src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_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_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 101     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 102     __m128i          ewitab;
 103     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 105     real             *ewtab;
 106     __m128           dummy_mask,cutoff_mask;
 107     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 108     __m128           one     = _mm_set1_ps(1.0);
 109     __m128           two     = _mm_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 128     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 129     beta2            = _mm_mul_ps(beta,beta);
 130     beta3            = _mm_mul_ps(beta,beta2);
 131     ewtab            = fr->ic->tabq_coul_FDV0;
 132     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 133     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 138     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 139     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->rcoulomb;
 144     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 145     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 146
 147     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 148     rvdw             = _mm_set1_ps(fr->rvdw);
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = jnrC = jnrD = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154     j_coord_offsetC = 0;
 155     j_coord_offsetD = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 182
 183         fix0             = _mm_setzero_ps();
 184         fiy0             = _mm_setzero_ps();
 185         fiz0             = _mm_setzero_ps();
 186         fix1             = _mm_setzero_ps();
 187         fiy1             = _mm_setzero_ps();
 188         fiz1             = _mm_setzero_ps();
 189         fix2             = _mm_setzero_ps();
 190         fiy2             = _mm_setzero_ps();
 191         fiz2             = _mm_setzero_ps();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm_setzero_ps();
 195         vvdwsum          = _mm_setzero_ps();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             j_coord_offsetA  = DIM*jnrA;
 207             j_coord_offsetB  = DIM*jnrB;
 208             j_coord_offsetC  = DIM*jnrC;
 209             j_coord_offsetD  = DIM*jnrD;
 210
 211             /* load j atom coordinates */
 212             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                               x+j_coord_offsetC,x+j_coord_offsetD,
 214                                               &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm_sub_ps(ix0,jx0);
 218             dy00             = _mm_sub_ps(iy0,jy0);
 219             dz00             = _mm_sub_ps(iz0,jz0);
 220             dx10             = _mm_sub_ps(ix1,jx0);
 221             dy10             = _mm_sub_ps(iy1,jy0);
 222             dz10             = _mm_sub_ps(iz1,jz0);
 223             dx20             = _mm_sub_ps(ix2,jx0);
 224             dy20             = _mm_sub_ps(iy2,jy0);
 225             dz20             = _mm_sub_ps(iz2,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 229             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 230             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 231
 232             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 233             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 234             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 235
 236             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 237             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 238             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 239
 240             /* Load parameters for j particles */
 241             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 242                                                               charge+jnrC+0,charge+jnrD+0);
 243             vdwjidx0A        = 2*vdwtype[jnrA+0];
 244             vdwjidx0B        = 2*vdwtype[jnrB+0];
 245             vdwjidx0C        = 2*vdwtype[jnrC+0];
 246             vdwjidx0D        = 2*vdwtype[jnrD+0];
 247
 248             fjx0             = _mm_setzero_ps();
 249             fjy0             = _mm_setzero_ps();
 250             fjz0             = _mm_setzero_ps();
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             if (gmx_mm_any_lt(rsq00,rcutoff2))
 257             {
 258
 259             r00              = _mm_mul_ps(rsq00,rinv00);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             qq00             = _mm_mul_ps(iq0,jq0);
 263             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 264                                          vdwparam+vdwioffset0+vdwjidx0B,
 265                                          vdwparam+vdwioffset0+vdwjidx0C,
 266                                          vdwparam+vdwioffset0+vdwjidx0D,
 267                                          &c6_00,&c12_00);
 268
 269             /* EWALD ELECTROSTATICS */
 270
 271             /* Analytical PME correction */
 272             zeta2            = _mm_mul_ps(beta2,rsq00);
 273             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 274             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 275             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 276             felec            = _mm_mul_ps(qq00,felec);
 277             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 278             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 279             velec            = _mm_mul_ps(qq00,velec);
 280
 281             /* LENNARD-JONES DISPERSION/REPULSION */
 282
 283             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 284             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 285             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 286             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 287                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 288             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 289
 290             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velec            = _mm_and_ps(velec,cutoff_mask);
 294             velecsum         = _mm_add_ps(velecsum,velec);
 295             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 296             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 297
 298             fscal            = _mm_add_ps(felec,fvdw);
 299
 300             fscal            = _mm_and_ps(fscal,cutoff_mask);
 301
 302              /* Update vectorial force */
 303             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 304             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 305             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 306
 307             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 308             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 309             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 310
 311             }
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             if (gmx_mm_any_lt(rsq10,rcutoff2))
 318             {
 319
 320             r10              = _mm_mul_ps(rsq10,rinv10);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_ps(iq1,jq0);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Analytical PME correction */
 328             zeta2            = _mm_mul_ps(beta2,rsq10);
 329             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 330             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 331             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 332             felec            = _mm_mul_ps(qq10,felec);
 333             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 334             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 335             velec            = _mm_mul_ps(qq10,velec);
 336
 337             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velec            = _mm_and_ps(velec,cutoff_mask);
 341             velecsum         = _mm_add_ps(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             fscal            = _mm_and_ps(fscal,cutoff_mask);
 346
 347              /* Update vectorial force */
 348             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 349             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 350             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 351
 352             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 353             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 354             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 355
 356             }
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             if (gmx_mm_any_lt(rsq20,rcutoff2))
 363             {
 364
 365             r20              = _mm_mul_ps(rsq20,rinv20);
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq20             = _mm_mul_ps(iq2,jq0);
 369
 370             /* EWALD ELECTROSTATICS */
 371
 372             /* Analytical PME correction */
 373             zeta2            = _mm_mul_ps(beta2,rsq20);
 374             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 375             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 376             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 377             felec            = _mm_mul_ps(qq20,felec);
 378             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 379             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 380             velec            = _mm_mul_ps(qq20,velec);
 381
 382             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velec            = _mm_and_ps(velec,cutoff_mask);
 386             velecsum         = _mm_add_ps(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             fscal            = _mm_and_ps(fscal,cutoff_mask);
 391
 392              /* Update vectorial force */
 393             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 394             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 395             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 396
 397             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 398             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 399             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 400
 401             }
 402
 403             fjptrA             = f+j_coord_offsetA;
 404             fjptrB             = f+j_coord_offsetB;
 405             fjptrC             = f+j_coord_offsetC;
 406             fjptrD             = f+j_coord_offsetD;
 407
 408             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 409
 410             /* Inner loop uses 117 flops */
 411         }
 412
 413         if(jidx<j_index_end)
 414         {
 415
 416             /* Get j neighbor index, and coordinate index */
 417             jnrlistA         = jjnr[jidx];
 418             jnrlistB         = jjnr[jidx+1];
 419             jnrlistC         = jjnr[jidx+2];
 420             jnrlistD         = jjnr[jidx+3];
 421             /* Sign of each element will be negative for non-real atoms.
 422              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 423              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 424              */
 425             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 426             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 427             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 428             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 429             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 430             j_coord_offsetA  = DIM*jnrA;
 431             j_coord_offsetB  = DIM*jnrB;
 432             j_coord_offsetC  = DIM*jnrC;
 433             j_coord_offsetD  = DIM*jnrD;
 434
 435             /* load j atom coordinates */
 436             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 437                                               x+j_coord_offsetC,x+j_coord_offsetD,
 438                                               &jx0,&jy0,&jz0);
 439
 440             /* Calculate displacement vector */
 441             dx00             = _mm_sub_ps(ix0,jx0);
 442             dy00             = _mm_sub_ps(iy0,jy0);
 443             dz00             = _mm_sub_ps(iz0,jz0);
 444             dx10             = _mm_sub_ps(ix1,jx0);
 445             dy10             = _mm_sub_ps(iy1,jy0);
 446             dz10             = _mm_sub_ps(iz1,jz0);
 447             dx20             = _mm_sub_ps(ix2,jx0);
 448             dy20             = _mm_sub_ps(iy2,jy0);
 449             dz20             = _mm_sub_ps(iz2,jz0);
 450
 451             /* Calculate squared distance and things based on it */
 452             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 453             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 454             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 455
 456             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 457             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 458             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 459
 460             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 461             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 462             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 463
 464             /* Load parameters for j particles */
 465             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 466                                                               charge+jnrC+0,charge+jnrD+0);
 467             vdwjidx0A        = 2*vdwtype[jnrA+0];
 468             vdwjidx0B        = 2*vdwtype[jnrB+0];
 469             vdwjidx0C        = 2*vdwtype[jnrC+0];
 470             vdwjidx0D        = 2*vdwtype[jnrD+0];
 471
 472             fjx0             = _mm_setzero_ps();
 473             fjy0             = _mm_setzero_ps();
 474             fjz0             = _mm_setzero_ps();
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             if (gmx_mm_any_lt(rsq00,rcutoff2))
 481             {
 482
 483             r00              = _mm_mul_ps(rsq00,rinv00);
 484             r00              = _mm_andnot_ps(dummy_mask,r00);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq00             = _mm_mul_ps(iq0,jq0);
 488             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 489                                          vdwparam+vdwioffset0+vdwjidx0B,
 490                                          vdwparam+vdwioffset0+vdwjidx0C,
 491                                          vdwparam+vdwioffset0+vdwjidx0D,
 492                                          &c6_00,&c12_00);
 493
 494             /* EWALD ELECTROSTATICS */
 495
 496             /* Analytical PME correction */
 497             zeta2            = _mm_mul_ps(beta2,rsq00);
 498             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 499             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 500             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 501             felec            = _mm_mul_ps(qq00,felec);
 502             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 503             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 504             velec            = _mm_mul_ps(qq00,velec);
 505
 506             /* LENNARD-JONES DISPERSION/REPULSION */
 507
 508             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 509             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 510             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 511             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 512                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 513             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 514
 515             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 516
 517             /* Update potential sum for this i atom from the interaction with this j atom. */
 518             velec            = _mm_and_ps(velec,cutoff_mask);
 519             velec            = _mm_andnot_ps(dummy_mask,velec);
 520             velecsum         = _mm_add_ps(velecsum,velec);
 521             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 522             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 523             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 524
 525             fscal            = _mm_add_ps(felec,fvdw);
 526
 527             fscal            = _mm_and_ps(fscal,cutoff_mask);
 528
 529             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 530
 531              /* Update vectorial force */
 532             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 533             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 534             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 535
 536             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 537             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 538             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 539
 540             }
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             if (gmx_mm_any_lt(rsq10,rcutoff2))
 547             {
 548
 549             r10              = _mm_mul_ps(rsq10,rinv10);
 550             r10              = _mm_andnot_ps(dummy_mask,r10);
 551
 552             /* Compute parameters for interactions between i and j atoms */
 553             qq10             = _mm_mul_ps(iq1,jq0);
 554
 555             /* EWALD ELECTROSTATICS */
 556
 557             /* Analytical PME correction */
 558             zeta2            = _mm_mul_ps(beta2,rsq10);
 559             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 560             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 561             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 562             felec            = _mm_mul_ps(qq10,felec);
 563             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 564             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 565             velec            = _mm_mul_ps(qq10,velec);
 566
 567             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             velec            = _mm_and_ps(velec,cutoff_mask);
 571             velec            = _mm_andnot_ps(dummy_mask,velec);
 572             velecsum         = _mm_add_ps(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_and_ps(fscal,cutoff_mask);
 577
 578             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 579
 580              /* Update vectorial force */
 581             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 582             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 583             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 584
 585             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 586             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 587             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 588
 589             }
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             if (gmx_mm_any_lt(rsq20,rcutoff2))
 596             {
 597
 598             r20              = _mm_mul_ps(rsq20,rinv20);
 599             r20              = _mm_andnot_ps(dummy_mask,r20);
 600
 601             /* Compute parameters for interactions between i and j atoms */
 602             qq20             = _mm_mul_ps(iq2,jq0);
 603
 604             /* EWALD ELECTROSTATICS */
 605
 606             /* Analytical PME correction */
 607             zeta2            = _mm_mul_ps(beta2,rsq20);
 608             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 609             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 610             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 611             felec            = _mm_mul_ps(qq20,felec);
 612             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 613             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 614             velec            = _mm_mul_ps(qq20,velec);
 615
 616             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velec            = _mm_and_ps(velec,cutoff_mask);
 620             velec            = _mm_andnot_ps(dummy_mask,velec);
 621             velecsum         = _mm_add_ps(velecsum,velec);
 622
 623             fscal            = felec;
 624
 625             fscal            = _mm_and_ps(fscal,cutoff_mask);
 626
 627             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 628
 629              /* Update vectorial force */
 630             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 631             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 632             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 633
 634             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 635             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 636             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 637
 638             }
 639
 640             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 641             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 642             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 643             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 644
 645             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 646
 647             /* Inner loop uses 120 flops */
 648         }
 649
 650         /* End of innermost loop */
 651
 652         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 653                                               f+i_coord_offset,fshift+i_shift_offset);
 654
 655         ggid                        = gid[iidx];
 656         /* Update potential energies */
 657         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 658         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 659
 660         /* Increment number of inner iterations */
 661         inneriter                  += j_index_end - j_index_start;
 662
 663         /* Outer loop uses 20 flops */
 664     }
 665
 666     /* Increment number of outer iterations */
 667     outeriter        += nri;
 668
 669     /* Update outer/inner flops */
 670
 671     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 672 }
 673 /*
 674  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_single
 675  * Electrostatics interaction: Ewald
 676  * VdW interaction:            LennardJones
 677  * Geometry:                   Water3-Particle
 678  * Calculate force/pot:        Force
 679  */
 680 void
 681 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_128_fma_single
 682                     (t_nblist                    * gmx_restrict       nlist,
 683                      rvec                        * gmx_restrict          xx,
 684                      rvec                        * gmx_restrict          ff,
 685                      t_forcerec                  * gmx_restrict          fr,
 686                      t_mdatoms                   * gmx_restrict     mdatoms,
 687                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 688                      t_nrnb                      * gmx_restrict        nrnb)
 689 {
 690     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 691      * just 0 for non-waters.
 692      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 693      * jnr indices corresponding to data put in the four positions in the SIMD register.
 694      */
 695     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 696     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 697     int              jnrA,jnrB,jnrC,jnrD;
 698     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 699     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 700     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 701     real             rcutoff_scalar;
 702     real             *shiftvec,*fshift,*x,*f;
 703     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 704     real             scratch[4*DIM];
 705     __m128           fscal,rcutoff,rcutoff2,jidxall;
 706     int              vdwioffset0;
 707     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 708     int              vdwioffset1;
 709     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 710     int              vdwioffset2;
 711     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 712     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 713     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 714     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 715     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 716     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 717     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 718     real             *charge;
 719     int              nvdwtype;
 720     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 721     int              *vdwtype;
 722     real             *vdwparam;
 723     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 724     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 725     __m128i          ewitab;
 726     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 727     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 728     real             *ewtab;
 729     __m128           dummy_mask,cutoff_mask;
 730     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 731     __m128           one     = _mm_set1_ps(1.0);
 732     __m128           two     = _mm_set1_ps(2.0);
 733     x                = xx[0];
 734     f                = ff[0];
 735
 736     nri              = nlist->nri;
 737     iinr             = nlist->iinr;
 738     jindex           = nlist->jindex;
 739     jjnr             = nlist->jjnr;
 740     shiftidx         = nlist->shift;
 741     gid              = nlist->gid;
 742     shiftvec         = fr->shift_vec[0];
 743     fshift           = fr->fshift[0];
 744     facel            = _mm_set1_ps(fr->epsfac);
 745     charge           = mdatoms->chargeA;
 746     nvdwtype         = fr->ntype;
 747     vdwparam         = fr->nbfp;
 748     vdwtype          = mdatoms->typeA;
 749
 750     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 751     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 752     beta2            = _mm_mul_ps(beta,beta);
 753     beta3            = _mm_mul_ps(beta,beta2);
 754     ewtab            = fr->ic->tabq_coul_F;
 755     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 756     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 757
 758     /* Setup water-specific parameters */
 759     inr              = nlist->iinr[0];
 760     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 761     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 762     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 763     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 764
 765     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 766     rcutoff_scalar   = fr->rcoulomb;
 767     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 768     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 769
 770     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 771     rvdw             = _mm_set1_ps(fr->rvdw);
 772
 773     /* Avoid stupid compiler warnings */
 774     jnrA = jnrB = jnrC = jnrD = 0;
 775     j_coord_offsetA = 0;
 776     j_coord_offsetB = 0;
 777     j_coord_offsetC = 0;
 778     j_coord_offsetD = 0;
 779
 780     outeriter        = 0;
 781     inneriter        = 0;
 782
 783     for(iidx=0;iidx<4*DIM;iidx++)
 784     {
 785         scratch[iidx] = 0.0;
 786     }
 787
 788     /* Start outer loop over neighborlists */
 789     for(iidx=0; iidx<nri; iidx++)
 790     {
 791         /* Load shift vector for this list */
 792         i_shift_offset   = DIM*shiftidx[iidx];
 793
 794         /* Load limits for loop over neighbors */
 795         j_index_start    = jindex[iidx];
 796         j_index_end      = jindex[iidx+1];
 797
 798         /* Get outer coordinate index */
 799         inr              = iinr[iidx];
 800         i_coord_offset   = DIM*inr;
 801
 802         /* Load i particle coords and add shift vector */
 803         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 804                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 805
 806         fix0             = _mm_setzero_ps();
 807         fiy0             = _mm_setzero_ps();
 808         fiz0             = _mm_setzero_ps();
 809         fix1             = _mm_setzero_ps();
 810         fiy1             = _mm_setzero_ps();
 811         fiz1             = _mm_setzero_ps();
 812         fix2             = _mm_setzero_ps();
 813         fiy2             = _mm_setzero_ps();
 814         fiz2             = _mm_setzero_ps();
 815
 816         /* Start inner kernel loop */
 817         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 818         {
 819
 820             /* Get j neighbor index, and coordinate index */
 821             jnrA             = jjnr[jidx];
 822             jnrB             = jjnr[jidx+1];
 823             jnrC             = jjnr[jidx+2];
 824             jnrD             = jjnr[jidx+3];
 825             j_coord_offsetA  = DIM*jnrA;
 826             j_coord_offsetB  = DIM*jnrB;
 827             j_coord_offsetC  = DIM*jnrC;
 828             j_coord_offsetD  = DIM*jnrD;
 829
 830             /* load j atom coordinates */
 831             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 832                                               x+j_coord_offsetC,x+j_coord_offsetD,
 833                                               &jx0,&jy0,&jz0);
 834
 835             /* Calculate displacement vector */
 836             dx00             = _mm_sub_ps(ix0,jx0);
 837             dy00             = _mm_sub_ps(iy0,jy0);
 838             dz00             = _mm_sub_ps(iz0,jz0);
 839             dx10             = _mm_sub_ps(ix1,jx0);
 840             dy10             = _mm_sub_ps(iy1,jy0);
 841             dz10             = _mm_sub_ps(iz1,jz0);
 842             dx20             = _mm_sub_ps(ix2,jx0);
 843             dy20             = _mm_sub_ps(iy2,jy0);
 844             dz20             = _mm_sub_ps(iz2,jz0);
 845
 846             /* Calculate squared distance and things based on it */
 847             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 848             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 849             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 850
 851             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 852             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 853             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 854
 855             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 856             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 857             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 858
 859             /* Load parameters for j particles */
 860             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 861                                                               charge+jnrC+0,charge+jnrD+0);
 862             vdwjidx0A        = 2*vdwtype[jnrA+0];
 863             vdwjidx0B        = 2*vdwtype[jnrB+0];
 864             vdwjidx0C        = 2*vdwtype[jnrC+0];
 865             vdwjidx0D        = 2*vdwtype[jnrD+0];
 866
 867             fjx0             = _mm_setzero_ps();
 868             fjy0             = _mm_setzero_ps();
 869             fjz0             = _mm_setzero_ps();
 870
 871             /**************************
 872              * CALCULATE INTERACTIONS *
 873              **************************/
 874
 875             if (gmx_mm_any_lt(rsq00,rcutoff2))
 876             {
 877
 878             r00              = _mm_mul_ps(rsq00,rinv00);
 879
 880             /* Compute parameters for interactions between i and j atoms */
 881             qq00             = _mm_mul_ps(iq0,jq0);
 882             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 883                                          vdwparam+vdwioffset0+vdwjidx0B,
 884                                          vdwparam+vdwioffset0+vdwjidx0C,
 885                                          vdwparam+vdwioffset0+vdwjidx0D,
 886                                          &c6_00,&c12_00);
 887
 888             /* EWALD ELECTROSTATICS */
 889
 890             /* Analytical PME correction */
 891             zeta2            = _mm_mul_ps(beta2,rsq00);
 892             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 893             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 894             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 895             felec            = _mm_mul_ps(qq00,felec);
 896
 897             /* LENNARD-JONES DISPERSION/REPULSION */
 898
 899             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 900             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 901
 902             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 903
 904             fscal            = _mm_add_ps(felec,fvdw);
 905
 906             fscal            = _mm_and_ps(fscal,cutoff_mask);
 907
 908              /* Update vectorial force */
 909             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 910             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 911             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 912
 913             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 914             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 915             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 916
 917             }
 918
 919             /**************************
 920              * CALCULATE INTERACTIONS *
 921              **************************/
 922
 923             if (gmx_mm_any_lt(rsq10,rcutoff2))
 924             {
 925
 926             r10              = _mm_mul_ps(rsq10,rinv10);
 927
 928             /* Compute parameters for interactions between i and j atoms */
 929             qq10             = _mm_mul_ps(iq1,jq0);
 930
 931             /* EWALD ELECTROSTATICS */
 932
 933             /* Analytical PME correction */
 934             zeta2            = _mm_mul_ps(beta2,rsq10);
 935             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 936             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 937             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 938             felec            = _mm_mul_ps(qq10,felec);
 939
 940             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 941
 942             fscal            = felec;
 943
 944             fscal            = _mm_and_ps(fscal,cutoff_mask);
 945
 946              /* Update vectorial force */
 947             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 948             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 949             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 950
 951             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 952             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 953             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 954
 955             }
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             if (gmx_mm_any_lt(rsq20,rcutoff2))
 962             {
 963
 964             r20              = _mm_mul_ps(rsq20,rinv20);
 965
 966             /* Compute parameters for interactions between i and j atoms */
 967             qq20             = _mm_mul_ps(iq2,jq0);
 968
 969             /* EWALD ELECTROSTATICS */
 970
 971             /* Analytical PME correction */
 972             zeta2            = _mm_mul_ps(beta2,rsq20);
 973             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 974             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 975             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 976             felec            = _mm_mul_ps(qq20,felec);
 977
 978             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 979
 980             fscal            = felec;
 981
 982             fscal            = _mm_and_ps(fscal,cutoff_mask);
 983
 984              /* Update vectorial force */
 985             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 986             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 987             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 988
 989             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 990             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 991             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 992
 993             }
 994
 995             fjptrA             = f+j_coord_offsetA;
 996             fjptrB             = f+j_coord_offsetB;
 997             fjptrC             = f+j_coord_offsetC;
 998             fjptrD             = f+j_coord_offsetD;
 999
1000             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1001
1002             /* Inner loop uses 100 flops */
1003         }
1004
1005         if(jidx<j_index_end)
1006         {
1007
1008             /* Get j neighbor index, and coordinate index */
1009             jnrlistA         = jjnr[jidx];
1010             jnrlistB         = jjnr[jidx+1];
1011             jnrlistC         = jjnr[jidx+2];
1012             jnrlistD         = jjnr[jidx+3];
1013             /* Sign of each element will be negative for non-real atoms.
1014              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1015              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1016              */
1017             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1018             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1019             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1020             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1021             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1022             j_coord_offsetA  = DIM*jnrA;
1023             j_coord_offsetB  = DIM*jnrB;
1024             j_coord_offsetC  = DIM*jnrC;
1025             j_coord_offsetD  = DIM*jnrD;
1026
1027             /* load j atom coordinates */
1028             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1029                                               x+j_coord_offsetC,x+j_coord_offsetD,
1030                                               &jx0,&jy0,&jz0);
1031
1032             /* Calculate displacement vector */
1033             dx00             = _mm_sub_ps(ix0,jx0);
1034             dy00             = _mm_sub_ps(iy0,jy0);
1035             dz00             = _mm_sub_ps(iz0,jz0);
1036             dx10             = _mm_sub_ps(ix1,jx0);
1037             dy10             = _mm_sub_ps(iy1,jy0);
1038             dz10             = _mm_sub_ps(iz1,jz0);
1039             dx20             = _mm_sub_ps(ix2,jx0);
1040             dy20             = _mm_sub_ps(iy2,jy0);
1041             dz20             = _mm_sub_ps(iz2,jz0);
1042
1043             /* Calculate squared distance and things based on it */
1044             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1045             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1046             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1047
1048             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1049             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1050             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1051
1052             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1053             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1054             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1055
1056             /* Load parameters for j particles */
1057             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1058                                                               charge+jnrC+0,charge+jnrD+0);
1059             vdwjidx0A        = 2*vdwtype[jnrA+0];
1060             vdwjidx0B        = 2*vdwtype[jnrB+0];
1061             vdwjidx0C        = 2*vdwtype[jnrC+0];
1062             vdwjidx0D        = 2*vdwtype[jnrD+0];
1063
1064             fjx0             = _mm_setzero_ps();
1065             fjy0             = _mm_setzero_ps();
1066             fjz0             = _mm_setzero_ps();
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             if (gmx_mm_any_lt(rsq00,rcutoff2))
1073             {
1074
1075             r00              = _mm_mul_ps(rsq00,rinv00);
1076             r00              = _mm_andnot_ps(dummy_mask,r00);
1077
1078             /* Compute parameters for interactions between i and j atoms */
1079             qq00             = _mm_mul_ps(iq0,jq0);
1080             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1081                                          vdwparam+vdwioffset0+vdwjidx0B,
1082                                          vdwparam+vdwioffset0+vdwjidx0C,
1083                                          vdwparam+vdwioffset0+vdwjidx0D,
1084                                          &c6_00,&c12_00);
1085
1086             /* EWALD ELECTROSTATICS */
1087
1088             /* Analytical PME correction */
1089             zeta2            = _mm_mul_ps(beta2,rsq00);
1090             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1091             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1092             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1093             felec            = _mm_mul_ps(qq00,felec);
1094
1095             /* LENNARD-JONES DISPERSION/REPULSION */
1096
1097             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1098             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1099
1100             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1101
1102             fscal            = _mm_add_ps(felec,fvdw);
1103
1104             fscal            = _mm_and_ps(fscal,cutoff_mask);
1105
1106             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1107
1108              /* Update vectorial force */
1109             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1110             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1111             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1112
1113             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1114             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1115             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1116
1117             }
1118
1119             /**************************
1120              * CALCULATE INTERACTIONS *
1121              **************************/
1122
1123             if (gmx_mm_any_lt(rsq10,rcutoff2))
1124             {
1125
1126             r10              = _mm_mul_ps(rsq10,rinv10);
1127             r10              = _mm_andnot_ps(dummy_mask,r10);
1128
1129             /* Compute parameters for interactions between i and j atoms */
1130             qq10             = _mm_mul_ps(iq1,jq0);
1131
1132             /* EWALD ELECTROSTATICS */
1133
1134             /* Analytical PME correction */
1135             zeta2            = _mm_mul_ps(beta2,rsq10);
1136             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1137             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1138             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1139             felec            = _mm_mul_ps(qq10,felec);
1140
1141             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1142
1143             fscal            = felec;
1144
1145             fscal            = _mm_and_ps(fscal,cutoff_mask);
1146
1147             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1148
1149              /* Update vectorial force */
1150             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1151             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1152             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1153
1154             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1155             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1156             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1157
1158             }
1159
1160             /**************************
1161              * CALCULATE INTERACTIONS *
1162              **************************/
1163
1164             if (gmx_mm_any_lt(rsq20,rcutoff2))
1165             {
1166
1167             r20              = _mm_mul_ps(rsq20,rinv20);
1168             r20              = _mm_andnot_ps(dummy_mask,r20);
1169
1170             /* Compute parameters for interactions between i and j atoms */
1171             qq20             = _mm_mul_ps(iq2,jq0);
1172
1173             /* EWALD ELECTROSTATICS */
1174
1175             /* Analytical PME correction */
1176             zeta2            = _mm_mul_ps(beta2,rsq20);
1177             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1178             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1179             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1180             felec            = _mm_mul_ps(qq20,felec);
1181
1182             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1183
1184             fscal            = felec;
1185
1186             fscal            = _mm_and_ps(fscal,cutoff_mask);
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190              /* Update vectorial force */
1191             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1192             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1193             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1194
1195             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1196             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1197             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1198
1199             }
1200
1201             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1202             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1203             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1204             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1205
1206             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1207
1208             /* Inner loop uses 103 flops */
1209         }
1210
1211         /* End of innermost loop */
1212
1213         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1214                                               f+i_coord_offset,fshift+i_shift_offset);
1215
1216         /* Increment number of inner iterations */
1217         inneriter                  += j_index_end - j_index_start;
1218
1219         /* Outer loop uses 18 flops */
1220     }
1221
1222     /* Increment number of outer iterations */
1223     outeriter        += nri;
1224
1225     /* Update outer/inner flops */
1226
1227     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*103);
1228 }