src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 101     real             *vftab;
 102     __m128i          ewitab;
 103     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 128
 129     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 130     ewtab            = fr->ic->tabq_coul_FDV0;
 131     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 132     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 137     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 138     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 139     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 166
 167         fix0             = _mm_setzero_pd();
 168         fiy0             = _mm_setzero_pd();
 169         fiz0             = _mm_setzero_pd();
 170         fix1             = _mm_setzero_pd();
 171         fiy1             = _mm_setzero_pd();
 172         fiz1             = _mm_setzero_pd();
 173         fix2             = _mm_setzero_pd();
 174         fiy2             = _mm_setzero_pd();
 175         fiz2             = _mm_setzero_pd();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_pd();
 179         vvdwsum          = _mm_setzero_pd();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_pd(ix0,jx0);
 197             dy00             = _mm_sub_pd(iy0,jy0);
 198             dz00             = _mm_sub_pd(iz0,jz0);
 199             dx10             = _mm_sub_pd(ix1,jx0);
 200             dy10             = _mm_sub_pd(iy1,jy0);
 201             dz10             = _mm_sub_pd(iz1,jz0);
 202             dx20             = _mm_sub_pd(ix2,jx0);
 203             dy20             = _mm_sub_pd(iy2,jy0);
 204             dz20             = _mm_sub_pd(iz2,jz0);
 205
 206             /* Calculate squared distance and things based on it */
 207             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 208             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 209             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 210
 211             rinv00           = avx128fma_invsqrt_d(rsq00);
 212             rinv10           = avx128fma_invsqrt_d(rsq10);
 213             rinv20           = avx128fma_invsqrt_d(rsq20);
 214
 215             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 216             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 217             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 221             vdwjidx0A        = 2*vdwtype[jnrA+0];
 222             vdwjidx0B        = 2*vdwtype[jnrB+0];
 223
 224             fjx0             = _mm_setzero_pd();
 225             fjy0             = _mm_setzero_pd();
 226             fjz0             = _mm_setzero_pd();
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             r00              = _mm_mul_pd(rsq00,rinv00);
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm_mul_pd(iq0,jq0);
 236             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 237                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 238
 239             /* Calculate table index by multiplying r with table scale and truncate to integer */
 240             rt               = _mm_mul_pd(r00,vftabscale);
 241             vfitab           = _mm_cvttpd_epi32(rt);
 242 #ifdef __XOP__
 243             vfeps            = _mm_frcz_pd(rt);
 244 #else
 245             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 246 #endif
 247             twovfeps         = _mm_add_pd(vfeps,vfeps);
 248             vfitab           = _mm_slli_epi32(vfitab,3);
 249
 250             /* EWALD ELECTROSTATICS */
 251
 252             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 253             ewrt             = _mm_mul_pd(r00,ewtabscale);
 254             ewitab           = _mm_cvttpd_epi32(ewrt);
 255 #ifdef __XOP__
 256             eweps            = _mm_frcz_pd(ewrt);
 257 #else
 258             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 259 #endif
 260             twoeweps         = _mm_add_pd(eweps,eweps);
 261             ewitab           = _mm_slli_epi32(ewitab,2);
 262             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 263             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 264             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 265             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 266             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 267             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 268             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 269             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 270             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 271             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 272
 273             /* CUBIC SPLINE TABLE DISPERSION */
 274             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 276             GMX_MM_TRANSPOSE2_PD(Y,F);
 277             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 278             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 279             GMX_MM_TRANSPOSE2_PD(G,H);
 280             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 281             VV               = _mm_macc_pd(vfeps,Fp,Y);
 282             vvdw6            = _mm_mul_pd(c6_00,VV);
 283             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 284             fvdw6            = _mm_mul_pd(c6_00,FF);
 285
 286             /* CUBIC SPLINE TABLE REPULSION */
 287             vfitab           = _mm_add_epi32(vfitab,ifour);
 288             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 289             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 290             GMX_MM_TRANSPOSE2_PD(Y,F);
 291             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 292             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 293             GMX_MM_TRANSPOSE2_PD(G,H);
 294             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 295             VV               = _mm_macc_pd(vfeps,Fp,Y);
 296             vvdw12           = _mm_mul_pd(c12_00,VV);
 297             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 298             fvdw12           = _mm_mul_pd(c12_00,FF);
 299             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 300             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm_add_pd(velecsum,velec);
 304             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 305
 306             fscal            = _mm_add_pd(felec,fvdw);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 310             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 311             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 312
 313             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 314             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 315             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r10              = _mm_mul_pd(rsq10,rinv10);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_pd(iq1,jq0);
 325
 326             /* EWALD ELECTROSTATICS */
 327
 328             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 329             ewrt             = _mm_mul_pd(r10,ewtabscale);
 330             ewitab           = _mm_cvttpd_epi32(ewrt);
 331 #ifdef __XOP__
 332             eweps            = _mm_frcz_pd(ewrt);
 333 #else
 334             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 335 #endif
 336             twoeweps         = _mm_add_pd(eweps,eweps);
 337             ewitab           = _mm_slli_epi32(ewitab,2);
 338             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 339             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 340             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 341             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 342             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 343             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 344             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 345             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 346             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 347             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm_add_pd(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354             /* Update vectorial force */
 355             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 356             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 357             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 358
 359             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 360             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 361             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r20              = _mm_mul_pd(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm_mul_pd(iq2,jq0);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 375             ewrt             = _mm_mul_pd(r20,ewtabscale);
 376             ewitab           = _mm_cvttpd_epi32(ewrt);
 377 #ifdef __XOP__
 378             eweps            = _mm_frcz_pd(ewrt);
 379 #else
 380             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 381 #endif
 382             twoeweps         = _mm_add_pd(eweps,eweps);
 383             ewitab           = _mm_slli_epi32(ewitab,2);
 384             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 385             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 386             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 387             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 388             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 389             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 390             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 391             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 392             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 393             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm_add_pd(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             /* Update vectorial force */
 401             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 402             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 403             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 404
 405             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 406             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 407             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 408
 409             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 410
 411             /* Inner loop uses 169 flops */
 412         }
 413
 414         if(jidx<j_index_end)
 415         {
 416
 417             jnrA             = jjnr[jidx];
 418             j_coord_offsetA  = DIM*jnrA;
 419
 420             /* load j atom coordinates */
 421             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 422                                               &jx0,&jy0,&jz0);
 423
 424             /* Calculate displacement vector */
 425             dx00             = _mm_sub_pd(ix0,jx0);
 426             dy00             = _mm_sub_pd(iy0,jy0);
 427             dz00             = _mm_sub_pd(iz0,jz0);
 428             dx10             = _mm_sub_pd(ix1,jx0);
 429             dy10             = _mm_sub_pd(iy1,jy0);
 430             dz10             = _mm_sub_pd(iz1,jz0);
 431             dx20             = _mm_sub_pd(ix2,jx0);
 432             dy20             = _mm_sub_pd(iy2,jy0);
 433             dz20             = _mm_sub_pd(iz2,jz0);
 434
 435             /* Calculate squared distance and things based on it */
 436             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 437             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 438             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 439
 440             rinv00           = avx128fma_invsqrt_d(rsq00);
 441             rinv10           = avx128fma_invsqrt_d(rsq10);
 442             rinv20           = avx128fma_invsqrt_d(rsq20);
 443
 444             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 445             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 446             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 447
 448             /* Load parameters for j particles */
 449             jq0              = _mm_load_sd(charge+jnrA+0);
 450             vdwjidx0A        = 2*vdwtype[jnrA+0];
 451
 452             fjx0             = _mm_setzero_pd();
 453             fjy0             = _mm_setzero_pd();
 454             fjz0             = _mm_setzero_pd();
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             r00              = _mm_mul_pd(rsq00,rinv00);
 461
 462             /* Compute parameters for interactions between i and j atoms */
 463             qq00             = _mm_mul_pd(iq0,jq0);
 464             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 465
 466             /* Calculate table index by multiplying r with table scale and truncate to integer */
 467             rt               = _mm_mul_pd(r00,vftabscale);
 468             vfitab           = _mm_cvttpd_epi32(rt);
 469 #ifdef __XOP__
 470             vfeps            = _mm_frcz_pd(rt);
 471 #else
 472             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 473 #endif
 474             twovfeps         = _mm_add_pd(vfeps,vfeps);
 475             vfitab           = _mm_slli_epi32(vfitab,3);
 476
 477             /* EWALD ELECTROSTATICS */
 478
 479             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 480             ewrt             = _mm_mul_pd(r00,ewtabscale);
 481             ewitab           = _mm_cvttpd_epi32(ewrt);
 482 #ifdef __XOP__
 483             eweps            = _mm_frcz_pd(ewrt);
 484 #else
 485             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 486 #endif
 487             twoeweps         = _mm_add_pd(eweps,eweps);
 488             ewitab           = _mm_slli_epi32(ewitab,2);
 489             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 490             ewtabD           = _mm_setzero_pd();
 491             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 492             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 493             ewtabFn          = _mm_setzero_pd();
 494             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 495             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 496             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 497             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 498             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 499
 500             /* CUBIC SPLINE TABLE DISPERSION */
 501             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 502             F                = _mm_setzero_pd();
 503             GMX_MM_TRANSPOSE2_PD(Y,F);
 504             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 505             H                = _mm_setzero_pd();
 506             GMX_MM_TRANSPOSE2_PD(G,H);
 507             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 508             VV               = _mm_macc_pd(vfeps,Fp,Y);
 509             vvdw6            = _mm_mul_pd(c6_00,VV);
 510             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 511             fvdw6            = _mm_mul_pd(c6_00,FF);
 512
 513             /* CUBIC SPLINE TABLE REPULSION */
 514             vfitab           = _mm_add_epi32(vfitab,ifour);
 515             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 516             F                = _mm_setzero_pd();
 517             GMX_MM_TRANSPOSE2_PD(Y,F);
 518             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 519             H                = _mm_setzero_pd();
 520             GMX_MM_TRANSPOSE2_PD(G,H);
 521             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 522             VV               = _mm_macc_pd(vfeps,Fp,Y);
 523             vvdw12           = _mm_mul_pd(c12_00,VV);
 524             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 525             fvdw12           = _mm_mul_pd(c12_00,FF);
 526             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 527             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 531             velecsum         = _mm_add_pd(velecsum,velec);
 532             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 533             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 534
 535             fscal            = _mm_add_pd(felec,fvdw);
 536
 537             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 538
 539             /* Update vectorial force */
 540             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 541             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 542             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 543
 544             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 545             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 546             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             r10              = _mm_mul_pd(rsq10,rinv10);
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq10             = _mm_mul_pd(iq1,jq0);
 556
 557             /* EWALD ELECTROSTATICS */
 558
 559             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 560             ewrt             = _mm_mul_pd(r10,ewtabscale);
 561             ewitab           = _mm_cvttpd_epi32(ewrt);
 562 #ifdef __XOP__
 563             eweps            = _mm_frcz_pd(ewrt);
 564 #else
 565             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 566 #endif
 567             twoeweps         = _mm_add_pd(eweps,eweps);
 568             ewitab           = _mm_slli_epi32(ewitab,2);
 569             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 570             ewtabD           = _mm_setzero_pd();
 571             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 572             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 573             ewtabFn          = _mm_setzero_pd();
 574             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 575             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 576             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 577             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 578             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 582             velecsum         = _mm_add_pd(velecsum,velec);
 583
 584             fscal            = felec;
 585
 586             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 587
 588             /* Update vectorial force */
 589             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 590             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 591             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 592
 593             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 594             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 595             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 596
 597             /**************************
 598              * CALCULATE INTERACTIONS *
 599              **************************/
 600
 601             r20              = _mm_mul_pd(rsq20,rinv20);
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq20             = _mm_mul_pd(iq2,jq0);
 605
 606             /* EWALD ELECTROSTATICS */
 607
 608             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 609             ewrt             = _mm_mul_pd(r20,ewtabscale);
 610             ewitab           = _mm_cvttpd_epi32(ewrt);
 611 #ifdef __XOP__
 612             eweps            = _mm_frcz_pd(ewrt);
 613 #else
 614             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 615 #endif
 616             twoeweps         = _mm_add_pd(eweps,eweps);
 617             ewitab           = _mm_slli_epi32(ewitab,2);
 618             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 619             ewtabD           = _mm_setzero_pd();
 620             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 621             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 622             ewtabFn          = _mm_setzero_pd();
 623             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 624             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 625             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 626             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 627             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 628
 629             /* Update potential sum for this i atom from the interaction with this j atom. */
 630             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 631             velecsum         = _mm_add_pd(velecsum,velec);
 632
 633             fscal            = felec;
 634
 635             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 636
 637             /* Update vectorial force */
 638             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 639             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 640             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 641
 642             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 643             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 644             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 645
 646             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 647
 648             /* Inner loop uses 169 flops */
 649         }
 650
 651         /* End of innermost loop */
 652
 653         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 654                                               f+i_coord_offset,fshift+i_shift_offset);
 655
 656         ggid                        = gid[iidx];
 657         /* Update potential energies */
 658         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 659         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 660
 661         /* Increment number of inner iterations */
 662         inneriter                  += j_index_end - j_index_start;
 663
 664         /* Outer loop uses 20 flops */
 665     }
 666
 667     /* Increment number of outer iterations */
 668     outeriter        += nri;
 669
 670     /* Update outer/inner flops */
 671
 672     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*169);
 673 }
 674 /*
 675  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 676  * Electrostatics interaction: Ewald
 677  * VdW interaction:            CubicSplineTable
 678  * Geometry:                   Water3-Particle
 679  * Calculate force/pot:        Force
 680  */
 681 void
 682 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 683                     (t_nblist                    * gmx_restrict       nlist,
 684                      rvec                        * gmx_restrict          xx,
 685                      rvec                        * gmx_restrict          ff,
 686                      struct t_forcerec           * gmx_restrict          fr,
 687                      t_mdatoms                   * gmx_restrict     mdatoms,
 688                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 689                      t_nrnb                      * gmx_restrict        nrnb)
 690 {
 691     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 692      * just 0 for non-waters.
 693      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 694      * jnr indices corresponding to data put in the four positions in the SIMD register.
 695      */
 696     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 697     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 698     int              jnrA,jnrB;
 699     int              j_coord_offsetA,j_coord_offsetB;
 700     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 701     real             rcutoff_scalar;
 702     real             *shiftvec,*fshift,*x,*f;
 703     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 704     int              vdwioffset0;
 705     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 706     int              vdwioffset1;
 707     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 708     int              vdwioffset2;
 709     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 710     int              vdwjidx0A,vdwjidx0B;
 711     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 712     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 713     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 714     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 715     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 716     real             *charge;
 717     int              nvdwtype;
 718     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 719     int              *vdwtype;
 720     real             *vdwparam;
 721     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 722     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 723     __m128i          vfitab;
 724     __m128i          ifour       = _mm_set1_epi32(4);
 725     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 726     real             *vftab;
 727     __m128i          ewitab;
 728     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 729     real             *ewtab;
 730     __m128d          dummy_mask,cutoff_mask;
 731     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 732     __m128d          one     = _mm_set1_pd(1.0);
 733     __m128d          two     = _mm_set1_pd(2.0);
 734     x                = xx[0];
 735     f                = ff[0];
 736
 737     nri              = nlist->nri;
 738     iinr             = nlist->iinr;
 739     jindex           = nlist->jindex;
 740     jjnr             = nlist->jjnr;
 741     shiftidx         = nlist->shift;
 742     gid              = nlist->gid;
 743     shiftvec         = fr->shift_vec[0];
 744     fshift           = fr->fshift[0];
 745     facel            = _mm_set1_pd(fr->ic->epsfac);
 746     charge           = mdatoms->chargeA;
 747     nvdwtype         = fr->ntype;
 748     vdwparam         = fr->nbfp;
 749     vdwtype          = mdatoms->typeA;
 750
 751     vftab            = kernel_data->table_vdw->data;
 752     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 753
 754     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 755     ewtab            = fr->ic->tabq_coul_F;
 756     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 757     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 758
 759     /* Setup water-specific parameters */
 760     inr              = nlist->iinr[0];
 761     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 762     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 763     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 764     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 765
 766     /* Avoid stupid compiler warnings */
 767     jnrA = jnrB = 0;
 768     j_coord_offsetA = 0;
 769     j_coord_offsetB = 0;
 770
 771     outeriter        = 0;
 772     inneriter        = 0;
 773
 774     /* Start outer loop over neighborlists */
 775     for(iidx=0; iidx<nri; iidx++)
 776     {
 777         /* Load shift vector for this list */
 778         i_shift_offset   = DIM*shiftidx[iidx];
 779
 780         /* Load limits for loop over neighbors */
 781         j_index_start    = jindex[iidx];
 782         j_index_end      = jindex[iidx+1];
 783
 784         /* Get outer coordinate index */
 785         inr              = iinr[iidx];
 786         i_coord_offset   = DIM*inr;
 787
 788         /* Load i particle coords and add shift vector */
 789         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 790                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 791
 792         fix0             = _mm_setzero_pd();
 793         fiy0             = _mm_setzero_pd();
 794         fiz0             = _mm_setzero_pd();
 795         fix1             = _mm_setzero_pd();
 796         fiy1             = _mm_setzero_pd();
 797         fiz1             = _mm_setzero_pd();
 798         fix2             = _mm_setzero_pd();
 799         fiy2             = _mm_setzero_pd();
 800         fiz2             = _mm_setzero_pd();
 801
 802         /* Start inner kernel loop */
 803         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 804         {
 805
 806             /* Get j neighbor index, and coordinate index */
 807             jnrA             = jjnr[jidx];
 808             jnrB             = jjnr[jidx+1];
 809             j_coord_offsetA  = DIM*jnrA;
 810             j_coord_offsetB  = DIM*jnrB;
 811
 812             /* load j atom coordinates */
 813             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 814                                               &jx0,&jy0,&jz0);
 815
 816             /* Calculate displacement vector */
 817             dx00             = _mm_sub_pd(ix0,jx0);
 818             dy00             = _mm_sub_pd(iy0,jy0);
 819             dz00             = _mm_sub_pd(iz0,jz0);
 820             dx10             = _mm_sub_pd(ix1,jx0);
 821             dy10             = _mm_sub_pd(iy1,jy0);
 822             dz10             = _mm_sub_pd(iz1,jz0);
 823             dx20             = _mm_sub_pd(ix2,jx0);
 824             dy20             = _mm_sub_pd(iy2,jy0);
 825             dz20             = _mm_sub_pd(iz2,jz0);
 826
 827             /* Calculate squared distance and things based on it */
 828             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 829             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 830             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 831
 832             rinv00           = avx128fma_invsqrt_d(rsq00);
 833             rinv10           = avx128fma_invsqrt_d(rsq10);
 834             rinv20           = avx128fma_invsqrt_d(rsq20);
 835
 836             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 837             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 838             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 839
 840             /* Load parameters for j particles */
 841             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 842             vdwjidx0A        = 2*vdwtype[jnrA+0];
 843             vdwjidx0B        = 2*vdwtype[jnrB+0];
 844
 845             fjx0             = _mm_setzero_pd();
 846             fjy0             = _mm_setzero_pd();
 847             fjz0             = _mm_setzero_pd();
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             r00              = _mm_mul_pd(rsq00,rinv00);
 854
 855             /* Compute parameters for interactions between i and j atoms */
 856             qq00             = _mm_mul_pd(iq0,jq0);
 857             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 858                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 859
 860             /* Calculate table index by multiplying r with table scale and truncate to integer */
 861             rt               = _mm_mul_pd(r00,vftabscale);
 862             vfitab           = _mm_cvttpd_epi32(rt);
 863 #ifdef __XOP__
 864             vfeps            = _mm_frcz_pd(rt);
 865 #else
 866             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 867 #endif
 868             twovfeps         = _mm_add_pd(vfeps,vfeps);
 869             vfitab           = _mm_slli_epi32(vfitab,3);
 870
 871             /* EWALD ELECTROSTATICS */
 872
 873             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 874             ewrt             = _mm_mul_pd(r00,ewtabscale);
 875             ewitab           = _mm_cvttpd_epi32(ewrt);
 876 #ifdef __XOP__
 877             eweps            = _mm_frcz_pd(ewrt);
 878 #else
 879             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 880 #endif
 881             twoeweps         = _mm_add_pd(eweps,eweps);
 882             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 883                                          &ewtabF,&ewtabFn);
 884             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 885             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 886
 887             /* CUBIC SPLINE TABLE DISPERSION */
 888             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 889             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 890             GMX_MM_TRANSPOSE2_PD(Y,F);
 891             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 892             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 893             GMX_MM_TRANSPOSE2_PD(G,H);
 894             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 895             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 896             fvdw6            = _mm_mul_pd(c6_00,FF);
 897
 898             /* CUBIC SPLINE TABLE REPULSION */
 899             vfitab           = _mm_add_epi32(vfitab,ifour);
 900             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 901             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 902             GMX_MM_TRANSPOSE2_PD(Y,F);
 903             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 904             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 905             GMX_MM_TRANSPOSE2_PD(G,H);
 906             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 907             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 908             fvdw12           = _mm_mul_pd(c12_00,FF);
 909             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 910
 911             fscal            = _mm_add_pd(felec,fvdw);
 912
 913             /* Update vectorial force */
 914             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 915             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 916             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 917
 918             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 919             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 920             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             r10              = _mm_mul_pd(rsq10,rinv10);
 927
 928             /* Compute parameters for interactions between i and j atoms */
 929             qq10             = _mm_mul_pd(iq1,jq0);
 930
 931             /* EWALD ELECTROSTATICS */
 932
 933             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 934             ewrt             = _mm_mul_pd(r10,ewtabscale);
 935             ewitab           = _mm_cvttpd_epi32(ewrt);
 936 #ifdef __XOP__
 937             eweps            = _mm_frcz_pd(ewrt);
 938 #else
 939             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 940 #endif
 941             twoeweps         = _mm_add_pd(eweps,eweps);
 942             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 943                                          &ewtabF,&ewtabFn);
 944             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 945             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 946
 947             fscal            = felec;
 948
 949             /* Update vectorial force */
 950             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 951             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 952             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 953
 954             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 955             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 956             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 957
 958             /**************************
 959              * CALCULATE INTERACTIONS *
 960              **************************/
 961
 962             r20              = _mm_mul_pd(rsq20,rinv20);
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq20             = _mm_mul_pd(iq2,jq0);
 966
 967             /* EWALD ELECTROSTATICS */
 968
 969             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 970             ewrt             = _mm_mul_pd(r20,ewtabscale);
 971             ewitab           = _mm_cvttpd_epi32(ewrt);
 972 #ifdef __XOP__
 973             eweps            = _mm_frcz_pd(ewrt);
 974 #else
 975             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 976 #endif
 977             twoeweps         = _mm_add_pd(eweps,eweps);
 978             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 979                                          &ewtabF,&ewtabFn);
 980             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 981             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 982
 983             fscal            = felec;
 984
 985             /* Update vectorial force */
 986             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 987             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 988             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 989
 990             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 991             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 992             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 993
 994             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 995
 996             /* Inner loop uses 146 flops */
 997         }
 998
 999         if(jidx<j_index_end)
1000         {
1001
1002             jnrA             = jjnr[jidx];
1003             j_coord_offsetA  = DIM*jnrA;
1004
1005             /* load j atom coordinates */
1006             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1007                                               &jx0,&jy0,&jz0);
1008
1009             /* Calculate displacement vector */
1010             dx00             = _mm_sub_pd(ix0,jx0);
1011             dy00             = _mm_sub_pd(iy0,jy0);
1012             dz00             = _mm_sub_pd(iz0,jz0);
1013             dx10             = _mm_sub_pd(ix1,jx0);
1014             dy10             = _mm_sub_pd(iy1,jy0);
1015             dz10             = _mm_sub_pd(iz1,jz0);
1016             dx20             = _mm_sub_pd(ix2,jx0);
1017             dy20             = _mm_sub_pd(iy2,jy0);
1018             dz20             = _mm_sub_pd(iz2,jz0);
1019
1020             /* Calculate squared distance and things based on it */
1021             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1022             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1023             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1024
1025             rinv00           = avx128fma_invsqrt_d(rsq00);
1026             rinv10           = avx128fma_invsqrt_d(rsq10);
1027             rinv20           = avx128fma_invsqrt_d(rsq20);
1028
1029             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1030             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1031             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1032
1033             /* Load parameters for j particles */
1034             jq0              = _mm_load_sd(charge+jnrA+0);
1035             vdwjidx0A        = 2*vdwtype[jnrA+0];
1036
1037             fjx0             = _mm_setzero_pd();
1038             fjy0             = _mm_setzero_pd();
1039             fjz0             = _mm_setzero_pd();
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             r00              = _mm_mul_pd(rsq00,rinv00);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq00             = _mm_mul_pd(iq0,jq0);
1049             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1050
1051             /* Calculate table index by multiplying r with table scale and truncate to integer */
1052             rt               = _mm_mul_pd(r00,vftabscale);
1053             vfitab           = _mm_cvttpd_epi32(rt);
1054 #ifdef __XOP__
1055             vfeps            = _mm_frcz_pd(rt);
1056 #else
1057             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1058 #endif
1059             twovfeps         = _mm_add_pd(vfeps,vfeps);
1060             vfitab           = _mm_slli_epi32(vfitab,3);
1061
1062             /* EWALD ELECTROSTATICS */
1063
1064             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1065             ewrt             = _mm_mul_pd(r00,ewtabscale);
1066             ewitab           = _mm_cvttpd_epi32(ewrt);
1067 #ifdef __XOP__
1068             eweps            = _mm_frcz_pd(ewrt);
1069 #else
1070             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1071 #endif
1072             twoeweps         = _mm_add_pd(eweps,eweps);
1073             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1074             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1075             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1076
1077             /* CUBIC SPLINE TABLE DISPERSION */
1078             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1079             F                = _mm_setzero_pd();
1080             GMX_MM_TRANSPOSE2_PD(Y,F);
1081             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1082             H                = _mm_setzero_pd();
1083             GMX_MM_TRANSPOSE2_PD(G,H);
1084             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1085             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1086             fvdw6            = _mm_mul_pd(c6_00,FF);
1087
1088             /* CUBIC SPLINE TABLE REPULSION */
1089             vfitab           = _mm_add_epi32(vfitab,ifour);
1090             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1091             F                = _mm_setzero_pd();
1092             GMX_MM_TRANSPOSE2_PD(Y,F);
1093             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1094             H                = _mm_setzero_pd();
1095             GMX_MM_TRANSPOSE2_PD(G,H);
1096             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1097             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1098             fvdw12           = _mm_mul_pd(c12_00,FF);
1099             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1100
1101             fscal            = _mm_add_pd(felec,fvdw);
1102
1103             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1104
1105             /* Update vectorial force */
1106             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1107             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1108             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1109
1110             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1111             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1112             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             r10              = _mm_mul_pd(rsq10,rinv10);
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             qq10             = _mm_mul_pd(iq1,jq0);
1122
1123             /* EWALD ELECTROSTATICS */
1124
1125             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1126             ewrt             = _mm_mul_pd(r10,ewtabscale);
1127             ewitab           = _mm_cvttpd_epi32(ewrt);
1128 #ifdef __XOP__
1129             eweps            = _mm_frcz_pd(ewrt);
1130 #else
1131             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1132 #endif
1133             twoeweps         = _mm_add_pd(eweps,eweps);
1134             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1135             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1136             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1137
1138             fscal            = felec;
1139
1140             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1141
1142             /* Update vectorial force */
1143             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1144             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1145             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1146
1147             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1148             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1149             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1150
1151             /**************************
1152              * CALCULATE INTERACTIONS *
1153              **************************/
1154
1155             r20              = _mm_mul_pd(rsq20,rinv20);
1156
1157             /* Compute parameters for interactions between i and j atoms */
1158             qq20             = _mm_mul_pd(iq2,jq0);
1159
1160             /* EWALD ELECTROSTATICS */
1161
1162             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1163             ewrt             = _mm_mul_pd(r20,ewtabscale);
1164             ewitab           = _mm_cvttpd_epi32(ewrt);
1165 #ifdef __XOP__
1166             eweps            = _mm_frcz_pd(ewrt);
1167 #else
1168             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1169 #endif
1170             twoeweps         = _mm_add_pd(eweps,eweps);
1171             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1172             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1173             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1174
1175             fscal            = felec;
1176
1177             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1178
1179             /* Update vectorial force */
1180             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1181             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1182             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1183
1184             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1185             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1186             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1187
1188             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1189
1190             /* Inner loop uses 146 flops */
1191         }
1192
1193         /* End of innermost loop */
1194
1195         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1196                                               f+i_coord_offset,fshift+i_shift_offset);
1197
1198         /* Increment number of inner iterations */
1199         inneriter                  += j_index_end - j_index_start;
1200
1201         /* Outer loop uses 18 flops */
1202     }
1203
1204     /* Increment number of outer iterations */
1205     outeriter        += nri;
1206
1207     /* Update outer/inner flops */
1208
1209     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*146);
1210 }