src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJ_GeomW4P1_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomW4P1_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              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 104     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 105     __m128i          ewitab;
 106     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 108     real             *ewtab;
 109     __m128           dummy_mask,cutoff_mask;
 110     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 111     __m128           one     = _mm_set1_ps(1.0);
 112     __m128           two     = _mm_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_ps(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 131     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 132     beta2            = _mm_mul_ps(beta,beta);
 133     beta3            = _mm_mul_ps(beta,beta2);
 134     ewtab            = fr->ic->tabq_coul_FDV0;
 135     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 136     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 141     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 142     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 143     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm_setzero_ps();
 179         fiy0             = _mm_setzero_ps();
 180         fiz0             = _mm_setzero_ps();
 181         fix1             = _mm_setzero_ps();
 182         fiy1             = _mm_setzero_ps();
 183         fiz1             = _mm_setzero_ps();
 184         fix2             = _mm_setzero_ps();
 185         fiy2             = _mm_setzero_ps();
 186         fiz2             = _mm_setzero_ps();
 187         fix3             = _mm_setzero_ps();
 188         fiy3             = _mm_setzero_ps();
 189         fiz3             = _mm_setzero_ps();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_ps();
 193         vvdwsum          = _mm_setzero_ps();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                               x+j_coord_offsetC,x+j_coord_offsetD,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm_sub_ps(ix0,jx0);
 216             dy00             = _mm_sub_ps(iy0,jy0);
 217             dz00             = _mm_sub_ps(iz0,jz0);
 218             dx10             = _mm_sub_ps(ix1,jx0);
 219             dy10             = _mm_sub_ps(iy1,jy0);
 220             dz10             = _mm_sub_ps(iz1,jz0);
 221             dx20             = _mm_sub_ps(ix2,jx0);
 222             dy20             = _mm_sub_ps(iy2,jy0);
 223             dz20             = _mm_sub_ps(iz2,jz0);
 224             dx30             = _mm_sub_ps(ix3,jx0);
 225             dy30             = _mm_sub_ps(iy3,jy0);
 226             dz30             = _mm_sub_ps(iz3,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 230             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 231             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 232             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 233
 234             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 235             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 236             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 237
 238             rinvsq00         = gmx_mm_inv_ps(rsq00);
 239             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 240             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 241             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 245                                                               charge+jnrC+0,charge+jnrD+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248             vdwjidx0C        = 2*vdwtype[jnrC+0];
 249             vdwjidx0D        = 2*vdwtype[jnrD+0];
 250
 251             fjx0             = _mm_setzero_ps();
 252             fjy0             = _mm_setzero_ps();
 253             fjz0             = _mm_setzero_ps();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 261                                          vdwparam+vdwioffset0+vdwjidx0B,
 262                                          vdwparam+vdwioffset0+vdwjidx0C,
 263                                          vdwparam+vdwioffset0+vdwjidx0D,
 264                                          &c6_00,&c12_00);
 265
 266             /* LENNARD-JONES DISPERSION/REPULSION */
 267
 268             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 269             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 270             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 271             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 272             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 276
 277             fscal            = fvdw;
 278
 279              /* Update vectorial force */
 280             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 281             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 282             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 283
 284             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 285             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 286             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             r10              = _mm_mul_ps(rsq10,rinv10);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq10             = _mm_mul_ps(iq1,jq0);
 296
 297             /* EWALD ELECTROSTATICS */
 298
 299             /* Analytical PME correction */
 300             zeta2            = _mm_mul_ps(beta2,rsq10);
 301             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 302             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 303             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 304             felec            = _mm_mul_ps(qq10,felec);
 305             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 306             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 307             velec            = _mm_mul_ps(qq10,velec);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velecsum         = _mm_add_ps(velecsum,velec);
 311
 312             fscal            = felec;
 313
 314              /* Update vectorial force */
 315             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 316             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 317             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 318
 319             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 320             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 321             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             r20              = _mm_mul_ps(rsq20,rinv20);
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq20             = _mm_mul_ps(iq2,jq0);
 331
 332             /* EWALD ELECTROSTATICS */
 333
 334             /* Analytical PME correction */
 335             zeta2            = _mm_mul_ps(beta2,rsq20);
 336             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 337             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 338             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 339             felec            = _mm_mul_ps(qq20,felec);
 340             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 341             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 342             velec            = _mm_mul_ps(qq20,velec);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349              /* Update vectorial force */
 350             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 351             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 352             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 353
 354             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 355             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 356             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             r30              = _mm_mul_ps(rsq30,rinv30);
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq30             = _mm_mul_ps(iq3,jq0);
 366
 367             /* EWALD ELECTROSTATICS */
 368
 369             /* Analytical PME correction */
 370             zeta2            = _mm_mul_ps(beta2,rsq30);
 371             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 372             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 373             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 374             felec            = _mm_mul_ps(qq30,felec);
 375             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 376             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 377             velec            = _mm_mul_ps(qq30,velec);
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384              /* Update vectorial force */
 385             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 386             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 387             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 388
 389             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 390             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 391             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 392
 393             fjptrA             = f+j_coord_offsetA;
 394             fjptrB             = f+j_coord_offsetB;
 395             fjptrC             = f+j_coord_offsetC;
 396             fjptrD             = f+j_coord_offsetD;
 397
 398             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 399
 400             /* Inner loop uses 122 flops */
 401         }
 402
 403         if(jidx<j_index_end)
 404         {
 405
 406             /* Get j neighbor index, and coordinate index */
 407             jnrlistA         = jjnr[jidx];
 408             jnrlistB         = jjnr[jidx+1];
 409             jnrlistC         = jjnr[jidx+2];
 410             jnrlistD         = jjnr[jidx+3];
 411             /* Sign of each element will be negative for non-real atoms.
 412              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 413              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 414              */
 415             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 416             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 417             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 418             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 419             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 420             j_coord_offsetA  = DIM*jnrA;
 421             j_coord_offsetB  = DIM*jnrB;
 422             j_coord_offsetC  = DIM*jnrC;
 423             j_coord_offsetD  = DIM*jnrD;
 424
 425             /* load j atom coordinates */
 426             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                               x+j_coord_offsetC,x+j_coord_offsetD,
 428                                               &jx0,&jy0,&jz0);
 429
 430             /* Calculate displacement vector */
 431             dx00             = _mm_sub_ps(ix0,jx0);
 432             dy00             = _mm_sub_ps(iy0,jy0);
 433             dz00             = _mm_sub_ps(iz0,jz0);
 434             dx10             = _mm_sub_ps(ix1,jx0);
 435             dy10             = _mm_sub_ps(iy1,jy0);
 436             dz10             = _mm_sub_ps(iz1,jz0);
 437             dx20             = _mm_sub_ps(ix2,jx0);
 438             dy20             = _mm_sub_ps(iy2,jy0);
 439             dz20             = _mm_sub_ps(iz2,jz0);
 440             dx30             = _mm_sub_ps(ix3,jx0);
 441             dy30             = _mm_sub_ps(iy3,jy0);
 442             dz30             = _mm_sub_ps(iz3,jz0);
 443
 444             /* Calculate squared distance and things based on it */
 445             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 446             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 447             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 448             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 449
 450             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 451             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 452             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 453
 454             rinvsq00         = gmx_mm_inv_ps(rsq00);
 455             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 456             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 457             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 458
 459             /* Load parameters for j particles */
 460             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 461                                                               charge+jnrC+0,charge+jnrD+0);
 462             vdwjidx0A        = 2*vdwtype[jnrA+0];
 463             vdwjidx0B        = 2*vdwtype[jnrB+0];
 464             vdwjidx0C        = 2*vdwtype[jnrC+0];
 465             vdwjidx0D        = 2*vdwtype[jnrD+0];
 466
 467             fjx0             = _mm_setzero_ps();
 468             fjy0             = _mm_setzero_ps();
 469             fjz0             = _mm_setzero_ps();
 470
 471             /**************************
 472              * CALCULATE INTERACTIONS *
 473              **************************/
 474
 475             /* Compute parameters for interactions between i and j atoms */
 476             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 477                                          vdwparam+vdwioffset0+vdwjidx0B,
 478                                          vdwparam+vdwioffset0+vdwjidx0C,
 479                                          vdwparam+vdwioffset0+vdwjidx0D,
 480                                          &c6_00,&c12_00);
 481
 482             /* LENNARD-JONES DISPERSION/REPULSION */
 483
 484             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 485             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 486             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 487             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 488             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 492             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 493
 494             fscal            = fvdw;
 495
 496             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 497
 498              /* Update vectorial force */
 499             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 500             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 501             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 502
 503             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 504             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 505             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             r10              = _mm_mul_ps(rsq10,rinv10);
 512             r10              = _mm_andnot_ps(dummy_mask,r10);
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq10             = _mm_mul_ps(iq1,jq0);
 516
 517             /* EWALD ELECTROSTATICS */
 518
 519             /* Analytical PME correction */
 520             zeta2            = _mm_mul_ps(beta2,rsq10);
 521             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 522             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 523             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 524             felec            = _mm_mul_ps(qq10,felec);
 525             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 526             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 527             velec            = _mm_mul_ps(qq10,velec);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm_andnot_ps(dummy_mask,velec);
 531             velecsum         = _mm_add_ps(velecsum,velec);
 532
 533             fscal            = felec;
 534
 535             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 536
 537              /* Update vectorial force */
 538             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 539             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 540             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 541
 542             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 543             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 544             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             r20              = _mm_mul_ps(rsq20,rinv20);
 551             r20              = _mm_andnot_ps(dummy_mask,r20);
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             qq20             = _mm_mul_ps(iq2,jq0);
 555
 556             /* EWALD ELECTROSTATICS */
 557
 558             /* Analytical PME correction */
 559             zeta2            = _mm_mul_ps(beta2,rsq20);
 560             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 561             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 562             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 563             felec            = _mm_mul_ps(qq20,felec);
 564             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 565             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 566             velec            = _mm_mul_ps(qq20,velec);
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm_andnot_ps(dummy_mask,velec);
 570             velecsum         = _mm_add_ps(velecsum,velec);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 575
 576              /* Update vectorial force */
 577             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 578             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 579             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 580
 581             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 582             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 583             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             r30              = _mm_mul_ps(rsq30,rinv30);
 590             r30              = _mm_andnot_ps(dummy_mask,r30);
 591
 592             /* Compute parameters for interactions between i and j atoms */
 593             qq30             = _mm_mul_ps(iq3,jq0);
 594
 595             /* EWALD ELECTROSTATICS */
 596
 597             /* Analytical PME correction */
 598             zeta2            = _mm_mul_ps(beta2,rsq30);
 599             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 600             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 601             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 602             felec            = _mm_mul_ps(qq30,felec);
 603             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 604             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 605             velec            = _mm_mul_ps(qq30,velec);
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm_andnot_ps(dummy_mask,velec);
 609             velecsum         = _mm_add_ps(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 614
 615              /* Update vectorial force */
 616             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 617             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 618             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 619
 620             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 621             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 622             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 623
 624             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 625             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 626             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 627             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 628
 629             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 630
 631             /* Inner loop uses 125 flops */
 632         }
 633
 634         /* End of innermost loop */
 635
 636         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 637                                               f+i_coord_offset,fshift+i_shift_offset);
 638
 639         ggid                        = gid[iidx];
 640         /* Update potential energies */
 641         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 642         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 643
 644         /* Increment number of inner iterations */
 645         inneriter                  += j_index_end - j_index_start;
 646
 647         /* Outer loop uses 26 flops */
 648     }
 649
 650     /* Increment number of outer iterations */
 651     outeriter        += nri;
 652
 653     /* Update outer/inner flops */
 654
 655     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*125);
 656 }
 657 /*
 658  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_128_fma_single
 659  * Electrostatics interaction: Ewald
 660  * VdW interaction:            LennardJones
 661  * Geometry:                   Water4-Particle
 662  * Calculate force/pot:        Force
 663  */
 664 void
 665 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_128_fma_single
 666                     (t_nblist                    * gmx_restrict       nlist,
 667                      rvec                        * gmx_restrict          xx,
 668                      rvec                        * gmx_restrict          ff,
 669                      t_forcerec                  * gmx_restrict          fr,
 670                      t_mdatoms                   * gmx_restrict     mdatoms,
 671                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 672                      t_nrnb                      * gmx_restrict        nrnb)
 673 {
 674     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 675      * just 0 for non-waters.
 676      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 677      * jnr indices corresponding to data put in the four positions in the SIMD register.
 678      */
 679     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 680     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 681     int              jnrA,jnrB,jnrC,jnrD;
 682     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 683     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 684     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 685     real             rcutoff_scalar;
 686     real             *shiftvec,*fshift,*x,*f;
 687     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 688     real             scratch[4*DIM];
 689     __m128           fscal,rcutoff,rcutoff2,jidxall;
 690     int              vdwioffset0;
 691     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 692     int              vdwioffset1;
 693     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 694     int              vdwioffset2;
 695     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 696     int              vdwioffset3;
 697     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 698     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 699     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 700     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 701     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 702     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 703     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 704     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 705     real             *charge;
 706     int              nvdwtype;
 707     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 708     int              *vdwtype;
 709     real             *vdwparam;
 710     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 711     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 712     __m128i          ewitab;
 713     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 714     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 715     real             *ewtab;
 716     __m128           dummy_mask,cutoff_mask;
 717     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 718     __m128           one     = _mm_set1_ps(1.0);
 719     __m128           two     = _mm_set1_ps(2.0);
 720     x                = xx[0];
 721     f                = ff[0];
 722
 723     nri              = nlist->nri;
 724     iinr             = nlist->iinr;
 725     jindex           = nlist->jindex;
 726     jjnr             = nlist->jjnr;
 727     shiftidx         = nlist->shift;
 728     gid              = nlist->gid;
 729     shiftvec         = fr->shift_vec[0];
 730     fshift           = fr->fshift[0];
 731     facel            = _mm_set1_ps(fr->epsfac);
 732     charge           = mdatoms->chargeA;
 733     nvdwtype         = fr->ntype;
 734     vdwparam         = fr->nbfp;
 735     vdwtype          = mdatoms->typeA;
 736
 737     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 738     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 739     beta2            = _mm_mul_ps(beta,beta);
 740     beta3            = _mm_mul_ps(beta,beta2);
 741     ewtab            = fr->ic->tabq_coul_F;
 742     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 743     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 744
 745     /* Setup water-specific parameters */
 746     inr              = nlist->iinr[0];
 747     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 748     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 749     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 750     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 751
 752     /* Avoid stupid compiler warnings */
 753     jnrA = jnrB = jnrC = jnrD = 0;
 754     j_coord_offsetA = 0;
 755     j_coord_offsetB = 0;
 756     j_coord_offsetC = 0;
 757     j_coord_offsetD = 0;
 758
 759     outeriter        = 0;
 760     inneriter        = 0;
 761
 762     for(iidx=0;iidx<4*DIM;iidx++)
 763     {
 764         scratch[iidx] = 0.0;
 765     }
 766
 767     /* Start outer loop over neighborlists */
 768     for(iidx=0; iidx<nri; iidx++)
 769     {
 770         /* Load shift vector for this list */
 771         i_shift_offset   = DIM*shiftidx[iidx];
 772
 773         /* Load limits for loop over neighbors */
 774         j_index_start    = jindex[iidx];
 775         j_index_end      = jindex[iidx+1];
 776
 777         /* Get outer coordinate index */
 778         inr              = iinr[iidx];
 779         i_coord_offset   = DIM*inr;
 780
 781         /* Load i particle coords and add shift vector */
 782         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 783                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 784
 785         fix0             = _mm_setzero_ps();
 786         fiy0             = _mm_setzero_ps();
 787         fiz0             = _mm_setzero_ps();
 788         fix1             = _mm_setzero_ps();
 789         fiy1             = _mm_setzero_ps();
 790         fiz1             = _mm_setzero_ps();
 791         fix2             = _mm_setzero_ps();
 792         fiy2             = _mm_setzero_ps();
 793         fiz2             = _mm_setzero_ps();
 794         fix3             = _mm_setzero_ps();
 795         fiy3             = _mm_setzero_ps();
 796         fiz3             = _mm_setzero_ps();
 797
 798         /* Start inner kernel loop */
 799         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 800         {
 801
 802             /* Get j neighbor index, and coordinate index */
 803             jnrA             = jjnr[jidx];
 804             jnrB             = jjnr[jidx+1];
 805             jnrC             = jjnr[jidx+2];
 806             jnrD             = jjnr[jidx+3];
 807             j_coord_offsetA  = DIM*jnrA;
 808             j_coord_offsetB  = DIM*jnrB;
 809             j_coord_offsetC  = DIM*jnrC;
 810             j_coord_offsetD  = DIM*jnrD;
 811
 812             /* load j atom coordinates */
 813             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 814                                               x+j_coord_offsetC,x+j_coord_offsetD,
 815                                               &jx0,&jy0,&jz0);
 816
 817             /* Calculate displacement vector */
 818             dx00             = _mm_sub_ps(ix0,jx0);
 819             dy00             = _mm_sub_ps(iy0,jy0);
 820             dz00             = _mm_sub_ps(iz0,jz0);
 821             dx10             = _mm_sub_ps(ix1,jx0);
 822             dy10             = _mm_sub_ps(iy1,jy0);
 823             dz10             = _mm_sub_ps(iz1,jz0);
 824             dx20             = _mm_sub_ps(ix2,jx0);
 825             dy20             = _mm_sub_ps(iy2,jy0);
 826             dz20             = _mm_sub_ps(iz2,jz0);
 827             dx30             = _mm_sub_ps(ix3,jx0);
 828             dy30             = _mm_sub_ps(iy3,jy0);
 829             dz30             = _mm_sub_ps(iz3,jz0);
 830
 831             /* Calculate squared distance and things based on it */
 832             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 833             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 834             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 835             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 836
 837             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 838             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 839             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 840
 841             rinvsq00         = gmx_mm_inv_ps(rsq00);
 842             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 843             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 844             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 845
 846             /* Load parameters for j particles */
 847             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 848                                                               charge+jnrC+0,charge+jnrD+0);
 849             vdwjidx0A        = 2*vdwtype[jnrA+0];
 850             vdwjidx0B        = 2*vdwtype[jnrB+0];
 851             vdwjidx0C        = 2*vdwtype[jnrC+0];
 852             vdwjidx0D        = 2*vdwtype[jnrD+0];
 853
 854             fjx0             = _mm_setzero_ps();
 855             fjy0             = _mm_setzero_ps();
 856             fjz0             = _mm_setzero_ps();
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 864                                          vdwparam+vdwioffset0+vdwjidx0B,
 865                                          vdwparam+vdwioffset0+vdwjidx0C,
 866                                          vdwparam+vdwioffset0+vdwjidx0D,
 867                                          &c6_00,&c12_00);
 868
 869             /* LENNARD-JONES DISPERSION/REPULSION */
 870
 871             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 872             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 873
 874             fscal            = fvdw;
 875
 876              /* Update vectorial force */
 877             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 878             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 879             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 880
 881             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 882             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 883             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             r10              = _mm_mul_ps(rsq10,rinv10);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq10             = _mm_mul_ps(iq1,jq0);
 893
 894             /* EWALD ELECTROSTATICS */
 895
 896             /* Analytical PME correction */
 897             zeta2            = _mm_mul_ps(beta2,rsq10);
 898             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 899             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 900             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 901             felec            = _mm_mul_ps(qq10,felec);
 902
 903             fscal            = felec;
 904
 905              /* Update vectorial force */
 906             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 907             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 908             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 909
 910             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 911             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 912             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             r20              = _mm_mul_ps(rsq20,rinv20);
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             qq20             = _mm_mul_ps(iq2,jq0);
 922
 923             /* EWALD ELECTROSTATICS */
 924
 925             /* Analytical PME correction */
 926             zeta2            = _mm_mul_ps(beta2,rsq20);
 927             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 928             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 929             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 930             felec            = _mm_mul_ps(qq20,felec);
 931
 932             fscal            = felec;
 933
 934              /* Update vectorial force */
 935             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 936             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 937             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 938
 939             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 940             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 941             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             r30              = _mm_mul_ps(rsq30,rinv30);
 948
 949             /* Compute parameters for interactions between i and j atoms */
 950             qq30             = _mm_mul_ps(iq3,jq0);
 951
 952             /* EWALD ELECTROSTATICS */
 953
 954             /* Analytical PME correction */
 955             zeta2            = _mm_mul_ps(beta2,rsq30);
 956             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 957             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 958             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 959             felec            = _mm_mul_ps(qq30,felec);
 960
 961             fscal            = felec;
 962
 963              /* Update vectorial force */
 964             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 965             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 966             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 967
 968             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 969             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 970             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 971
 972             fjptrA             = f+j_coord_offsetA;
 973             fjptrB             = f+j_coord_offsetB;
 974             fjptrC             = f+j_coord_offsetC;
 975             fjptrD             = f+j_coord_offsetD;
 976
 977             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 978
 979             /* Inner loop uses 114 flops */
 980         }
 981
 982         if(jidx<j_index_end)
 983         {
 984
 985             /* Get j neighbor index, and coordinate index */
 986             jnrlistA         = jjnr[jidx];
 987             jnrlistB         = jjnr[jidx+1];
 988             jnrlistC         = jjnr[jidx+2];
 989             jnrlistD         = jjnr[jidx+3];
 990             /* Sign of each element will be negative for non-real atoms.
 991              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 992              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 993              */
 994             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 995             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 996             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 997             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 998             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 999             j_coord_offsetA  = DIM*jnrA;
1000             j_coord_offsetB  = DIM*jnrB;
1001             j_coord_offsetC  = DIM*jnrC;
1002             j_coord_offsetD  = DIM*jnrD;
1003
1004             /* load j atom coordinates */
1005             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1006                                               x+j_coord_offsetC,x+j_coord_offsetD,
1007                                               &jx0,&jy0,&jz0);
1008
1009             /* Calculate displacement vector */
1010             dx00             = _mm_sub_ps(ix0,jx0);
1011             dy00             = _mm_sub_ps(iy0,jy0);
1012             dz00             = _mm_sub_ps(iz0,jz0);
1013             dx10             = _mm_sub_ps(ix1,jx0);
1014             dy10             = _mm_sub_ps(iy1,jy0);
1015             dz10             = _mm_sub_ps(iz1,jz0);
1016             dx20             = _mm_sub_ps(ix2,jx0);
1017             dy20             = _mm_sub_ps(iy2,jy0);
1018             dz20             = _mm_sub_ps(iz2,jz0);
1019             dx30             = _mm_sub_ps(ix3,jx0);
1020             dy30             = _mm_sub_ps(iy3,jy0);
1021             dz30             = _mm_sub_ps(iz3,jz0);
1022
1023             /* Calculate squared distance and things based on it */
1024             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1025             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1026             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1027             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1028
1029             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1030             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1031             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1032
1033             rinvsq00         = gmx_mm_inv_ps(rsq00);
1034             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1035             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1036             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1037
1038             /* Load parameters for j particles */
1039             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1040                                                               charge+jnrC+0,charge+jnrD+0);
1041             vdwjidx0A        = 2*vdwtype[jnrA+0];
1042             vdwjidx0B        = 2*vdwtype[jnrB+0];
1043             vdwjidx0C        = 2*vdwtype[jnrC+0];
1044             vdwjidx0D        = 2*vdwtype[jnrD+0];
1045
1046             fjx0             = _mm_setzero_ps();
1047             fjy0             = _mm_setzero_ps();
1048             fjz0             = _mm_setzero_ps();
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1056                                          vdwparam+vdwioffset0+vdwjidx0B,
1057                                          vdwparam+vdwioffset0+vdwjidx0C,
1058                                          vdwparam+vdwioffset0+vdwjidx0D,
1059                                          &c6_00,&c12_00);
1060
1061             /* LENNARD-JONES DISPERSION/REPULSION */
1062
1063             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1064             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1065
1066             fscal            = fvdw;
1067
1068             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1069
1070              /* Update vectorial force */
1071             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1072             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1073             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1074
1075             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1076             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1077             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             r10              = _mm_mul_ps(rsq10,rinv10);
1084             r10              = _mm_andnot_ps(dummy_mask,r10);
1085
1086             /* Compute parameters for interactions between i and j atoms */
1087             qq10             = _mm_mul_ps(iq1,jq0);
1088
1089             /* EWALD ELECTROSTATICS */
1090
1091             /* Analytical PME correction */
1092             zeta2            = _mm_mul_ps(beta2,rsq10);
1093             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1094             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1095             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1096             felec            = _mm_mul_ps(qq10,felec);
1097
1098             fscal            = felec;
1099
1100             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1101
1102              /* Update vectorial force */
1103             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1104             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1105             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1106
1107             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1108             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1109             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             r20              = _mm_mul_ps(rsq20,rinv20);
1116             r20              = _mm_andnot_ps(dummy_mask,r20);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq20             = _mm_mul_ps(iq2,jq0);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Analytical PME correction */
1124             zeta2            = _mm_mul_ps(beta2,rsq20);
1125             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1126             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1127             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1128             felec            = _mm_mul_ps(qq20,felec);
1129
1130             fscal            = felec;
1131
1132             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1133
1134              /* Update vectorial force */
1135             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1136             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1137             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1138
1139             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1140             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1141             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1142
1143             /**************************
1144              * CALCULATE INTERACTIONS *
1145              **************************/
1146
1147             r30              = _mm_mul_ps(rsq30,rinv30);
1148             r30              = _mm_andnot_ps(dummy_mask,r30);
1149
1150             /* Compute parameters for interactions between i and j atoms */
1151             qq30             = _mm_mul_ps(iq3,jq0);
1152
1153             /* EWALD ELECTROSTATICS */
1154
1155             /* Analytical PME correction */
1156             zeta2            = _mm_mul_ps(beta2,rsq30);
1157             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1158             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1159             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1160             felec            = _mm_mul_ps(qq30,felec);
1161
1162             fscal            = felec;
1163
1164             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1165
1166              /* Update vectorial force */
1167             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1168             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1169             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1170
1171             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1172             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1173             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1174
1175             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1176             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1177             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1178             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1179
1180             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1181
1182             /* Inner loop uses 117 flops */
1183         }
1184
1185         /* End of innermost loop */
1186
1187         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1188                                               f+i_coord_offset,fshift+i_shift_offset);
1189
1190         /* Increment number of inner iterations */
1191         inneriter                  += j_index_end - j_index_start;
1192
1193         /* Outer loop uses 24 flops */
1194     }
1195
1196     /* Increment number of outer iterations */
1197     outeriter        += nri;
1198
1199     /* Update outer/inner flops */
1200
1201     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*117);
1202 }