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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 104     real             *vftab;
 105     __m128i          ewitab;
 106     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     real             *ewtab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 131
 132     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 133     ewtab            = fr->ic->tabq_coul_FDV0;
 134     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 135     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 140     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 141     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 142     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_pd();
 171         fiy0             = _mm_setzero_pd();
 172         fiz0             = _mm_setzero_pd();
 173         fix1             = _mm_setzero_pd();
 174         fiy1             = _mm_setzero_pd();
 175         fiz1             = _mm_setzero_pd();
 176         fix2             = _mm_setzero_pd();
 177         fiy2             = _mm_setzero_pd();
 178         fiz2             = _mm_setzero_pd();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_pd();
 182         vvdwsum          = _mm_setzero_pd();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_pd(ix0,jx0);
 200             dy00             = _mm_sub_pd(iy0,jy0);
 201             dz00             = _mm_sub_pd(iz0,jz0);
 202             dx10             = _mm_sub_pd(ix1,jx0);
 203             dy10             = _mm_sub_pd(iy1,jy0);
 204             dz10             = _mm_sub_pd(iz1,jz0);
 205             dx20             = _mm_sub_pd(ix2,jx0);
 206             dy20             = _mm_sub_pd(iy2,jy0);
 207             dz20             = _mm_sub_pd(iz2,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 213
 214             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 215             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 216             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 217
 218             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 219             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226
 227             fjx0             = _mm_setzero_pd();
 228             fjy0             = _mm_setzero_pd();
 229             fjz0             = _mm_setzero_pd();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             r00              = _mm_mul_pd(rsq00,rinv00);
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq00             = _mm_mul_pd(iq0,jq0);
 239             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 240                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 241
 242             /* Calculate table index by multiplying r with table scale and truncate to integer */
 243             rt               = _mm_mul_pd(r00,vftabscale);
 244             vfitab           = _mm_cvttpd_epi32(rt);
 245 #ifdef __XOP__
 246             vfeps            = _mm_frcz_pd(rt);
 247 #else
 248             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 249 #endif
 250             twovfeps         = _mm_add_pd(vfeps,vfeps);
 251             vfitab           = _mm_slli_epi32(vfitab,3);
 252
 253             /* EWALD ELECTROSTATICS */
 254
 255             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 256             ewrt             = _mm_mul_pd(r00,ewtabscale);
 257             ewitab           = _mm_cvttpd_epi32(ewrt);
 258 #ifdef __XOP__
 259             eweps            = _mm_frcz_pd(ewrt);
 260 #else
 261             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 262 #endif
 263             twoeweps         = _mm_add_pd(eweps,eweps);
 264             ewitab           = _mm_slli_epi32(ewitab,2);
 265             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 266             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 267             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 268             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 269             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 270             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 271             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 272             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 273             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 274             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 275
 276             /* CUBIC SPLINE TABLE DISPERSION */
 277             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 278             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 279             GMX_MM_TRANSPOSE2_PD(Y,F);
 280             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 281             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 282             GMX_MM_TRANSPOSE2_PD(G,H);
 283             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 284             VV               = _mm_macc_pd(vfeps,Fp,Y);
 285             vvdw6            = _mm_mul_pd(c6_00,VV);
 286             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 287             fvdw6            = _mm_mul_pd(c6_00,FF);
 288
 289             /* CUBIC SPLINE TABLE REPULSION */
 290             vfitab           = _mm_add_epi32(vfitab,ifour);
 291             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 292             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 293             GMX_MM_TRANSPOSE2_PD(Y,F);
 294             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 295             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 296             GMX_MM_TRANSPOSE2_PD(G,H);
 297             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 298             VV               = _mm_macc_pd(vfeps,Fp,Y);
 299             vvdw12           = _mm_mul_pd(c12_00,VV);
 300             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 301             fvdw12           = _mm_mul_pd(c12_00,FF);
 302             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 303             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velecsum         = _mm_add_pd(velecsum,velec);
 307             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 308
 309             fscal            = _mm_add_pd(felec,fvdw);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 313             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 314             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 315
 316             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 317             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 318             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r10              = _mm_mul_pd(rsq10,rinv10);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm_mul_pd(iq1,jq0);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _mm_mul_pd(r10,ewtabscale);
 333             ewitab           = _mm_cvttpd_epi32(ewrt);
 334 #ifdef __XOP__
 335             eweps            = _mm_frcz_pd(ewrt);
 336 #else
 337             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 338 #endif
 339             twoeweps         = _mm_add_pd(eweps,eweps);
 340             ewitab           = _mm_slli_epi32(ewitab,2);
 341             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 342             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 343             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 344             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 345             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 346             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 347             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 348             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 349             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 350             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velecsum         = _mm_add_pd(velecsum,velec);
 354
 355             fscal            = felec;
 356
 357             /* Update vectorial force */
 358             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 359             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 360             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 361
 362             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 363             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 364             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             r20              = _mm_mul_pd(rsq20,rinv20);
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq20             = _mm_mul_pd(iq2,jq0);
 374
 375             /* EWALD ELECTROSTATICS */
 376
 377             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 378             ewrt             = _mm_mul_pd(r20,ewtabscale);
 379             ewitab           = _mm_cvttpd_epi32(ewrt);
 380 #ifdef __XOP__
 381             eweps            = _mm_frcz_pd(ewrt);
 382 #else
 383             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 384 #endif
 385             twoeweps         = _mm_add_pd(eweps,eweps);
 386             ewitab           = _mm_slli_epi32(ewitab,2);
 387             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 388             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 389             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 390             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 391             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 392             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 393             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 394             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 395             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 396             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velecsum         = _mm_add_pd(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             /* Update vectorial force */
 404             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 405             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 406             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 407
 408             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 409             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 410             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 411
 412             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 413
 414             /* Inner loop uses 169 flops */
 415         }
 416
 417         if(jidx<j_index_end)
 418         {
 419
 420             jnrA             = jjnr[jidx];
 421             j_coord_offsetA  = DIM*jnrA;
 422
 423             /* load j atom coordinates */
 424             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 425                                               &jx0,&jy0,&jz0);
 426
 427             /* Calculate displacement vector */
 428             dx00             = _mm_sub_pd(ix0,jx0);
 429             dy00             = _mm_sub_pd(iy0,jy0);
 430             dz00             = _mm_sub_pd(iz0,jz0);
 431             dx10             = _mm_sub_pd(ix1,jx0);
 432             dy10             = _mm_sub_pd(iy1,jy0);
 433             dz10             = _mm_sub_pd(iz1,jz0);
 434             dx20             = _mm_sub_pd(ix2,jx0);
 435             dy20             = _mm_sub_pd(iy2,jy0);
 436             dz20             = _mm_sub_pd(iz2,jz0);
 437
 438             /* Calculate squared distance and things based on it */
 439             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 440             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 441             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 442
 443             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 444             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 445             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 446
 447             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 448             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 449             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 450
 451             /* Load parameters for j particles */
 452             jq0              = _mm_load_sd(charge+jnrA+0);
 453             vdwjidx0A        = 2*vdwtype[jnrA+0];
 454
 455             fjx0             = _mm_setzero_pd();
 456             fjy0             = _mm_setzero_pd();
 457             fjz0             = _mm_setzero_pd();
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             r00              = _mm_mul_pd(rsq00,rinv00);
 464
 465             /* Compute parameters for interactions between i and j atoms */
 466             qq00             = _mm_mul_pd(iq0,jq0);
 467             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 468
 469             /* Calculate table index by multiplying r with table scale and truncate to integer */
 470             rt               = _mm_mul_pd(r00,vftabscale);
 471             vfitab           = _mm_cvttpd_epi32(rt);
 472 #ifdef __XOP__
 473             vfeps            = _mm_frcz_pd(rt);
 474 #else
 475             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 476 #endif
 477             twovfeps         = _mm_add_pd(vfeps,vfeps);
 478             vfitab           = _mm_slli_epi32(vfitab,3);
 479
 480             /* EWALD ELECTROSTATICS */
 481
 482             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 483             ewrt             = _mm_mul_pd(r00,ewtabscale);
 484             ewitab           = _mm_cvttpd_epi32(ewrt);
 485 #ifdef __XOP__
 486             eweps            = _mm_frcz_pd(ewrt);
 487 #else
 488             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 489 #endif
 490             twoeweps         = _mm_add_pd(eweps,eweps);
 491             ewitab           = _mm_slli_epi32(ewitab,2);
 492             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 493             ewtabD           = _mm_setzero_pd();
 494             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 495             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 496             ewtabFn          = _mm_setzero_pd();
 497             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 498             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 499             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 500             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 501             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 502
 503             /* CUBIC SPLINE TABLE DISPERSION */
 504             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 505             F                = _mm_setzero_pd();
 506             GMX_MM_TRANSPOSE2_PD(Y,F);
 507             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 508             H                = _mm_setzero_pd();
 509             GMX_MM_TRANSPOSE2_PD(G,H);
 510             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 511             VV               = _mm_macc_pd(vfeps,Fp,Y);
 512             vvdw6            = _mm_mul_pd(c6_00,VV);
 513             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 514             fvdw6            = _mm_mul_pd(c6_00,FF);
 515
 516             /* CUBIC SPLINE TABLE REPULSION */
 517             vfitab           = _mm_add_epi32(vfitab,ifour);
 518             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 519             F                = _mm_setzero_pd();
 520             GMX_MM_TRANSPOSE2_PD(Y,F);
 521             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 522             H                = _mm_setzero_pd();
 523             GMX_MM_TRANSPOSE2_PD(G,H);
 524             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 525             VV               = _mm_macc_pd(vfeps,Fp,Y);
 526             vvdw12           = _mm_mul_pd(c12_00,VV);
 527             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 528             fvdw12           = _mm_mul_pd(c12_00,FF);
 529             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 530             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 534             velecsum         = _mm_add_pd(velecsum,velec);
 535             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 536             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 537
 538             fscal            = _mm_add_pd(felec,fvdw);
 539
 540             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 541
 542             /* Update vectorial force */
 543             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 544             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 545             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 546
 547             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 548             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 549             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             r10              = _mm_mul_pd(rsq10,rinv10);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq10             = _mm_mul_pd(iq1,jq0);
 559
 560             /* EWALD ELECTROSTATICS */
 561
 562             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 563             ewrt             = _mm_mul_pd(r10,ewtabscale);
 564             ewitab           = _mm_cvttpd_epi32(ewrt);
 565 #ifdef __XOP__
 566             eweps            = _mm_frcz_pd(ewrt);
 567 #else
 568             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 569 #endif
 570             twoeweps         = _mm_add_pd(eweps,eweps);
 571             ewitab           = _mm_slli_epi32(ewitab,2);
 572             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 573             ewtabD           = _mm_setzero_pd();
 574             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 575             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 576             ewtabFn          = _mm_setzero_pd();
 577             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 578             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 579             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 580             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 581             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 585             velecsum         = _mm_add_pd(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 590
 591             /* Update vectorial force */
 592             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 593             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 594             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 595
 596             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 597             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 598             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 599
 600             /**************************
 601              * CALCULATE INTERACTIONS *
 602              **************************/
 603
 604             r20              = _mm_mul_pd(rsq20,rinv20);
 605
 606             /* Compute parameters for interactions between i and j atoms */
 607             qq20             = _mm_mul_pd(iq2,jq0);
 608
 609             /* EWALD ELECTROSTATICS */
 610
 611             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 612             ewrt             = _mm_mul_pd(r20,ewtabscale);
 613             ewitab           = _mm_cvttpd_epi32(ewrt);
 614 #ifdef __XOP__
 615             eweps            = _mm_frcz_pd(ewrt);
 616 #else
 617             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 618 #endif
 619             twoeweps         = _mm_add_pd(eweps,eweps);
 620             ewitab           = _mm_slli_epi32(ewitab,2);
 621             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 622             ewtabD           = _mm_setzero_pd();
 623             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 624             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 625             ewtabFn          = _mm_setzero_pd();
 626             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 627             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 628             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 629             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 630             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 631
 632             /* Update potential sum for this i atom from the interaction with this j atom. */
 633             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 634             velecsum         = _mm_add_pd(velecsum,velec);
 635
 636             fscal            = felec;
 637
 638             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 639
 640             /* Update vectorial force */
 641             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 642             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 643             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 644
 645             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 646             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 647             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 648
 649             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 650
 651             /* Inner loop uses 169 flops */
 652         }
 653
 654         /* End of innermost loop */
 655
 656         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 657                                               f+i_coord_offset,fshift+i_shift_offset);
 658
 659         ggid                        = gid[iidx];
 660         /* Update potential energies */
 661         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 662         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 663
 664         /* Increment number of inner iterations */
 665         inneriter                  += j_index_end - j_index_start;
 666
 667         /* Outer loop uses 20 flops */
 668     }
 669
 670     /* Increment number of outer iterations */
 671     outeriter        += nri;
 672
 673     /* Update outer/inner flops */
 674
 675     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*169);
 676 }
 677 /*
 678  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 679  * Electrostatics interaction: Ewald
 680  * VdW interaction:            CubicSplineTable
 681  * Geometry:                   Water3-Particle
 682  * Calculate force/pot:        Force
 683  */
 684 void
 685 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 686                     (t_nblist                    * gmx_restrict       nlist,
 687                      rvec                        * gmx_restrict          xx,
 688                      rvec                        * gmx_restrict          ff,
 689                      t_forcerec                  * gmx_restrict          fr,
 690                      t_mdatoms                   * gmx_restrict     mdatoms,
 691                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 692                      t_nrnb                      * gmx_restrict        nrnb)
 693 {
 694     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 695      * just 0 for non-waters.
 696      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 697      * jnr indices corresponding to data put in the four positions in the SIMD register.
 698      */
 699     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 700     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 701     int              jnrA,jnrB;
 702     int              j_coord_offsetA,j_coord_offsetB;
 703     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 704     real             rcutoff_scalar;
 705     real             *shiftvec,*fshift,*x,*f;
 706     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 707     int              vdwioffset0;
 708     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 709     int              vdwioffset1;
 710     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 711     int              vdwioffset2;
 712     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 713     int              vdwjidx0A,vdwjidx0B;
 714     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 715     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 716     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 717     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 718     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 719     real             *charge;
 720     int              nvdwtype;
 721     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 722     int              *vdwtype;
 723     real             *vdwparam;
 724     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 725     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 726     __m128i          vfitab;
 727     __m128i          ifour       = _mm_set1_epi32(4);
 728     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 729     real             *vftab;
 730     __m128i          ewitab;
 731     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 732     real             *ewtab;
 733     __m128d          dummy_mask,cutoff_mask;
 734     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 735     __m128d          one     = _mm_set1_pd(1.0);
 736     __m128d          two     = _mm_set1_pd(2.0);
 737     x                = xx[0];
 738     f                = ff[0];
 739
 740     nri              = nlist->nri;
 741     iinr             = nlist->iinr;
 742     jindex           = nlist->jindex;
 743     jjnr             = nlist->jjnr;
 744     shiftidx         = nlist->shift;
 745     gid              = nlist->gid;
 746     shiftvec         = fr->shift_vec[0];
 747     fshift           = fr->fshift[0];
 748     facel            = _mm_set1_pd(fr->epsfac);
 749     charge           = mdatoms->chargeA;
 750     nvdwtype         = fr->ntype;
 751     vdwparam         = fr->nbfp;
 752     vdwtype          = mdatoms->typeA;
 753
 754     vftab            = kernel_data->table_vdw->data;
 755     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 756
 757     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 758     ewtab            = fr->ic->tabq_coul_F;
 759     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 760     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 761
 762     /* Setup water-specific parameters */
 763     inr              = nlist->iinr[0];
 764     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 765     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 766     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 767     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 768
 769     /* Avoid stupid compiler warnings */
 770     jnrA = jnrB = 0;
 771     j_coord_offsetA = 0;
 772     j_coord_offsetB = 0;
 773
 774     outeriter        = 0;
 775     inneriter        = 0;
 776
 777     /* Start outer loop over neighborlists */
 778     for(iidx=0; iidx<nri; iidx++)
 779     {
 780         /* Load shift vector for this list */
 781         i_shift_offset   = DIM*shiftidx[iidx];
 782
 783         /* Load limits for loop over neighbors */
 784         j_index_start    = jindex[iidx];
 785         j_index_end      = jindex[iidx+1];
 786
 787         /* Get outer coordinate index */
 788         inr              = iinr[iidx];
 789         i_coord_offset   = DIM*inr;
 790
 791         /* Load i particle coords and add shift vector */
 792         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 793                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 794
 795         fix0             = _mm_setzero_pd();
 796         fiy0             = _mm_setzero_pd();
 797         fiz0             = _mm_setzero_pd();
 798         fix1             = _mm_setzero_pd();
 799         fiy1             = _mm_setzero_pd();
 800         fiz1             = _mm_setzero_pd();
 801         fix2             = _mm_setzero_pd();
 802         fiy2             = _mm_setzero_pd();
 803         fiz2             = _mm_setzero_pd();
 804
 805         /* Start inner kernel loop */
 806         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 807         {
 808
 809             /* Get j neighbor index, and coordinate index */
 810             jnrA             = jjnr[jidx];
 811             jnrB             = jjnr[jidx+1];
 812             j_coord_offsetA  = DIM*jnrA;
 813             j_coord_offsetB  = DIM*jnrB;
 814
 815             /* load j atom coordinates */
 816             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 817                                               &jx0,&jy0,&jz0);
 818
 819             /* Calculate displacement vector */
 820             dx00             = _mm_sub_pd(ix0,jx0);
 821             dy00             = _mm_sub_pd(iy0,jy0);
 822             dz00             = _mm_sub_pd(iz0,jz0);
 823             dx10             = _mm_sub_pd(ix1,jx0);
 824             dy10             = _mm_sub_pd(iy1,jy0);
 825             dz10             = _mm_sub_pd(iz1,jz0);
 826             dx20             = _mm_sub_pd(ix2,jx0);
 827             dy20             = _mm_sub_pd(iy2,jy0);
 828             dz20             = _mm_sub_pd(iz2,jz0);
 829
 830             /* Calculate squared distance and things based on it */
 831             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 832             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 833             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 834
 835             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 836             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 837             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 838
 839             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 840             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 841             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 842
 843             /* Load parameters for j particles */
 844             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 845             vdwjidx0A        = 2*vdwtype[jnrA+0];
 846             vdwjidx0B        = 2*vdwtype[jnrB+0];
 847
 848             fjx0             = _mm_setzero_pd();
 849             fjy0             = _mm_setzero_pd();
 850             fjz0             = _mm_setzero_pd();
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             r00              = _mm_mul_pd(rsq00,rinv00);
 857
 858             /* Compute parameters for interactions between i and j atoms */
 859             qq00             = _mm_mul_pd(iq0,jq0);
 860             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 861                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 862
 863             /* Calculate table index by multiplying r with table scale and truncate to integer */
 864             rt               = _mm_mul_pd(r00,vftabscale);
 865             vfitab           = _mm_cvttpd_epi32(rt);
 866 #ifdef __XOP__
 867             vfeps            = _mm_frcz_pd(rt);
 868 #else
 869             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 870 #endif
 871             twovfeps         = _mm_add_pd(vfeps,vfeps);
 872             vfitab           = _mm_slli_epi32(vfitab,3);
 873
 874             /* EWALD ELECTROSTATICS */
 875
 876             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 877             ewrt             = _mm_mul_pd(r00,ewtabscale);
 878             ewitab           = _mm_cvttpd_epi32(ewrt);
 879 #ifdef __XOP__
 880             eweps            = _mm_frcz_pd(ewrt);
 881 #else
 882             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 883 #endif
 884             twoeweps         = _mm_add_pd(eweps,eweps);
 885             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 886                                          &ewtabF,&ewtabFn);
 887             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 888             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 889
 890             /* CUBIC SPLINE TABLE DISPERSION */
 891             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 892             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 893             GMX_MM_TRANSPOSE2_PD(Y,F);
 894             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 895             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 896             GMX_MM_TRANSPOSE2_PD(G,H);
 897             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 898             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 899             fvdw6            = _mm_mul_pd(c6_00,FF);
 900
 901             /* CUBIC SPLINE TABLE REPULSION */
 902             vfitab           = _mm_add_epi32(vfitab,ifour);
 903             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 904             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 905             GMX_MM_TRANSPOSE2_PD(Y,F);
 906             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 907             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 908             GMX_MM_TRANSPOSE2_PD(G,H);
 909             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 910             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 911             fvdw12           = _mm_mul_pd(c12_00,FF);
 912             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 913
 914             fscal            = _mm_add_pd(felec,fvdw);
 915
 916             /* Update vectorial force */
 917             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 918             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 919             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 920
 921             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 922             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 923             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 924
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             r10              = _mm_mul_pd(rsq10,rinv10);
 930
 931             /* Compute parameters for interactions between i and j atoms */
 932             qq10             = _mm_mul_pd(iq1,jq0);
 933
 934             /* EWALD ELECTROSTATICS */
 935
 936             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 937             ewrt             = _mm_mul_pd(r10,ewtabscale);
 938             ewitab           = _mm_cvttpd_epi32(ewrt);
 939 #ifdef __XOP__
 940             eweps            = _mm_frcz_pd(ewrt);
 941 #else
 942             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 943 #endif
 944             twoeweps         = _mm_add_pd(eweps,eweps);
 945             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 946                                          &ewtabF,&ewtabFn);
 947             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 948             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 949
 950             fscal            = felec;
 951
 952             /* Update vectorial force */
 953             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 954             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 955             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 956
 957             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 958             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 959             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             r20              = _mm_mul_pd(rsq20,rinv20);
 966
 967             /* Compute parameters for interactions between i and j atoms */
 968             qq20             = _mm_mul_pd(iq2,jq0);
 969
 970             /* EWALD ELECTROSTATICS */
 971
 972             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 973             ewrt             = _mm_mul_pd(r20,ewtabscale);
 974             ewitab           = _mm_cvttpd_epi32(ewrt);
 975 #ifdef __XOP__
 976             eweps            = _mm_frcz_pd(ewrt);
 977 #else
 978             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 979 #endif
 980             twoeweps         = _mm_add_pd(eweps,eweps);
 981             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 982                                          &ewtabF,&ewtabFn);
 983             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 984             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 985
 986             fscal            = felec;
 987
 988             /* Update vectorial force */
 989             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 990             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 991             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 992
 993             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 994             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 995             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 996
 997             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 998
 999             /* Inner loop uses 146 flops */
1000         }
1001
1002         if(jidx<j_index_end)
1003         {
1004
1005             jnrA             = jjnr[jidx];
1006             j_coord_offsetA  = DIM*jnrA;
1007
1008             /* load j atom coordinates */
1009             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1010                                               &jx0,&jy0,&jz0);
1011
1012             /* Calculate displacement vector */
1013             dx00             = _mm_sub_pd(ix0,jx0);
1014             dy00             = _mm_sub_pd(iy0,jy0);
1015             dz00             = _mm_sub_pd(iz0,jz0);
1016             dx10             = _mm_sub_pd(ix1,jx0);
1017             dy10             = _mm_sub_pd(iy1,jy0);
1018             dz10             = _mm_sub_pd(iz1,jz0);
1019             dx20             = _mm_sub_pd(ix2,jx0);
1020             dy20             = _mm_sub_pd(iy2,jy0);
1021             dz20             = _mm_sub_pd(iz2,jz0);
1022
1023             /* Calculate squared distance and things based on it */
1024             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1025             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1026             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1027
1028             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1029             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1030             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1031
1032             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1033             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1034             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1035
1036             /* Load parameters for j particles */
1037             jq0              = _mm_load_sd(charge+jnrA+0);
1038             vdwjidx0A        = 2*vdwtype[jnrA+0];
1039
1040             fjx0             = _mm_setzero_pd();
1041             fjy0             = _mm_setzero_pd();
1042             fjz0             = _mm_setzero_pd();
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             r00              = _mm_mul_pd(rsq00,rinv00);
1049
1050             /* Compute parameters for interactions between i and j atoms */
1051             qq00             = _mm_mul_pd(iq0,jq0);
1052             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1053
1054             /* Calculate table index by multiplying r with table scale and truncate to integer */
1055             rt               = _mm_mul_pd(r00,vftabscale);
1056             vfitab           = _mm_cvttpd_epi32(rt);
1057 #ifdef __XOP__
1058             vfeps            = _mm_frcz_pd(rt);
1059 #else
1060             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1061 #endif
1062             twovfeps         = _mm_add_pd(vfeps,vfeps);
1063             vfitab           = _mm_slli_epi32(vfitab,3);
1064
1065             /* EWALD ELECTROSTATICS */
1066
1067             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1068             ewrt             = _mm_mul_pd(r00,ewtabscale);
1069             ewitab           = _mm_cvttpd_epi32(ewrt);
1070 #ifdef __XOP__
1071             eweps            = _mm_frcz_pd(ewrt);
1072 #else
1073             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1074 #endif
1075             twoeweps         = _mm_add_pd(eweps,eweps);
1076             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1077             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1078             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1079
1080             /* CUBIC SPLINE TABLE DISPERSION */
1081             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1082             F                = _mm_setzero_pd();
1083             GMX_MM_TRANSPOSE2_PD(Y,F);
1084             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1085             H                = _mm_setzero_pd();
1086             GMX_MM_TRANSPOSE2_PD(G,H);
1087             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1088             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1089             fvdw6            = _mm_mul_pd(c6_00,FF);
1090
1091             /* CUBIC SPLINE TABLE REPULSION */
1092             vfitab           = _mm_add_epi32(vfitab,ifour);
1093             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1094             F                = _mm_setzero_pd();
1095             GMX_MM_TRANSPOSE2_PD(Y,F);
1096             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1097             H                = _mm_setzero_pd();
1098             GMX_MM_TRANSPOSE2_PD(G,H);
1099             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1100             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1101             fvdw12           = _mm_mul_pd(c12_00,FF);
1102             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1103
1104             fscal            = _mm_add_pd(felec,fvdw);
1105
1106             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1107
1108             /* Update vectorial force */
1109             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1110             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1111             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1112
1113             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1114             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1115             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1116
1117             /**************************
1118              * CALCULATE INTERACTIONS *
1119              **************************/
1120
1121             r10              = _mm_mul_pd(rsq10,rinv10);
1122
1123             /* Compute parameters for interactions between i and j atoms */
1124             qq10             = _mm_mul_pd(iq1,jq0);
1125
1126             /* EWALD ELECTROSTATICS */
1127
1128             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1129             ewrt             = _mm_mul_pd(r10,ewtabscale);
1130             ewitab           = _mm_cvttpd_epi32(ewrt);
1131 #ifdef __XOP__
1132             eweps            = _mm_frcz_pd(ewrt);
1133 #else
1134             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1135 #endif
1136             twoeweps         = _mm_add_pd(eweps,eweps);
1137             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1138             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1139             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1140
1141             fscal            = felec;
1142
1143             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1144
1145             /* Update vectorial force */
1146             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1147             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1148             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1149
1150             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1151             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1152             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             r20              = _mm_mul_pd(rsq20,rinv20);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq20             = _mm_mul_pd(iq2,jq0);
1162
1163             /* EWALD ELECTROSTATICS */
1164
1165             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1166             ewrt             = _mm_mul_pd(r20,ewtabscale);
1167             ewitab           = _mm_cvttpd_epi32(ewrt);
1168 #ifdef __XOP__
1169             eweps            = _mm_frcz_pd(ewrt);
1170 #else
1171             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1172 #endif
1173             twoeweps         = _mm_add_pd(eweps,eweps);
1174             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1175             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1176             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1177
1178             fscal            = felec;
1179
1180             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1181
1182             /* Update vectorial force */
1183             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1184             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1185             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1186
1187             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1188             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1189             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1190
1191             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1192
1193             /* Inner loop uses 146 flops */
1194         }
1195
1196         /* End of innermost loop */
1197
1198         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1199                                               f+i_coord_offset,fshift+i_shift_offset);
1200
1201         /* Increment number of inner iterations */
1202         inneriter                  += j_index_end - j_index_start;
1203
1204         /* Outer loop uses 18 flops */
1205     }
1206
1207     /* Increment number of outer iterations */
1208     outeriter        += nri;
1209
1210     /* Update outer/inner flops */
1211
1212     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*146);
1213 }