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 338             felec            = _mm_mul_ps(qq00,felec);
 339             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 340             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 341             velec            = _mm_mul_ps(qq00,velec);
 342
 343             /* LENNARD-JONES DISPERSION/REPULSION */
 344
 345             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 346             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 347             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 348             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 349             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velec            = _mm_andnot_ps(dummy_mask,velec);
 353             velecsum         = _mm_add_ps(velecsum,velec);
 354             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 355             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 356
 357             fscal            = _mm_add_ps(felec,fvdw);
 358
 359             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 360
 361              /* Update vectorial force */
 362             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 363             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 364             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 365
 366             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 367             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 368             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 369             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 370             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 371                                                    _mm_mul_ps(dx00,fscal),
 372                                                    _mm_mul_ps(dy00,fscal),
 373                                                    _mm_mul_ps(dz00,fscal));
 374
 375             /* Inner loop uses 42 flops */
 376         }
 377
 378         /* End of innermost loop */
 379
 380         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 381                                               f+i_coord_offset,fshift+i_shift_offset);
 382
 383         ggid                        = gid[iidx];
 384         /* Update potential energies */
 385         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 386         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 387
 388         /* Increment number of inner iterations */
 389         inneriter                  += j_index_end - j_index_start;
 390
 391         /* Outer loop uses 9 flops */
 392     }
 393
 394     /* Increment number of outer iterations */
 395     outeriter        += nri;
 396
 397     /* Update outer/inner flops */
 398
 399     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*42);
 400 }
 401 /*
 402  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_single
 403  * Electrostatics interaction: Ewald
 404  * VdW interaction:            LennardJones
 405  * Geometry:                   Particle-Particle
 406  * Calculate force/pot:        Force
 407  */
 408 void
 409 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_single
 410                     (t_nblist                    * gmx_restrict       nlist,
 411                      rvec                        * gmx_restrict          xx,
 412                      rvec                        * gmx_restrict          ff,
 413                      t_forcerec                  * gmx_restrict          fr,
 414                      t_mdatoms                   * gmx_restrict     mdatoms,
 415                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 416                      t_nrnb                      * gmx_restrict        nrnb)
 417 {
 418     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 419      * just 0 for non-waters.
 420      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 421      * jnr indices corresponding to data put in the four positions in the SIMD register.
 422      */
 423     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 424     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 425     int              jnrA,jnrB,jnrC,jnrD;
 426     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 427     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 428     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 429     real             rcutoff_scalar;
 430     real             *shiftvec,*fshift,*x,*f;
 431     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 432     real             scratch[4*DIM];
 433     __m128           fscal,rcutoff,rcutoff2,jidxall;
 434     int              vdwioffset0;
 435     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 436     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 437     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 438     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 439     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 440     real             *charge;
 441     int              nvdwtype;
 442     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 443     int              *vdwtype;
 444     real             *vdwparam;
 445     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 446     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 447     __m128i          ewitab;
 448     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 449     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 450     real             *ewtab;
 451     __m128           dummy_mask,cutoff_mask;
 452     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 453     __m128           one     = _mm_set1_ps(1.0);
 454     __m128           two     = _mm_set1_ps(2.0);
 455     x                = xx[0];
 456     f                = ff[0];
 457
 458     nri              = nlist->nri;
 459     iinr             = nlist->iinr;
 460     jindex           = nlist->jindex;
 461     jjnr             = nlist->jjnr;
 462     shiftidx         = nlist->shift;
 463     gid              = nlist->gid;
 464     shiftvec         = fr->shift_vec[0];
 465     fshift           = fr->fshift[0];
 466     facel            = _mm_set1_ps(fr->epsfac);
 467     charge           = mdatoms->chargeA;
 468     nvdwtype         = fr->ntype;
 469     vdwparam         = fr->nbfp;
 470     vdwtype          = mdatoms->typeA;
 471
 472     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 473     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 474     beta2            = _mm_mul_ps(beta,beta);
 475     beta3            = _mm_mul_ps(beta,beta2);
 476     ewtab            = fr->ic->tabq_coul_F;
 477     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 478     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 479
 480     /* Avoid stupid compiler warnings */
 481     jnrA = jnrB = jnrC = jnrD = 0;
 482     j_coord_offsetA = 0;
 483     j_coord_offsetB = 0;
 484     j_coord_offsetC = 0;
 485     j_coord_offsetD = 0;
 486
 487     outeriter        = 0;
 488     inneriter        = 0;
 489
 490     for(iidx=0;iidx<4*DIM;iidx++)
 491     {
 492         scratch[iidx] = 0.0;
 493     }
 494
 495     /* Start outer loop over neighborlists */
 496     for(iidx=0; iidx<nri; iidx++)
 497     {
 498         /* Load shift vector for this list */
 499         i_shift_offset   = DIM*shiftidx[iidx];
 500
 501         /* Load limits for loop over neighbors */
 502         j_index_start    = jindex[iidx];
 503         j_index_end      = jindex[iidx+1];
 504
 505         /* Get outer coordinate index */
 506         inr              = iinr[iidx];
 507         i_coord_offset   = DIM*inr;
 508
 509         /* Load i particle coords and add shift vector */
 510         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 511
 512         fix0             = _mm_setzero_ps();
 513         fiy0             = _mm_setzero_ps();
 514         fiz0             = _mm_setzero_ps();
 515
 516         /* Load parameters for i particles */
 517         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 518         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 519
 520         /* Start inner kernel loop */
 521         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 522         {
 523
 524             /* Get j neighbor index, and coordinate index */
 525             jnrA             = jjnr[jidx];
 526             jnrB             = jjnr[jidx+1];
 527             jnrC             = jjnr[jidx+2];
 528             jnrD             = jjnr[jidx+3];
 529             j_coord_offsetA  = DIM*jnrA;
 530             j_coord_offsetB  = DIM*jnrB;
 531             j_coord_offsetC  = DIM*jnrC;
 532             j_coord_offsetD  = DIM*jnrD;
 533
 534             /* load j atom coordinates */
 535             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 536                                               x+j_coord_offsetC,x+j_coord_offsetD,
 537                                               &jx0,&jy0,&jz0);
 538
 539             /* Calculate displacement vector */
 540             dx00             = _mm_sub_ps(ix0,jx0);
 541             dy00             = _mm_sub_ps(iy0,jy0);
 542             dz00             = _mm_sub_ps(iz0,jz0);
 543
 544             /* Calculate squared distance and things based on it */
 545             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 546
 547             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 548
 549             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 550
 551             /* Load parameters for j particles */
 552             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 553                                                               charge+jnrC+0,charge+jnrD+0);
 554             vdwjidx0A        = 2*vdwtype[jnrA+0];
 555             vdwjidx0B        = 2*vdwtype[jnrB+0];
 556             vdwjidx0C        = 2*vdwtype[jnrC+0];
 557             vdwjidx0D        = 2*vdwtype[jnrD+0];
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r00              = _mm_mul_ps(rsq00,rinv00);
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             qq00             = _mm_mul_ps(iq0,jq0);
 567             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 568                                          vdwparam+vdwioffset0+vdwjidx0B,
 569                                          vdwparam+vdwioffset0+vdwjidx0C,
 570                                          vdwparam+vdwioffset0+vdwjidx0D,
 571                                          &c6_00,&c12_00);
 572
 573             /* EWALD ELECTROSTATICS */
 574
 575             /* Analytical PME correction */
 576             zeta2            = _mm_mul_ps(beta2,rsq00);
 577             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 578             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 579             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 580             felec            = _mm_mul_ps(qq00,felec);
 581
 582             /* LENNARD-JONES DISPERSION/REPULSION */
 583
 584             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 585             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 586
 587             fscal            = _mm_add_ps(felec,fvdw);
 588
 589              /* Update vectorial force */
 590             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 591             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 592             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 593
 594             fjptrA             = f+j_coord_offsetA;
 595             fjptrB             = f+j_coord_offsetB;
 596             fjptrC             = f+j_coord_offsetC;
 597             fjptrD             = f+j_coord_offsetD;
 598             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 599                                                    _mm_mul_ps(dx00,fscal),
 600                                                    _mm_mul_ps(dy00,fscal),
 601                                                    _mm_mul_ps(dz00,fscal));
 602
 603             /* Inner loop uses 35 flops */
 604         }
 605
 606         if(jidx<j_index_end)
 607         {
 608
 609             /* Get j neighbor index, and coordinate index */
 610             jnrlistA         = jjnr[jidx];
 611             jnrlistB         = jjnr[jidx+1];
 612             jnrlistC         = jjnr[jidx+2];
 613             jnrlistD         = jjnr[jidx+3];
 614             /* Sign of each element will be negative for non-real atoms.
 615              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 616              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 617              */
 618             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 619             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 620             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 621             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 622             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 623             j_coord_offsetA  = DIM*jnrA;
 624             j_coord_offsetB  = DIM*jnrB;
 625             j_coord_offsetC  = DIM*jnrC;
 626             j_coord_offsetD  = DIM*jnrD;
 627
 628             /* load j atom coordinates */
 629             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 630                                               x+j_coord_offsetC,x+j_coord_offsetD,
 631                                               &jx0,&jy0,&jz0);
 632
 633             /* Calculate displacement vector */
 634             dx00             = _mm_sub_ps(ix0,jx0);
 635             dy00             = _mm_sub_ps(iy0,jy0);
 636             dz00             = _mm_sub_ps(iz0,jz0);
 637
 638             /* Calculate squared distance and things based on it */
 639             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 640
 641             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 642
 643             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 644
 645             /* Load parameters for j particles */
 646             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 647                                                               charge+jnrC+0,charge+jnrD+0);
 648             vdwjidx0A        = 2*vdwtype[jnrA+0];
 649             vdwjidx0B        = 2*vdwtype[jnrB+0];
 650             vdwjidx0C        = 2*vdwtype[jnrC+0];
 651             vdwjidx0D        = 2*vdwtype[jnrD+0];
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             r00              = _mm_mul_ps(rsq00,rinv00);
 658             r00              = _mm_andnot_ps(dummy_mask,r00);
 659
 660             /* Compute parameters for interactions between i and j atoms */
 661             qq00             = _mm_mul_ps(iq0,jq0);
 662             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 663                                          vdwparam+vdwioffset0+vdwjidx0B,
 664                                          vdwparam+vdwioffset0+vdwjidx0C,
 665                                          vdwparam+vdwioffset0+vdwjidx0D,
 666                                          &c6_00,&c12_00);
 667
 668             /* EWALD ELECTROSTATICS */
 669
 670             /* Analytical PME correction */
 671             zeta2            = _mm_mul_ps(beta2,rsq00);
 672             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 673             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 674             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 675             felec            = _mm_mul_ps(qq00,felec);
 676
 677             /* LENNARD-JONES DISPERSION/REPULSION */
 678
 679             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 680             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 681
 682             fscal            = _mm_add_ps(felec,fvdw);
 683
 684             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 685
 686              /* Update vectorial force */
 687             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 688             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 689             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 690
 691             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 692             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 693             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 694             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 695             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 696                                                    _mm_mul_ps(dx00,fscal),
 697                                                    _mm_mul_ps(dy00,fscal),
 698                                                    _mm_mul_ps(dz00,fscal));
 699
 700             /* Inner loop uses 36 flops */
 701         }
 702
 703         /* End of innermost loop */
 704
 705         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 706                                               f+i_coord_offset,fshift+i_shift_offset);
 707
 708         /* Increment number of inner iterations */
 709         inneriter                  += j_index_end - j_index_start;
 710
 711         /* Outer loop uses 7 flops */
 712     }
 713
 714     /* Increment number of outer iterations */
 715     outeriter        += nri;
 716
 717     /* Update outer/inner flops */
 718
 719     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*36);
 720 }