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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 102     real             *vftab;
 103     __m128i          ewitab;
 104     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     real             *ewtab;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     vftab            = kernel_data->table_vdw->data;
 128     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 129
 130     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 131     ewtab            = fr->ic->tabq_coul_FDV0;
 132     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 133     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 138     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 139     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 167
 168         fix0             = _mm_setzero_pd();
 169         fiy0             = _mm_setzero_pd();
 170         fiz0             = _mm_setzero_pd();
 171         fix1             = _mm_setzero_pd();
 172         fiy1             = _mm_setzero_pd();
 173         fiz1             = _mm_setzero_pd();
 174         fix2             = _mm_setzero_pd();
 175         fiy2             = _mm_setzero_pd();
 176         fiz2             = _mm_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_pd();
 180         vvdwsum          = _mm_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 209             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 211
 212             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 213             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 214             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 215
 216             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 217             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 218             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224
 225             fjx0             = _mm_setzero_pd();
 226             fjy0             = _mm_setzero_pd();
 227             fjz0             = _mm_setzero_pd();
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             r00              = _mm_mul_pd(rsq00,rinv00);
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             qq00             = _mm_mul_pd(iq0,jq0);
 237             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 238                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 239
 240             /* Calculate table index by multiplying r with table scale and truncate to integer */
 241             rt               = _mm_mul_pd(r00,vftabscale);
 242             vfitab           = _mm_cvttpd_epi32(rt);
 243 #ifdef __XOP__
 244             vfeps            = _mm_frcz_pd(rt);
 245 #else
 246             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 247 #endif
 248             twovfeps         = _mm_add_pd(vfeps,vfeps);
 249             vfitab           = _mm_slli_epi32(vfitab,3);
 250
 251             /* EWALD ELECTROSTATICS */
 252
 253             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 254             ewrt             = _mm_mul_pd(r00,ewtabscale);
 255             ewitab           = _mm_cvttpd_epi32(ewrt);
 256 #ifdef __XOP__
 257             eweps            = _mm_frcz_pd(ewrt);
 258 #else
 259             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 260 #endif
 261             twoeweps         = _mm_add_pd(eweps,eweps);
 262             ewitab           = _mm_slli_epi32(ewitab,2);
 263             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 264             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 265             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 266             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 267             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 268             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 269             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 270             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 271             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 272             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 273
 274             /* CUBIC SPLINE TABLE DISPERSION */
 275             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 277             GMX_MM_TRANSPOSE2_PD(Y,F);
 278             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 279             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 280             GMX_MM_TRANSPOSE2_PD(G,H);
 281             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 282             VV               = _mm_macc_pd(vfeps,Fp,Y);
 283             vvdw6            = _mm_mul_pd(c6_00,VV);
 284             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 285             fvdw6            = _mm_mul_pd(c6_00,FF);
 286
 287             /* CUBIC SPLINE TABLE REPULSION */
 288             vfitab           = _mm_add_epi32(vfitab,ifour);
 289             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 290             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 291             GMX_MM_TRANSPOSE2_PD(Y,F);
 292             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 293             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 294             GMX_MM_TRANSPOSE2_PD(G,H);
 295             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 296             VV               = _mm_macc_pd(vfeps,Fp,Y);
 297             vvdw12           = _mm_mul_pd(c12_00,VV);
 298             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 299             fvdw12           = _mm_mul_pd(c12_00,FF);
 300             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 301             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velecsum         = _mm_add_pd(velecsum,velec);
 305             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 306
 307             fscal            = _mm_add_pd(felec,fvdw);
 308
 309             /* Update vectorial force */
 310             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 311             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 312             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 313
 314             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 315             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 316             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             r10              = _mm_mul_pd(rsq10,rinv10);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq10             = _mm_mul_pd(iq1,jq0);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 330             ewrt             = _mm_mul_pd(r10,ewtabscale);
 331             ewitab           = _mm_cvttpd_epi32(ewrt);
 332 #ifdef __XOP__
 333             eweps            = _mm_frcz_pd(ewrt);
 334 #else
 335             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 336 #endif
 337             twoeweps         = _mm_add_pd(eweps,eweps);
 338             ewitab           = _mm_slli_epi32(ewitab,2);
 339             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 340             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 341             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 342             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 343             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 344             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 345             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 346             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 347             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 348             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velecsum         = _mm_add_pd(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             /* Update vectorial force */
 356             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 357             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 358             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 359
 360             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 361             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 362             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             r20              = _mm_mul_pd(rsq20,rinv20);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq20             = _mm_mul_pd(iq2,jq0);
 372
 373             /* EWALD ELECTROSTATICS */
 374
 375             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 376             ewrt             = _mm_mul_pd(r20,ewtabscale);
 377             ewitab           = _mm_cvttpd_epi32(ewrt);
 378 #ifdef __XOP__
 379             eweps            = _mm_frcz_pd(ewrt);
 380 #else
 381             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 382 #endif
 383             twoeweps         = _mm_add_pd(eweps,eweps);
 384             ewitab           = _mm_slli_epi32(ewitab,2);
 385             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 386             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 387             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 388             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 389             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 390             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 391             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 392             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 393             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 394             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 395
 396             /* Update potential sum for this i atom from the interaction with this j atom. */
 397             velecsum         = _mm_add_pd(velecsum,velec);
 398
 399             fscal            = felec;
 400
 401             /* Update vectorial force */
 402             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 403             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 404             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 405
 406             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 407             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 408             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 409
 410             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 411
 412             /* Inner loop uses 169 flops */
 413         }
 414
 415         if(jidx<j_index_end)
 416         {
 417
 418             jnrA             = jjnr[jidx];
 419             j_coord_offsetA  = DIM*jnrA;
 420
 421             /* load j atom coordinates */
 422             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 423                                               &jx0,&jy0,&jz0);
 424
 425             /* Calculate displacement vector */
 426             dx00             = _mm_sub_pd(ix0,jx0);
 427             dy00             = _mm_sub_pd(iy0,jy0);
 428             dz00             = _mm_sub_pd(iz0,jz0);
 429             dx10             = _mm_sub_pd(ix1,jx0);
 430             dy10             = _mm_sub_pd(iy1,jy0);
 431             dz10             = _mm_sub_pd(iz1,jz0);
 432             dx20             = _mm_sub_pd(ix2,jx0);
 433             dy20             = _mm_sub_pd(iy2,jy0);
 434             dz20             = _mm_sub_pd(iz2,jz0);
 435
 436             /* Calculate squared distance and things based on it */
 437             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 438             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 439             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 440
 441             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 442             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 443             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 444
 445             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 446             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 447             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 448
 449             /* Load parameters for j particles */
 450             jq0              = _mm_load_sd(charge+jnrA+0);
 451             vdwjidx0A        = 2*vdwtype[jnrA+0];
 452
 453             fjx0             = _mm_setzero_pd();
 454             fjy0             = _mm_setzero_pd();
 455             fjz0             = _mm_setzero_pd();
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             r00              = _mm_mul_pd(rsq00,rinv00);
 462
 463             /* Compute parameters for interactions between i and j atoms */
 464             qq00             = _mm_mul_pd(iq0,jq0);
 465             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 466
 467             /* Calculate table index by multiplying r with table scale and truncate to integer */
 468             rt               = _mm_mul_pd(r00,vftabscale);
 469             vfitab           = _mm_cvttpd_epi32(rt);
 470 #ifdef __XOP__
 471             vfeps            = _mm_frcz_pd(rt);
 472 #else
 473             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 474 #endif
 475             twovfeps         = _mm_add_pd(vfeps,vfeps);
 476             vfitab           = _mm_slli_epi32(vfitab,3);
 477
 478             /* EWALD ELECTROSTATICS */
 479
 480             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 481             ewrt             = _mm_mul_pd(r00,ewtabscale);
 482             ewitab           = _mm_cvttpd_epi32(ewrt);
 483 #ifdef __XOP__
 484             eweps            = _mm_frcz_pd(ewrt);
 485 #else
 486             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 487 #endif
 488             twoeweps         = _mm_add_pd(eweps,eweps);
 489             ewitab           = _mm_slli_epi32(ewitab,2);
 490             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 491             ewtabD           = _mm_setzero_pd();
 492             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 493             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 494             ewtabFn          = _mm_setzero_pd();
 495             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 496             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 497             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 498             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 499             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 500
 501             /* CUBIC SPLINE TABLE DISPERSION */
 502             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 503             F                = _mm_setzero_pd();
 504             GMX_MM_TRANSPOSE2_PD(Y,F);
 505             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 506             H                = _mm_setzero_pd();
 507             GMX_MM_TRANSPOSE2_PD(G,H);
 508             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 509             VV               = _mm_macc_pd(vfeps,Fp,Y);
 510             vvdw6            = _mm_mul_pd(c6_00,VV);
 511             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 512             fvdw6            = _mm_mul_pd(c6_00,FF);
 513
 514             /* CUBIC SPLINE TABLE REPULSION */
 515             vfitab           = _mm_add_epi32(vfitab,ifour);
 516             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 517             F                = _mm_setzero_pd();
 518             GMX_MM_TRANSPOSE2_PD(Y,F);
 519             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 520             H                = _mm_setzero_pd();
 521             GMX_MM_TRANSPOSE2_PD(G,H);
 522             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 523             VV               = _mm_macc_pd(vfeps,Fp,Y);
 524             vvdw12           = _mm_mul_pd(c12_00,VV);
 525             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 526             fvdw12           = _mm_mul_pd(c12_00,FF);
 527             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 528             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 532             velecsum         = _mm_add_pd(velecsum,velec);
 533             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 534             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 535
 536             fscal            = _mm_add_pd(felec,fvdw);
 537
 538             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 539
 540             /* Update vectorial force */
 541             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 542             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 543             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 544
 545             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 546             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 547             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             r10              = _mm_mul_pd(rsq10,rinv10);
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             qq10             = _mm_mul_pd(iq1,jq0);
 557
 558             /* EWALD ELECTROSTATICS */
 559
 560             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 561             ewrt             = _mm_mul_pd(r10,ewtabscale);
 562             ewitab           = _mm_cvttpd_epi32(ewrt);
 563 #ifdef __XOP__
 564             eweps            = _mm_frcz_pd(ewrt);
 565 #else
 566             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 567 #endif
 568             twoeweps         = _mm_add_pd(eweps,eweps);
 569             ewitab           = _mm_slli_epi32(ewitab,2);
 570             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 571             ewtabD           = _mm_setzero_pd();
 572             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 573             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 574             ewtabFn          = _mm_setzero_pd();
 575             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 576             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 577             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 578             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 579             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 583             velecsum         = _mm_add_pd(velecsum,velec);
 584
 585             fscal            = felec;
 586
 587             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 588
 589             /* Update vectorial force */
 590             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 591             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 592             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 593
 594             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 595             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 596             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r20              = _mm_mul_pd(rsq20,rinv20);
 603
 604             /* Compute parameters for interactions between i and j atoms */
 605             qq20             = _mm_mul_pd(iq2,jq0);
 606
 607             /* EWALD ELECTROSTATICS */
 608
 609             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 610             ewrt             = _mm_mul_pd(r20,ewtabscale);
 611             ewitab           = _mm_cvttpd_epi32(ewrt);
 612 #ifdef __XOP__
 613             eweps            = _mm_frcz_pd(ewrt);
 614 #else
 615             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 616 #endif
 617             twoeweps         = _mm_add_pd(eweps,eweps);
 618             ewitab           = _mm_slli_epi32(ewitab,2);
 619             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 620             ewtabD           = _mm_setzero_pd();
 621             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 622             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 623             ewtabFn          = _mm_setzero_pd();
 624             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 625             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 626             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 627             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 628             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 629
 630             /* Update potential sum for this i atom from the interaction with this j atom. */
 631             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 632             velecsum         = _mm_add_pd(velecsum,velec);
 633
 634             fscal            = felec;
 635
 636             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 637
 638             /* Update vectorial force */
 639             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 640             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 641             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 642
 643             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 644             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 645             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 646
 647             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 648
 649             /* Inner loop uses 169 flops */
 650         }
 651
 652         /* End of innermost loop */
 653
 654         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 655                                               f+i_coord_offset,fshift+i_shift_offset);
 656
 657         ggid                        = gid[iidx];
 658         /* Update potential energies */
 659         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 660         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 661
 662         /* Increment number of inner iterations */
 663         inneriter                  += j_index_end - j_index_start;
 664
 665         /* Outer loop uses 20 flops */
 666     }
 667
 668     /* Increment number of outer iterations */
 669     outeriter        += nri;
 670
 671     /* Update outer/inner flops */
 672
 673     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*169);
 674 }
 675 /*
 676  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 677  * Electrostatics interaction: Ewald
 678  * VdW interaction:            CubicSplineTable
 679  * Geometry:                   Water3-Particle
 680  * Calculate force/pot:        Force
 681  */
 682 void
 683 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 684                     (t_nblist                    * gmx_restrict       nlist,
 685                      rvec                        * gmx_restrict          xx,
 686                      rvec                        * gmx_restrict          ff,
 687                      t_forcerec                  * gmx_restrict          fr,
 688                      t_mdatoms                   * gmx_restrict     mdatoms,
 689                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 690                      t_nrnb                      * gmx_restrict        nrnb)
 691 {
 692     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 693      * just 0 for non-waters.
 694      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 695      * jnr indices corresponding to data put in the four positions in the SIMD register.
 696      */
 697     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 698     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 699     int              jnrA,jnrB;
 700     int              j_coord_offsetA,j_coord_offsetB;
 701     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 702     real             rcutoff_scalar;
 703     real             *shiftvec,*fshift,*x,*f;
 704     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 705     int              vdwioffset0;
 706     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 707     int              vdwioffset1;
 708     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 709     int              vdwioffset2;
 710     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 711     int              vdwjidx0A,vdwjidx0B;
 712     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 713     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 714     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 715     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 716     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 717     real             *charge;
 718     int              nvdwtype;
 719     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 720     int              *vdwtype;
 721     real             *vdwparam;
 722     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 723     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 724     __m128i          vfitab;
 725     __m128i          ifour       = _mm_set1_epi32(4);
 726     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 727     real             *vftab;
 728     __m128i          ewitab;
 729     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 730     real             *ewtab;
 731     __m128d          dummy_mask,cutoff_mask;
 732     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 733     __m128d          one     = _mm_set1_pd(1.0);
 734     __m128d          two     = _mm_set1_pd(2.0);
 735     x                = xx[0];
 736     f                = ff[0];
 737
 738     nri              = nlist->nri;
 739     iinr             = nlist->iinr;
 740     jindex           = nlist->jindex;
 741     jjnr             = nlist->jjnr;
 742     shiftidx         = nlist->shift;
 743     gid              = nlist->gid;
 744     shiftvec         = fr->shift_vec[0];
 745     fshift           = fr->fshift[0];
 746     facel            = _mm_set1_pd(fr->epsfac);
 747     charge           = mdatoms->chargeA;
 748     nvdwtype         = fr->ntype;
 749     vdwparam         = fr->nbfp;
 750     vdwtype          = mdatoms->typeA;
 751
 752     vftab            = kernel_data->table_vdw->data;
 753     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 754
 755     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 756     ewtab            = fr->ic->tabq_coul_F;
 757     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 758     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 759
 760     /* Setup water-specific parameters */
 761     inr              = nlist->iinr[0];
 762     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 763     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 764     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 765     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 766
 767     /* Avoid stupid compiler warnings */
 768     jnrA = jnrB = 0;
 769     j_coord_offsetA = 0;
 770     j_coord_offsetB = 0;
 771
 772     outeriter        = 0;
 773     inneriter        = 0;
 774
 775     /* Start outer loop over neighborlists */
 776     for(iidx=0; iidx<nri; iidx++)
 777     {
 778         /* Load shift vector for this list */
 779         i_shift_offset   = DIM*shiftidx[iidx];
 780
 781         /* Load limits for loop over neighbors */
 782         j_index_start    = jindex[iidx];
 783         j_index_end      = jindex[iidx+1];
 784
 785         /* Get outer coordinate index */
 786         inr              = iinr[iidx];
 787         i_coord_offset   = DIM*inr;
 788
 789         /* Load i particle coords and add shift vector */
 790         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 791                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 792
 793         fix0             = _mm_setzero_pd();
 794         fiy0             = _mm_setzero_pd();
 795         fiz0             = _mm_setzero_pd();
 796         fix1             = _mm_setzero_pd();
 797         fiy1             = _mm_setzero_pd();
 798         fiz1             = _mm_setzero_pd();
 799         fix2             = _mm_setzero_pd();
 800         fiy2             = _mm_setzero_pd();
 801         fiz2             = _mm_setzero_pd();
 802
 803         /* Start inner kernel loop */
 804         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 805         {
 806
 807             /* Get j neighbor index, and coordinate index */
 808             jnrA             = jjnr[jidx];
 809             jnrB             = jjnr[jidx+1];
 810             j_coord_offsetA  = DIM*jnrA;
 811             j_coord_offsetB  = DIM*jnrB;
 812
 813             /* load j atom coordinates */
 814             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 815                                               &jx0,&jy0,&jz0);
 816
 817             /* Calculate displacement vector */
 818             dx00             = _mm_sub_pd(ix0,jx0);
 819             dy00             = _mm_sub_pd(iy0,jy0);
 820             dz00             = _mm_sub_pd(iz0,jz0);
 821             dx10             = _mm_sub_pd(ix1,jx0);
 822             dy10             = _mm_sub_pd(iy1,jy0);
 823             dz10             = _mm_sub_pd(iz1,jz0);
 824             dx20             = _mm_sub_pd(ix2,jx0);
 825             dy20             = _mm_sub_pd(iy2,jy0);
 826             dz20             = _mm_sub_pd(iz2,jz0);
 827
 828             /* Calculate squared distance and things based on it */
 829             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 830             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 831             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 832
 833             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 834             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 835             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 836
 837             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 838             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 839             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 840
 841             /* Load parameters for j particles */
 842             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 843             vdwjidx0A        = 2*vdwtype[jnrA+0];
 844             vdwjidx0B        = 2*vdwtype[jnrB+0];
 845
 846             fjx0             = _mm_setzero_pd();
 847             fjy0             = _mm_setzero_pd();
 848             fjz0             = _mm_setzero_pd();
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             r00              = _mm_mul_pd(rsq00,rinv00);
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             qq00             = _mm_mul_pd(iq0,jq0);
 858             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 859                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 860
 861             /* Calculate table index by multiplying r with table scale and truncate to integer */
 862             rt               = _mm_mul_pd(r00,vftabscale);
 863             vfitab           = _mm_cvttpd_epi32(rt);
 864 #ifdef __XOP__
 865             vfeps            = _mm_frcz_pd(rt);
 866 #else
 867             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 868 #endif
 869             twovfeps         = _mm_add_pd(vfeps,vfeps);
 870             vfitab           = _mm_slli_epi32(vfitab,3);
 871
 872             /* EWALD ELECTROSTATICS */
 873
 874             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 875             ewrt             = _mm_mul_pd(r00,ewtabscale);
 876             ewitab           = _mm_cvttpd_epi32(ewrt);
 877 #ifdef __XOP__
 878             eweps            = _mm_frcz_pd(ewrt);
 879 #else
 880             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 881 #endif
 882             twoeweps         = _mm_add_pd(eweps,eweps);
 883             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 884                                          &ewtabF,&ewtabFn);
 885             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 886             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 887
 888             /* CUBIC SPLINE TABLE DISPERSION */
 889             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 890             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 891             GMX_MM_TRANSPOSE2_PD(Y,F);
 892             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 893             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 894             GMX_MM_TRANSPOSE2_PD(G,H);
 895             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 896             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 897             fvdw6            = _mm_mul_pd(c6_00,FF);
 898
 899             /* CUBIC SPLINE TABLE REPULSION */
 900             vfitab           = _mm_add_epi32(vfitab,ifour);
 901             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 902             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 903             GMX_MM_TRANSPOSE2_PD(Y,F);
 904             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 905             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 906             GMX_MM_TRANSPOSE2_PD(G,H);
 907             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 908             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 909             fvdw12           = _mm_mul_pd(c12_00,FF);
 910             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 911
 912             fscal            = _mm_add_pd(felec,fvdw);
 913
 914             /* Update vectorial force */
 915             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 916             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 917             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 918
 919             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 920             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 921             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             r10              = _mm_mul_pd(rsq10,rinv10);
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             qq10             = _mm_mul_pd(iq1,jq0);
 931
 932             /* EWALD ELECTROSTATICS */
 933
 934             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 935             ewrt             = _mm_mul_pd(r10,ewtabscale);
 936             ewitab           = _mm_cvttpd_epi32(ewrt);
 937 #ifdef __XOP__
 938             eweps            = _mm_frcz_pd(ewrt);
 939 #else
 940             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 941 #endif
 942             twoeweps         = _mm_add_pd(eweps,eweps);
 943             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 944                                          &ewtabF,&ewtabFn);
 945             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 946             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 947
 948             fscal            = felec;
 949
 950             /* Update vectorial force */
 951             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 952             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 953             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 954
 955             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 956             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 957             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             r20              = _mm_mul_pd(rsq20,rinv20);
 964
 965             /* Compute parameters for interactions between i and j atoms */
 966             qq20             = _mm_mul_pd(iq2,jq0);
 967
 968             /* EWALD ELECTROSTATICS */
 969
 970             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 971             ewrt             = _mm_mul_pd(r20,ewtabscale);
 972             ewitab           = _mm_cvttpd_epi32(ewrt);
 973 #ifdef __XOP__
 974             eweps            = _mm_frcz_pd(ewrt);
 975 #else
 976             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 977 #endif
 978             twoeweps         = _mm_add_pd(eweps,eweps);
 979             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 980                                          &ewtabF,&ewtabFn);
 981             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 982             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 983
 984             fscal            = felec;
 985
 986             /* Update vectorial force */
 987             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 988             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 989             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 990
 991             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 992             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 993             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 994
 995             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 996
 997             /* Inner loop uses 146 flops */
 998         }
 999
1000         if(jidx<j_index_end)
1001         {
1002
1003             jnrA             = jjnr[jidx];
1004             j_coord_offsetA  = DIM*jnrA;
1005
1006             /* load j atom coordinates */
1007             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1008                                               &jx0,&jy0,&jz0);
1009
1010             /* Calculate displacement vector */
1011             dx00             = _mm_sub_pd(ix0,jx0);
1012             dy00             = _mm_sub_pd(iy0,jy0);
1013             dz00             = _mm_sub_pd(iz0,jz0);
1014             dx10             = _mm_sub_pd(ix1,jx0);
1015             dy10             = _mm_sub_pd(iy1,jy0);
1016             dz10             = _mm_sub_pd(iz1,jz0);
1017             dx20             = _mm_sub_pd(ix2,jx0);
1018             dy20             = _mm_sub_pd(iy2,jy0);
1019             dz20             = _mm_sub_pd(iz2,jz0);
1020
1021             /* Calculate squared distance and things based on it */
1022             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1023             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1024             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1025
1026             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1027             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1028             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1029
1030             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1031             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1032             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1033
1034             /* Load parameters for j particles */
1035             jq0              = _mm_load_sd(charge+jnrA+0);
1036             vdwjidx0A        = 2*vdwtype[jnrA+0];
1037
1038             fjx0             = _mm_setzero_pd();
1039             fjy0             = _mm_setzero_pd();
1040             fjz0             = _mm_setzero_pd();
1041
1042             /**************************
1043              * CALCULATE INTERACTIONS *
1044              **************************/
1045
1046             r00              = _mm_mul_pd(rsq00,rinv00);
1047
1048             /* Compute parameters for interactions between i and j atoms */
1049             qq00             = _mm_mul_pd(iq0,jq0);
1050             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1051
1052             /* Calculate table index by multiplying r with table scale and truncate to integer */
1053             rt               = _mm_mul_pd(r00,vftabscale);
1054             vfitab           = _mm_cvttpd_epi32(rt);
1055 #ifdef __XOP__
1056             vfeps            = _mm_frcz_pd(rt);
1057 #else
1058             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1059 #endif
1060             twovfeps         = _mm_add_pd(vfeps,vfeps);
1061             vfitab           = _mm_slli_epi32(vfitab,3);
1062
1063             /* EWALD ELECTROSTATICS */
1064
1065             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1066             ewrt             = _mm_mul_pd(r00,ewtabscale);
1067             ewitab           = _mm_cvttpd_epi32(ewrt);
1068 #ifdef __XOP__
1069             eweps            = _mm_frcz_pd(ewrt);
1070 #else
1071             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1072 #endif
1073             twoeweps         = _mm_add_pd(eweps,eweps);
1074             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1075             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1076             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1077
1078             /* CUBIC SPLINE TABLE DISPERSION */
1079             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1080             F                = _mm_setzero_pd();
1081             GMX_MM_TRANSPOSE2_PD(Y,F);
1082             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1083             H                = _mm_setzero_pd();
1084             GMX_MM_TRANSPOSE2_PD(G,H);
1085             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1086             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1087             fvdw6            = _mm_mul_pd(c6_00,FF);
1088
1089             /* CUBIC SPLINE TABLE REPULSION */
1090             vfitab           = _mm_add_epi32(vfitab,ifour);
1091             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1092             F                = _mm_setzero_pd();
1093             GMX_MM_TRANSPOSE2_PD(Y,F);
1094             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1095             H                = _mm_setzero_pd();
1096             GMX_MM_TRANSPOSE2_PD(G,H);
1097             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1098             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1099             fvdw12           = _mm_mul_pd(c12_00,FF);
1100             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1101
1102             fscal            = _mm_add_pd(felec,fvdw);
1103
1104             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1105
1106             /* Update vectorial force */
1107             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1108             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1109             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1110
1111             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1112             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1113             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1114
1115             /**************************
1116              * CALCULATE INTERACTIONS *
1117              **************************/
1118
1119             r10              = _mm_mul_pd(rsq10,rinv10);
1120
1121             /* Compute parameters for interactions between i and j atoms */
1122             qq10             = _mm_mul_pd(iq1,jq0);
1123
1124             /* EWALD ELECTROSTATICS */
1125
1126             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1127             ewrt             = _mm_mul_pd(r10,ewtabscale);
1128             ewitab           = _mm_cvttpd_epi32(ewrt);
1129 #ifdef __XOP__
1130             eweps            = _mm_frcz_pd(ewrt);
1131 #else
1132             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1133 #endif
1134             twoeweps         = _mm_add_pd(eweps,eweps);
1135             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1136             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1137             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1138
1139             fscal            = felec;
1140
1141             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1142
1143             /* Update vectorial force */
1144             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1145             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1146             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1147
1148             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1149             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1150             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1151
1152             /**************************
1153              * CALCULATE INTERACTIONS *
1154              **************************/
1155
1156             r20              = _mm_mul_pd(rsq20,rinv20);
1157
1158             /* Compute parameters for interactions between i and j atoms */
1159             qq20             = _mm_mul_pd(iq2,jq0);
1160
1161             /* EWALD ELECTROSTATICS */
1162
1163             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1164             ewrt             = _mm_mul_pd(r20,ewtabscale);
1165             ewitab           = _mm_cvttpd_epi32(ewrt);
1166 #ifdef __XOP__
1167             eweps            = _mm_frcz_pd(ewrt);
1168 #else
1169             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1170 #endif
1171             twoeweps         = _mm_add_pd(eweps,eweps);
1172             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1173             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1174             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1179
1180             /* Update vectorial force */
1181             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1182             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1183             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1184
1185             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1186             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1187             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1188
1189             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1190
1191             /* Inner loop uses 146 flops */
1192         }
1193
1194         /* End of innermost loop */
1195
1196         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1197                                               f+i_coord_offset,fshift+i_shift_offset);
1198
1199         /* Increment number of inner iterations */
1200         inneriter                  += j_index_end - j_index_start;
1201
1202         /* Outer loop uses 18 flops */
1203     }
1204
1205     /* Increment number of outer iterations */
1206     outeriter        += nri;
1207
1208     /* Update outer/inner flops */
1209
1210     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*146);
1211 }