src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_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          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 108     real             *vftab;
 109     __m128i          ewitab;
 110     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 112     real             *ewtab;
 113     __m128           dummy_mask,cutoff_mask;
 114     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 115     __m128           one     = _mm_set1_ps(1.0);
 116     __m128           two     = _mm_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_vdw->data;
 135     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 136
 137     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 138     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 139     beta2            = _mm_mul_ps(beta,beta);
 140     beta3            = _mm_mul_ps(beta,beta2);
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 148     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 149     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 150     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     for(iidx=0;iidx<4*DIM;iidx++)
 163     {
 164         scratch[iidx] = 0.0;
 165     }
 166
 167     /* Start outer loop over neighborlists */
 168     for(iidx=0; iidx<nri; iidx++)
 169     {
 170         /* Load shift vector for this list */
 171         i_shift_offset   = DIM*shiftidx[iidx];
 172
 173         /* Load limits for loop over neighbors */
 174         j_index_start    = jindex[iidx];
 175         j_index_end      = jindex[iidx+1];
 176
 177         /* Get outer coordinate index */
 178         inr              = iinr[iidx];
 179         i_coord_offset   = DIM*inr;
 180
 181         /* Load i particle coords and add shift vector */
 182         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 183                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 184
 185         fix0             = _mm_setzero_ps();
 186         fiy0             = _mm_setzero_ps();
 187         fiz0             = _mm_setzero_ps();
 188         fix1             = _mm_setzero_ps();
 189         fiy1             = _mm_setzero_ps();
 190         fiz1             = _mm_setzero_ps();
 191         fix2             = _mm_setzero_ps();
 192         fiy2             = _mm_setzero_ps();
 193         fiz2             = _mm_setzero_ps();
 194         fix3             = _mm_setzero_ps();
 195         fiy3             = _mm_setzero_ps();
 196         fiz3             = _mm_setzero_ps();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm_setzero_ps();
 200         vvdwsum          = _mm_setzero_ps();
 201
 202         /* Start inner kernel loop */
 203         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 204         {
 205
 206             /* Get j neighbor index, and coordinate index */
 207             jnrA             = jjnr[jidx];
 208             jnrB             = jjnr[jidx+1];
 209             jnrC             = jjnr[jidx+2];
 210             jnrD             = jjnr[jidx+3];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215
 216             /* load j atom coordinates */
 217             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                               x+j_coord_offsetC,x+j_coord_offsetD,
 219                                               &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm_sub_ps(ix0,jx0);
 223             dy00             = _mm_sub_ps(iy0,jy0);
 224             dz00             = _mm_sub_ps(iz0,jz0);
 225             dx10             = _mm_sub_ps(ix1,jx0);
 226             dy10             = _mm_sub_ps(iy1,jy0);
 227             dz10             = _mm_sub_ps(iz1,jz0);
 228             dx20             = _mm_sub_ps(ix2,jx0);
 229             dy20             = _mm_sub_ps(iy2,jy0);
 230             dz20             = _mm_sub_ps(iz2,jz0);
 231             dx30             = _mm_sub_ps(ix3,jx0);
 232             dy30             = _mm_sub_ps(iy3,jy0);
 233             dz30             = _mm_sub_ps(iz3,jz0);
 234
 235             /* Calculate squared distance and things based on it */
 236             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 237             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 238             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 239             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 240
 241             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 242             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 243             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 244             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 245
 246             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 247             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 248             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 249
 250             /* Load parameters for j particles */
 251             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 252                                                               charge+jnrC+0,charge+jnrD+0);
 253             vdwjidx0A        = 2*vdwtype[jnrA+0];
 254             vdwjidx0B        = 2*vdwtype[jnrB+0];
 255             vdwjidx0C        = 2*vdwtype[jnrC+0];
 256             vdwjidx0D        = 2*vdwtype[jnrD+0];
 257
 258             fjx0             = _mm_setzero_ps();
 259             fjy0             = _mm_setzero_ps();
 260             fjz0             = _mm_setzero_ps();
 261
 262             /**************************
 263              * CALCULATE INTERACTIONS *
 264              **************************/
 265
 266             r00              = _mm_mul_ps(rsq00,rinv00);
 267
 268             /* Compute parameters for interactions between i and j atoms */
 269             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 270                                          vdwparam+vdwioffset0+vdwjidx0B,
 271                                          vdwparam+vdwioffset0+vdwjidx0C,
 272                                          vdwparam+vdwioffset0+vdwjidx0D,
 273                                          &c6_00,&c12_00);
 274
 275             /* Calculate table index by multiplying r with table scale and truncate to integer */
 276             rt               = _mm_mul_ps(r00,vftabscale);
 277             vfitab           = _mm_cvttps_epi32(rt);
 278 #ifdef __XOP__
 279             vfeps            = _mm_frcz_ps(rt);
 280 #else
 281             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 282 #endif
 283             twovfeps         = _mm_add_ps(vfeps,vfeps);
 284             vfitab           = _mm_slli_epi32(vfitab,3);
 285
 286             /* CUBIC SPLINE TABLE DISPERSION */
 287             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 288             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 289             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 290             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 291             _MM_TRANSPOSE4_PS(Y,F,G,H);
 292             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 293             VV               = _mm_macc_ps(vfeps,Fp,Y);
 294             vvdw6            = _mm_mul_ps(c6_00,VV);
 295             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 296             fvdw6            = _mm_mul_ps(c6_00,FF);
 297
 298             /* CUBIC SPLINE TABLE REPULSION */
 299             vfitab           = _mm_add_epi32(vfitab,ifour);
 300             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 301             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 302             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 303             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 304             _MM_TRANSPOSE4_PS(Y,F,G,H);
 305             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 306             VV               = _mm_macc_ps(vfeps,Fp,Y);
 307             vvdw12           = _mm_mul_ps(c12_00,VV);
 308             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 309             fvdw12           = _mm_mul_ps(c12_00,FF);
 310             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 311             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 315
 316             fscal            = fvdw;
 317
 318              /* Update vectorial force */
 319             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 320             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 321             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 322
 323             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 324             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 325             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             r10              = _mm_mul_ps(rsq10,rinv10);
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq10             = _mm_mul_ps(iq1,jq0);
 335
 336             /* EWALD ELECTROSTATICS */
 337
 338             /* Analytical PME correction */
 339             zeta2            = _mm_mul_ps(beta2,rsq10);
 340             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 341             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 342             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 343             felec            = _mm_mul_ps(qq10,felec);
 344             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 345             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 346             velec            = _mm_mul_ps(qq10,velec);
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353              /* Update vectorial force */
 354             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 355             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 356             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 357
 358             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 359             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 360             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             r20              = _mm_mul_ps(rsq20,rinv20);
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq20             = _mm_mul_ps(iq2,jq0);
 370
 371             /* EWALD ELECTROSTATICS */
 372
 373             /* Analytical PME correction */
 374             zeta2            = _mm_mul_ps(beta2,rsq20);
 375             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 376             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 377             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 378             felec            = _mm_mul_ps(qq20,felec);
 379             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 380             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 381             velec            = _mm_mul_ps(qq20,velec);
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             velecsum         = _mm_add_ps(velecsum,velec);
 385
 386             fscal            = felec;
 387
 388              /* Update vectorial force */
 389             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 390             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 391             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 392
 393             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 394             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 395             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 396
 397             /**************************
 398              * CALCULATE INTERACTIONS *
 399              **************************/
 400
 401             r30              = _mm_mul_ps(rsq30,rinv30);
 402
 403             /* Compute parameters for interactions between i and j atoms */
 404             qq30             = _mm_mul_ps(iq3,jq0);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Analytical PME correction */
 409             zeta2            = _mm_mul_ps(beta2,rsq30);
 410             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 411             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 412             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 413             felec            = _mm_mul_ps(qq30,felec);
 414             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 415             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 416             velec            = _mm_mul_ps(qq30,velec);
 417
 418             /* Update potential sum for this i atom from the interaction with this j atom. */
 419             velecsum         = _mm_add_ps(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423              /* Update vectorial force */
 424             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 425             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 426             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 427
 428             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 429             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 430             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 431
 432             fjptrA             = f+j_coord_offsetA;
 433             fjptrB             = f+j_coord_offsetB;
 434             fjptrC             = f+j_coord_offsetC;
 435             fjptrD             = f+j_coord_offsetD;
 436
 437             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 438
 439             /* Inner loop uses 146 flops */
 440         }
 441
 442         if(jidx<j_index_end)
 443         {
 444
 445             /* Get j neighbor index, and coordinate index */
 446             jnrlistA         = jjnr[jidx];
 447             jnrlistB         = jjnr[jidx+1];
 448             jnrlistC         = jjnr[jidx+2];
 449             jnrlistD         = jjnr[jidx+3];
 450             /* Sign of each element will be negative for non-real atoms.
 451              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 452              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 453              */
 454             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 455             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 456             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 457             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 458             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 459             j_coord_offsetA  = DIM*jnrA;
 460             j_coord_offsetB  = DIM*jnrB;
 461             j_coord_offsetC  = DIM*jnrC;
 462             j_coord_offsetD  = DIM*jnrD;
 463
 464             /* load j atom coordinates */
 465             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 466                                               x+j_coord_offsetC,x+j_coord_offsetD,
 467                                               &jx0,&jy0,&jz0);
 468
 469             /* Calculate displacement vector */
 470             dx00             = _mm_sub_ps(ix0,jx0);
 471             dy00             = _mm_sub_ps(iy0,jy0);
 472             dz00             = _mm_sub_ps(iz0,jz0);
 473             dx10             = _mm_sub_ps(ix1,jx0);
 474             dy10             = _mm_sub_ps(iy1,jy0);
 475             dz10             = _mm_sub_ps(iz1,jz0);
 476             dx20             = _mm_sub_ps(ix2,jx0);
 477             dy20             = _mm_sub_ps(iy2,jy0);
 478             dz20             = _mm_sub_ps(iz2,jz0);
 479             dx30             = _mm_sub_ps(ix3,jx0);
 480             dy30             = _mm_sub_ps(iy3,jy0);
 481             dz30             = _mm_sub_ps(iz3,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 485             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 486             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 487             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 488
 489             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 490             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 491             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 492             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 493
 494             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 495             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 496             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 497
 498             /* Load parameters for j particles */
 499             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 500                                                               charge+jnrC+0,charge+jnrD+0);
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502             vdwjidx0B        = 2*vdwtype[jnrB+0];
 503             vdwjidx0C        = 2*vdwtype[jnrC+0];
 504             vdwjidx0D        = 2*vdwtype[jnrD+0];
 505
 506             fjx0             = _mm_setzero_ps();
 507             fjy0             = _mm_setzero_ps();
 508             fjz0             = _mm_setzero_ps();
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             r00              = _mm_mul_ps(rsq00,rinv00);
 515             r00              = _mm_andnot_ps(dummy_mask,r00);
 516
 517             /* Compute parameters for interactions between i and j atoms */
 518             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 519                                          vdwparam+vdwioffset0+vdwjidx0B,
 520                                          vdwparam+vdwioffset0+vdwjidx0C,
 521                                          vdwparam+vdwioffset0+vdwjidx0D,
 522                                          &c6_00,&c12_00);
 523
 524             /* Calculate table index by multiplying r with table scale and truncate to integer */
 525             rt               = _mm_mul_ps(r00,vftabscale);
 526             vfitab           = _mm_cvttps_epi32(rt);
 527 #ifdef __XOP__
 528             vfeps            = _mm_frcz_ps(rt);
 529 #else
 530             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 531 #endif
 532             twovfeps         = _mm_add_ps(vfeps,vfeps);
 533             vfitab           = _mm_slli_epi32(vfitab,3);
 534
 535             /* CUBIC SPLINE TABLE DISPERSION */
 536             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 537             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 538             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 539             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 540             _MM_TRANSPOSE4_PS(Y,F,G,H);
 541             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 542             VV               = _mm_macc_ps(vfeps,Fp,Y);
 543             vvdw6            = _mm_mul_ps(c6_00,VV);
 544             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 545             fvdw6            = _mm_mul_ps(c6_00,FF);
 546
 547             /* CUBIC SPLINE TABLE REPULSION */
 548             vfitab           = _mm_add_epi32(vfitab,ifour);
 549             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 550             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 551             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 552             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 553             _MM_TRANSPOSE4_PS(Y,F,G,H);
 554             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 555             VV               = _mm_macc_ps(vfeps,Fp,Y);
 556             vvdw12           = _mm_mul_ps(c12_00,VV);
 557             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 558             fvdw12           = _mm_mul_ps(c12_00,FF);
 559             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 560             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 564             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 565
 566             fscal            = fvdw;
 567
 568             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 569
 570              /* Update vectorial force */
 571             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 572             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 573             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 574
 575             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 576             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 577             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             r10              = _mm_mul_ps(rsq10,rinv10);
 584             r10              = _mm_andnot_ps(dummy_mask,r10);
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq10             = _mm_mul_ps(iq1,jq0);
 588
 589             /* EWALD ELECTROSTATICS */
 590
 591             /* Analytical PME correction */
 592             zeta2            = _mm_mul_ps(beta2,rsq10);
 593             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 594             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 595             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 596             felec            = _mm_mul_ps(qq10,felec);
 597             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 598             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 599             velec            = _mm_mul_ps(qq10,velec);
 600
 601             /* Update potential sum for this i atom from the interaction with this j atom. */
 602             velec            = _mm_andnot_ps(dummy_mask,velec);
 603             velecsum         = _mm_add_ps(velecsum,velec);
 604
 605             fscal            = felec;
 606
 607             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 608
 609              /* Update vectorial force */
 610             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 611             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 612             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 613
 614             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 615             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 616             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             r20              = _mm_mul_ps(rsq20,rinv20);
 623             r20              = _mm_andnot_ps(dummy_mask,r20);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq20             = _mm_mul_ps(iq2,jq0);
 627
 628             /* EWALD ELECTROSTATICS */
 629
 630             /* Analytical PME correction */
 631             zeta2            = _mm_mul_ps(beta2,rsq20);
 632             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 633             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 634             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 635             felec            = _mm_mul_ps(qq20,felec);
 636             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 637             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 638             velec            = _mm_mul_ps(qq20,velec);
 639
 640             /* Update potential sum for this i atom from the interaction with this j atom. */
 641             velec            = _mm_andnot_ps(dummy_mask,velec);
 642             velecsum         = _mm_add_ps(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 647
 648              /* Update vectorial force */
 649             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 650             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 651             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 652
 653             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 654             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 655             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             r30              = _mm_mul_ps(rsq30,rinv30);
 662             r30              = _mm_andnot_ps(dummy_mask,r30);
 663
 664             /* Compute parameters for interactions between i and j atoms */
 665             qq30             = _mm_mul_ps(iq3,jq0);
 666
 667             /* EWALD ELECTROSTATICS */
 668
 669             /* Analytical PME correction */
 670             zeta2            = _mm_mul_ps(beta2,rsq30);
 671             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 672             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 673             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 674             felec            = _mm_mul_ps(qq30,felec);
 675             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 676             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv30);
 677             velec            = _mm_mul_ps(qq30,velec);
 678
 679             /* Update potential sum for this i atom from the interaction with this j atom. */
 680             velec            = _mm_andnot_ps(dummy_mask,velec);
 681             velecsum         = _mm_add_ps(velecsum,velec);
 682
 683             fscal            = felec;
 684
 685             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 686
 687              /* Update vectorial force */
 688             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 689             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 690             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 691
 692             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 693             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 694             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 695
 696             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 697             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 698             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 699             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 700
 701             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 702
 703             /* Inner loop uses 150 flops */
 704         }
 705
 706         /* End of innermost loop */
 707
 708         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 709                                               f+i_coord_offset,fshift+i_shift_offset);
 710
 711         ggid                        = gid[iidx];
 712         /* Update potential energies */
 713         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 714         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 715
 716         /* Increment number of inner iterations */
 717         inneriter                  += j_index_end - j_index_start;
 718
 719         /* Outer loop uses 26 flops */
 720     }
 721
 722     /* Increment number of outer iterations */
 723     outeriter        += nri;
 724
 725     /* Update outer/inner flops */
 726
 727     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*150);
 728 }
 729 /*
 730  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 731  * Electrostatics interaction: Ewald
 732  * VdW interaction:            CubicSplineTable
 733  * Geometry:                   Water4-Particle
 734  * Calculate force/pot:        Force
 735  */
 736 void
 737 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 738                     (t_nblist                    * gmx_restrict       nlist,
 739                      rvec                        * gmx_restrict          xx,
 740                      rvec                        * gmx_restrict          ff,
 741                      t_forcerec                  * gmx_restrict          fr,
 742                      t_mdatoms                   * gmx_restrict     mdatoms,
 743                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 744                      t_nrnb                      * gmx_restrict        nrnb)
 745 {
 746     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 747      * just 0 for non-waters.
 748      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 749      * jnr indices corresponding to data put in the four positions in the SIMD register.
 750      */
 751     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 752     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 753     int              jnrA,jnrB,jnrC,jnrD;
 754     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 755     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 756     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 757     real             rcutoff_scalar;
 758     real             *shiftvec,*fshift,*x,*f;
 759     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 760     real             scratch[4*DIM];
 761     __m128           fscal,rcutoff,rcutoff2,jidxall;
 762     int              vdwioffset0;
 763     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 764     int              vdwioffset1;
 765     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 766     int              vdwioffset2;
 767     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 768     int              vdwioffset3;
 769     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 770     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 771     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 772     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 773     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 774     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 775     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 776     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 777     real             *charge;
 778     int              nvdwtype;
 779     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 780     int              *vdwtype;
 781     real             *vdwparam;
 782     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 783     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 784     __m128i          vfitab;
 785     __m128i          ifour       = _mm_set1_epi32(4);
 786     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 787     real             *vftab;
 788     __m128i          ewitab;
 789     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 790     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 791     real             *ewtab;
 792     __m128           dummy_mask,cutoff_mask;
 793     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 794     __m128           one     = _mm_set1_ps(1.0);
 795     __m128           two     = _mm_set1_ps(2.0);
 796     x                = xx[0];
 797     f                = ff[0];
 798
 799     nri              = nlist->nri;
 800     iinr             = nlist->iinr;
 801     jindex           = nlist->jindex;
 802     jjnr             = nlist->jjnr;
 803     shiftidx         = nlist->shift;
 804     gid              = nlist->gid;
 805     shiftvec         = fr->shift_vec[0];
 806     fshift           = fr->fshift[0];
 807     facel            = _mm_set1_ps(fr->epsfac);
 808     charge           = mdatoms->chargeA;
 809     nvdwtype         = fr->ntype;
 810     vdwparam         = fr->nbfp;
 811     vdwtype          = mdatoms->typeA;
 812
 813     vftab            = kernel_data->table_vdw->data;
 814     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 815
 816     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 817     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 818     beta2            = _mm_mul_ps(beta,beta);
 819     beta3            = _mm_mul_ps(beta,beta2);
 820     ewtab            = fr->ic->tabq_coul_F;
 821     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 822     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 823
 824     /* Setup water-specific parameters */
 825     inr              = nlist->iinr[0];
 826     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 827     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 828     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 829     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 830
 831     /* Avoid stupid compiler warnings */
 832     jnrA = jnrB = jnrC = jnrD = 0;
 833     j_coord_offsetA = 0;
 834     j_coord_offsetB = 0;
 835     j_coord_offsetC = 0;
 836     j_coord_offsetD = 0;
 837
 838     outeriter        = 0;
 839     inneriter        = 0;
 840
 841     for(iidx=0;iidx<4*DIM;iidx++)
 842     {
 843         scratch[iidx] = 0.0;
 844     }
 845
 846     /* Start outer loop over neighborlists */
 847     for(iidx=0; iidx<nri; iidx++)
 848     {
 849         /* Load shift vector for this list */
 850         i_shift_offset   = DIM*shiftidx[iidx];
 851
 852         /* Load limits for loop over neighbors */
 853         j_index_start    = jindex[iidx];
 854         j_index_end      = jindex[iidx+1];
 855
 856         /* Get outer coordinate index */
 857         inr              = iinr[iidx];
 858         i_coord_offset   = DIM*inr;
 859
 860         /* Load i particle coords and add shift vector */
 861         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 862                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 863
 864         fix0             = _mm_setzero_ps();
 865         fiy0             = _mm_setzero_ps();
 866         fiz0             = _mm_setzero_ps();
 867         fix1             = _mm_setzero_ps();
 868         fiy1             = _mm_setzero_ps();
 869         fiz1             = _mm_setzero_ps();
 870         fix2             = _mm_setzero_ps();
 871         fiy2             = _mm_setzero_ps();
 872         fiz2             = _mm_setzero_ps();
 873         fix3             = _mm_setzero_ps();
 874         fiy3             = _mm_setzero_ps();
 875         fiz3             = _mm_setzero_ps();
 876
 877         /* Start inner kernel loop */
 878         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 879         {
 880
 881             /* Get j neighbor index, and coordinate index */
 882             jnrA             = jjnr[jidx];
 883             jnrB             = jjnr[jidx+1];
 884             jnrC             = jjnr[jidx+2];
 885             jnrD             = jjnr[jidx+3];
 886             j_coord_offsetA  = DIM*jnrA;
 887             j_coord_offsetB  = DIM*jnrB;
 888             j_coord_offsetC  = DIM*jnrC;
 889             j_coord_offsetD  = DIM*jnrD;
 890
 891             /* load j atom coordinates */
 892             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 893                                               x+j_coord_offsetC,x+j_coord_offsetD,
 894                                               &jx0,&jy0,&jz0);
 895
 896             /* Calculate displacement vector */
 897             dx00             = _mm_sub_ps(ix0,jx0);
 898             dy00             = _mm_sub_ps(iy0,jy0);
 899             dz00             = _mm_sub_ps(iz0,jz0);
 900             dx10             = _mm_sub_ps(ix1,jx0);
 901             dy10             = _mm_sub_ps(iy1,jy0);
 902             dz10             = _mm_sub_ps(iz1,jz0);
 903             dx20             = _mm_sub_ps(ix2,jx0);
 904             dy20             = _mm_sub_ps(iy2,jy0);
 905             dz20             = _mm_sub_ps(iz2,jz0);
 906             dx30             = _mm_sub_ps(ix3,jx0);
 907             dy30             = _mm_sub_ps(iy3,jy0);
 908             dz30             = _mm_sub_ps(iz3,jz0);
 909
 910             /* Calculate squared distance and things based on it */
 911             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 912             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 913             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 914             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 915
 916             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 917             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 918             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 919             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 920
 921             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 922             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 923             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 924
 925             /* Load parameters for j particles */
 926             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 927                                                               charge+jnrC+0,charge+jnrD+0);
 928             vdwjidx0A        = 2*vdwtype[jnrA+0];
 929             vdwjidx0B        = 2*vdwtype[jnrB+0];
 930             vdwjidx0C        = 2*vdwtype[jnrC+0];
 931             vdwjidx0D        = 2*vdwtype[jnrD+0];
 932
 933             fjx0             = _mm_setzero_ps();
 934             fjy0             = _mm_setzero_ps();
 935             fjz0             = _mm_setzero_ps();
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             r00              = _mm_mul_ps(rsq00,rinv00);
 942
 943             /* Compute parameters for interactions between i and j atoms */
 944             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 945                                          vdwparam+vdwioffset0+vdwjidx0B,
 946                                          vdwparam+vdwioffset0+vdwjidx0C,
 947                                          vdwparam+vdwioffset0+vdwjidx0D,
 948                                          &c6_00,&c12_00);
 949
 950             /* Calculate table index by multiplying r with table scale and truncate to integer */
 951             rt               = _mm_mul_ps(r00,vftabscale);
 952             vfitab           = _mm_cvttps_epi32(rt);
 953 #ifdef __XOP__
 954             vfeps            = _mm_frcz_ps(rt);
 955 #else
 956             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 957 #endif
 958             twovfeps         = _mm_add_ps(vfeps,vfeps);
 959             vfitab           = _mm_slli_epi32(vfitab,3);
 960
 961             /* CUBIC SPLINE TABLE DISPERSION */
 962             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 963             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 964             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 965             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 966             _MM_TRANSPOSE4_PS(Y,F,G,H);
 967             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 968             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 969             fvdw6            = _mm_mul_ps(c6_00,FF);
 970
 971             /* CUBIC SPLINE TABLE REPULSION */
 972             vfitab           = _mm_add_epi32(vfitab,ifour);
 973             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 974             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 975             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 976             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 977             _MM_TRANSPOSE4_PS(Y,F,G,H);
 978             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 979             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 980             fvdw12           = _mm_mul_ps(c12_00,FF);
 981             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 982
 983             fscal            = fvdw;
 984
 985              /* Update vectorial force */
 986             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 987             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 988             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 989
 990             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 991             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 992             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             r10              = _mm_mul_ps(rsq10,rinv10);
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq10             = _mm_mul_ps(iq1,jq0);
1002
1003             /* EWALD ELECTROSTATICS */
1004
1005             /* Analytical PME correction */
1006             zeta2            = _mm_mul_ps(beta2,rsq10);
1007             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1008             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1009             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1010             felec            = _mm_mul_ps(qq10,felec);
1011
1012             fscal            = felec;
1013
1014              /* Update vectorial force */
1015             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1016             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1017             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1018
1019             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1020             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1021             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1022
1023             /**************************
1024              * CALCULATE INTERACTIONS *
1025              **************************/
1026
1027             r20              = _mm_mul_ps(rsq20,rinv20);
1028
1029             /* Compute parameters for interactions between i and j atoms */
1030             qq20             = _mm_mul_ps(iq2,jq0);
1031
1032             /* EWALD ELECTROSTATICS */
1033
1034             /* Analytical PME correction */
1035             zeta2            = _mm_mul_ps(beta2,rsq20);
1036             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1037             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1038             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1039             felec            = _mm_mul_ps(qq20,felec);
1040
1041             fscal            = felec;
1042
1043              /* Update vectorial force */
1044             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1045             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1046             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1047
1048             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1049             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1050             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             r30              = _mm_mul_ps(rsq30,rinv30);
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             qq30             = _mm_mul_ps(iq3,jq0);
1060
1061             /* EWALD ELECTROSTATICS */
1062
1063             /* Analytical PME correction */
1064             zeta2            = _mm_mul_ps(beta2,rsq30);
1065             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1066             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1067             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1068             felec            = _mm_mul_ps(qq30,felec);
1069
1070             fscal            = felec;
1071
1072              /* Update vectorial force */
1073             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1074             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1075             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1076
1077             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1078             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1079             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1080
1081             fjptrA             = f+j_coord_offsetA;
1082             fjptrB             = f+j_coord_offsetB;
1083             fjptrC             = f+j_coord_offsetC;
1084             fjptrD             = f+j_coord_offsetD;
1085
1086             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1087
1088             /* Inner loop uses 135 flops */
1089         }
1090
1091         if(jidx<j_index_end)
1092         {
1093
1094             /* Get j neighbor index, and coordinate index */
1095             jnrlistA         = jjnr[jidx];
1096             jnrlistB         = jjnr[jidx+1];
1097             jnrlistC         = jjnr[jidx+2];
1098             jnrlistD         = jjnr[jidx+3];
1099             /* Sign of each element will be negative for non-real atoms.
1100              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1101              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1102              */
1103             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1104             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1105             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1106             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1107             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1108             j_coord_offsetA  = DIM*jnrA;
1109             j_coord_offsetB  = DIM*jnrB;
1110             j_coord_offsetC  = DIM*jnrC;
1111             j_coord_offsetD  = DIM*jnrD;
1112
1113             /* load j atom coordinates */
1114             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1115                                               x+j_coord_offsetC,x+j_coord_offsetD,
1116                                               &jx0,&jy0,&jz0);
1117
1118             /* Calculate displacement vector */
1119             dx00             = _mm_sub_ps(ix0,jx0);
1120             dy00             = _mm_sub_ps(iy0,jy0);
1121             dz00             = _mm_sub_ps(iz0,jz0);
1122             dx10             = _mm_sub_ps(ix1,jx0);
1123             dy10             = _mm_sub_ps(iy1,jy0);
1124             dz10             = _mm_sub_ps(iz1,jz0);
1125             dx20             = _mm_sub_ps(ix2,jx0);
1126             dy20             = _mm_sub_ps(iy2,jy0);
1127             dz20             = _mm_sub_ps(iz2,jz0);
1128             dx30             = _mm_sub_ps(ix3,jx0);
1129             dy30             = _mm_sub_ps(iy3,jy0);
1130             dz30             = _mm_sub_ps(iz3,jz0);
1131
1132             /* Calculate squared distance and things based on it */
1133             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1134             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1135             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1136             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1137
1138             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1139             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1140             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1141             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1142
1143             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1144             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1145             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1146
1147             /* Load parameters for j particles */
1148             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1149                                                               charge+jnrC+0,charge+jnrD+0);
1150             vdwjidx0A        = 2*vdwtype[jnrA+0];
1151             vdwjidx0B        = 2*vdwtype[jnrB+0];
1152             vdwjidx0C        = 2*vdwtype[jnrC+0];
1153             vdwjidx0D        = 2*vdwtype[jnrD+0];
1154
1155             fjx0             = _mm_setzero_ps();
1156             fjy0             = _mm_setzero_ps();
1157             fjz0             = _mm_setzero_ps();
1158
1159             /**************************
1160              * CALCULATE INTERACTIONS *
1161              **************************/
1162
1163             r00              = _mm_mul_ps(rsq00,rinv00);
1164             r00              = _mm_andnot_ps(dummy_mask,r00);
1165
1166             /* Compute parameters for interactions between i and j atoms */
1167             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1168                                          vdwparam+vdwioffset0+vdwjidx0B,
1169                                          vdwparam+vdwioffset0+vdwjidx0C,
1170                                          vdwparam+vdwioffset0+vdwjidx0D,
1171                                          &c6_00,&c12_00);
1172
1173             /* Calculate table index by multiplying r with table scale and truncate to integer */
1174             rt               = _mm_mul_ps(r00,vftabscale);
1175             vfitab           = _mm_cvttps_epi32(rt);
1176 #ifdef __XOP__
1177             vfeps            = _mm_frcz_ps(rt);
1178 #else
1179             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1180 #endif
1181             twovfeps         = _mm_add_ps(vfeps,vfeps);
1182             vfitab           = _mm_slli_epi32(vfitab,3);
1183
1184             /* CUBIC SPLINE TABLE DISPERSION */
1185             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1186             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1187             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1188             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1189             _MM_TRANSPOSE4_PS(Y,F,G,H);
1190             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1191             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1192             fvdw6            = _mm_mul_ps(c6_00,FF);
1193
1194             /* CUBIC SPLINE TABLE REPULSION */
1195             vfitab           = _mm_add_epi32(vfitab,ifour);
1196             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1197             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1198             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1199             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1200             _MM_TRANSPOSE4_PS(Y,F,G,H);
1201             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1202             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1203             fvdw12           = _mm_mul_ps(c12_00,FF);
1204             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1205
1206             fscal            = fvdw;
1207
1208             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1209
1210              /* Update vectorial force */
1211             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1212             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1213             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1214
1215             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1216             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1217             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             r10              = _mm_mul_ps(rsq10,rinv10);
1224             r10              = _mm_andnot_ps(dummy_mask,r10);
1225
1226             /* Compute parameters for interactions between i and j atoms */
1227             qq10             = _mm_mul_ps(iq1,jq0);
1228
1229             /* EWALD ELECTROSTATICS */
1230
1231             /* Analytical PME correction */
1232             zeta2            = _mm_mul_ps(beta2,rsq10);
1233             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1234             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1235             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1236             felec            = _mm_mul_ps(qq10,felec);
1237
1238             fscal            = felec;
1239
1240             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1241
1242              /* Update vectorial force */
1243             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1244             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1245             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1246
1247             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1248             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1249             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1250
1251             /**************************
1252              * CALCULATE INTERACTIONS *
1253              **************************/
1254
1255             r20              = _mm_mul_ps(rsq20,rinv20);
1256             r20              = _mm_andnot_ps(dummy_mask,r20);
1257
1258             /* Compute parameters for interactions between i and j atoms */
1259             qq20             = _mm_mul_ps(iq2,jq0);
1260
1261             /* EWALD ELECTROSTATICS */
1262
1263             /* Analytical PME correction */
1264             zeta2            = _mm_mul_ps(beta2,rsq20);
1265             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1266             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1267             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1268             felec            = _mm_mul_ps(qq20,felec);
1269
1270             fscal            = felec;
1271
1272             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1273
1274              /* Update vectorial force */
1275             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1276             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1277             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1278
1279             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1280             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1281             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1282
1283             /**************************
1284              * CALCULATE INTERACTIONS *
1285              **************************/
1286
1287             r30              = _mm_mul_ps(rsq30,rinv30);
1288             r30              = _mm_andnot_ps(dummy_mask,r30);
1289
1290             /* Compute parameters for interactions between i and j atoms */
1291             qq30             = _mm_mul_ps(iq3,jq0);
1292
1293             /* EWALD ELECTROSTATICS */
1294
1295             /* Analytical PME correction */
1296             zeta2            = _mm_mul_ps(beta2,rsq30);
1297             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1298             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1299             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1300             felec            = _mm_mul_ps(qq30,felec);
1301
1302             fscal            = felec;
1303
1304             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1305
1306              /* Update vectorial force */
1307             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1308             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1309             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1310
1311             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1312             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1313             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1314
1315             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1316             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1317             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1318             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1319
1320             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1321
1322             /* Inner loop uses 139 flops */
1323         }
1324
1325         /* End of innermost loop */
1326
1327         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1328                                               f+i_coord_offset,fshift+i_shift_offset);
1329
1330         /* Increment number of inner iterations */
1331         inneriter                  += j_index_end - j_index_start;
1332
1333         /* Outer loop uses 24 flops */
1334     }
1335
1336     /* Increment number of outer iterations */
1337     outeriter        += nri;
1338
1339     /* Update outer/inner flops */
1340
1341     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*139);
1342 }