src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_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_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_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     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_elec_vdw->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 161
 162         fix0             = _mm_setzero_pd();
 163         fiy0             = _mm_setzero_pd();
 164         fiz0             = _mm_setzero_pd();
 165         fix1             = _mm_setzero_pd();
 166         fiy1             = _mm_setzero_pd();
 167         fiz1             = _mm_setzero_pd();
 168         fix2             = _mm_setzero_pd();
 169         fiy2             = _mm_setzero_pd();
 170         fiz2             = _mm_setzero_pd();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_pd();
 174         vvdwsum          = _mm_setzero_pd();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_pd(ix0,jx0);
 192             dy00             = _mm_sub_pd(iy0,jy0);
 193             dz00             = _mm_sub_pd(iz0,jz0);
 194             dx10             = _mm_sub_pd(ix1,jx0);
 195             dy10             = _mm_sub_pd(iy1,jy0);
 196             dz10             = _mm_sub_pd(iz1,jz0);
 197             dx20             = _mm_sub_pd(ix2,jx0);
 198             dy20             = _mm_sub_pd(iy2,jy0);
 199             dz20             = _mm_sub_pd(iz2,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 203             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 204             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 205
 206             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 207             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 208             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 209
 210             /* Load parameters for j particles */
 211             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214
 215             fjx0             = _mm_setzero_pd();
 216             fjy0             = _mm_setzero_pd();
 217             fjz0             = _mm_setzero_pd();
 218
 219             /**************************
 220              * CALCULATE INTERACTIONS *
 221              **************************/
 222
 223             r00              = _mm_mul_pd(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_pd(iq0,jq0);
 227             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 228                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 229
 230             /* Calculate table index by multiplying r with table scale and truncate to integer */
 231             rt               = _mm_mul_pd(r00,vftabscale);
 232             vfitab           = _mm_cvttpd_epi32(rt);
 233 #ifdef __XOP__
 234             vfeps            = _mm_frcz_pd(rt);
 235 #else
 236             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 237 #endif
 238             twovfeps         = _mm_add_pd(vfeps,vfeps);
 239             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 240
 241             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 242             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 243             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 244             GMX_MM_TRANSPOSE2_PD(Y,F);
 245             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 246             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 247             GMX_MM_TRANSPOSE2_PD(G,H);
 248             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 249             VV               = _mm_macc_pd(vfeps,Fp,Y);
 250             velec            = _mm_mul_pd(qq00,VV);
 251             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 252             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 253
 254             /* CUBIC SPLINE TABLE DISPERSION */
 255             vfitab           = _mm_add_epi32(vfitab,ifour);
 256             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 257             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 258             GMX_MM_TRANSPOSE2_PD(Y,F);
 259             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 260             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 261             GMX_MM_TRANSPOSE2_PD(G,H);
 262             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 263             VV               = _mm_macc_pd(vfeps,Fp,Y);
 264             vvdw6            = _mm_mul_pd(c6_00,VV);
 265             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 266             fvdw6            = _mm_mul_pd(c6_00,FF);
 267
 268             /* CUBIC SPLINE TABLE REPULSION */
 269             vfitab           = _mm_add_epi32(vfitab,ifour);
 270             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 271             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 272             GMX_MM_TRANSPOSE2_PD(Y,F);
 273             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 274             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 275             GMX_MM_TRANSPOSE2_PD(G,H);
 276             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 277             VV               = _mm_macc_pd(vfeps,Fp,Y);
 278             vvdw12           = _mm_mul_pd(c12_00,VV);
 279             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 280             fvdw12           = _mm_mul_pd(c12_00,FF);
 281             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 282             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velecsum         = _mm_add_pd(velecsum,velec);
 286             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 287
 288             fscal            = _mm_add_pd(felec,fvdw);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 292             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 293             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 294
 295             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 296             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 297             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r10              = _mm_mul_pd(rsq10,rinv10);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_pd(iq1,jq0);
 307
 308             /* Calculate table index by multiplying r with table scale and truncate to integer */
 309             rt               = _mm_mul_pd(r10,vftabscale);
 310             vfitab           = _mm_cvttpd_epi32(rt);
 311 #ifdef __XOP__
 312             vfeps            = _mm_frcz_pd(rt);
 313 #else
 314             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 315 #endif
 316             twovfeps         = _mm_add_pd(vfeps,vfeps);
 317             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 318
 319             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 320             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 321             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 322             GMX_MM_TRANSPOSE2_PD(Y,F);
 323             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 324             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 325             GMX_MM_TRANSPOSE2_PD(G,H);
 326             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 327             VV               = _mm_macc_pd(vfeps,Fp,Y);
 328             velec            = _mm_mul_pd(qq10,VV);
 329             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 330             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velecsum         = _mm_add_pd(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             /* Update vectorial force */
 338             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 339             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 340             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 341
 342             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 343             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 344             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             r20              = _mm_mul_pd(rsq20,rinv20);
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq20             = _mm_mul_pd(iq2,jq0);
 354
 355             /* Calculate table index by multiplying r with table scale and truncate to integer */
 356             rt               = _mm_mul_pd(r20,vftabscale);
 357             vfitab           = _mm_cvttpd_epi32(rt);
 358 #ifdef __XOP__
 359             vfeps            = _mm_frcz_pd(rt);
 360 #else
 361             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 362 #endif
 363             twovfeps         = _mm_add_pd(vfeps,vfeps);
 364             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 365
 366             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 367             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 368             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 369             GMX_MM_TRANSPOSE2_PD(Y,F);
 370             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 371             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 372             GMX_MM_TRANSPOSE2_PD(G,H);
 373             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 374             VV               = _mm_macc_pd(vfeps,Fp,Y);
 375             velec            = _mm_mul_pd(qq20,VV);
 376             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 377             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm_add_pd(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             /* Update vectorial force */
 385             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 386             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 387             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 388
 389             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 390             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 391             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 392
 393             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 394
 395             /* Inner loop uses 171 flops */
 396         }
 397
 398         if(jidx<j_index_end)
 399         {
 400
 401             jnrA             = jjnr[jidx];
 402             j_coord_offsetA  = DIM*jnrA;
 403
 404             /* load j atom coordinates */
 405             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 406                                               &jx0,&jy0,&jz0);
 407
 408             /* Calculate displacement vector */
 409             dx00             = _mm_sub_pd(ix0,jx0);
 410             dy00             = _mm_sub_pd(iy0,jy0);
 411             dz00             = _mm_sub_pd(iz0,jz0);
 412             dx10             = _mm_sub_pd(ix1,jx0);
 413             dy10             = _mm_sub_pd(iy1,jy0);
 414             dz10             = _mm_sub_pd(iz1,jz0);
 415             dx20             = _mm_sub_pd(ix2,jx0);
 416             dy20             = _mm_sub_pd(iy2,jy0);
 417             dz20             = _mm_sub_pd(iz2,jz0);
 418
 419             /* Calculate squared distance and things based on it */
 420             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 421             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 422             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 423
 424             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 425             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 426             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 427
 428             /* Load parameters for j particles */
 429             jq0              = _mm_load_sd(charge+jnrA+0);
 430             vdwjidx0A        = 2*vdwtype[jnrA+0];
 431
 432             fjx0             = _mm_setzero_pd();
 433             fjy0             = _mm_setzero_pd();
 434             fjz0             = _mm_setzero_pd();
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             r00              = _mm_mul_pd(rsq00,rinv00);
 441
 442             /* Compute parameters for interactions between i and j atoms */
 443             qq00             = _mm_mul_pd(iq0,jq0);
 444             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 445
 446             /* Calculate table index by multiplying r with table scale and truncate to integer */
 447             rt               = _mm_mul_pd(r00,vftabscale);
 448             vfitab           = _mm_cvttpd_epi32(rt);
 449 #ifdef __XOP__
 450             vfeps            = _mm_frcz_pd(rt);
 451 #else
 452             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 453 #endif
 454             twovfeps         = _mm_add_pd(vfeps,vfeps);
 455             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 456
 457             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 458             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 459             F                = _mm_setzero_pd();
 460             GMX_MM_TRANSPOSE2_PD(Y,F);
 461             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 462             H                = _mm_setzero_pd();
 463             GMX_MM_TRANSPOSE2_PD(G,H);
 464             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 465             VV               = _mm_macc_pd(vfeps,Fp,Y);
 466             velec            = _mm_mul_pd(qq00,VV);
 467             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 468             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 469
 470             /* CUBIC SPLINE TABLE DISPERSION */
 471             vfitab           = _mm_add_epi32(vfitab,ifour);
 472             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 473             F                = _mm_setzero_pd();
 474             GMX_MM_TRANSPOSE2_PD(Y,F);
 475             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 476             H                = _mm_setzero_pd();
 477             GMX_MM_TRANSPOSE2_PD(G,H);
 478             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 479             VV               = _mm_macc_pd(vfeps,Fp,Y);
 480             vvdw6            = _mm_mul_pd(c6_00,VV);
 481             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 482             fvdw6            = _mm_mul_pd(c6_00,FF);
 483
 484             /* CUBIC SPLINE TABLE REPULSION */
 485             vfitab           = _mm_add_epi32(vfitab,ifour);
 486             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 487             F                = _mm_setzero_pd();
 488             GMX_MM_TRANSPOSE2_PD(Y,F);
 489             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 490             H                = _mm_setzero_pd();
 491             GMX_MM_TRANSPOSE2_PD(G,H);
 492             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 493             VV               = _mm_macc_pd(vfeps,Fp,Y);
 494             vvdw12           = _mm_mul_pd(c12_00,VV);
 495             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 496             fvdw12           = _mm_mul_pd(c12_00,FF);
 497             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 498             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 502             velecsum         = _mm_add_pd(velecsum,velec);
 503             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 504             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 505
 506             fscal            = _mm_add_pd(felec,fvdw);
 507
 508             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 509
 510             /* Update vectorial force */
 511             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 512             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 513             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 514
 515             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 516             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 517             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             r10              = _mm_mul_pd(rsq10,rinv10);
 524
 525             /* Compute parameters for interactions between i and j atoms */
 526             qq10             = _mm_mul_pd(iq1,jq0);
 527
 528             /* Calculate table index by multiplying r with table scale and truncate to integer */
 529             rt               = _mm_mul_pd(r10,vftabscale);
 530             vfitab           = _mm_cvttpd_epi32(rt);
 531 #ifdef __XOP__
 532             vfeps            = _mm_frcz_pd(rt);
 533 #else
 534             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 535 #endif
 536             twovfeps         = _mm_add_pd(vfeps,vfeps);
 537             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 538
 539             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 540             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 541             F                = _mm_setzero_pd();
 542             GMX_MM_TRANSPOSE2_PD(Y,F);
 543             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 544             H                = _mm_setzero_pd();
 545             GMX_MM_TRANSPOSE2_PD(G,H);
 546             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 547             VV               = _mm_macc_pd(vfeps,Fp,Y);
 548             velec            = _mm_mul_pd(qq10,VV);
 549             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 550             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 554             velecsum         = _mm_add_pd(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 559
 560             /* Update vectorial force */
 561             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 562             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 563             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 564
 565             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 566             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 567             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             r20              = _mm_mul_pd(rsq20,rinv20);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq20             = _mm_mul_pd(iq2,jq0);
 577
 578             /* Calculate table index by multiplying r with table scale and truncate to integer */
 579             rt               = _mm_mul_pd(r20,vftabscale);
 580             vfitab           = _mm_cvttpd_epi32(rt);
 581 #ifdef __XOP__
 582             vfeps            = _mm_frcz_pd(rt);
 583 #else
 584             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 585 #endif
 586             twovfeps         = _mm_add_pd(vfeps,vfeps);
 587             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 588
 589             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 590             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 591             F                = _mm_setzero_pd();
 592             GMX_MM_TRANSPOSE2_PD(Y,F);
 593             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 594             H                = _mm_setzero_pd();
 595             GMX_MM_TRANSPOSE2_PD(G,H);
 596             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 597             VV               = _mm_macc_pd(vfeps,Fp,Y);
 598             velec            = _mm_mul_pd(qq20,VV);
 599             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 600             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 604             velecsum         = _mm_add_pd(velecsum,velec);
 605
 606             fscal            = felec;
 607
 608             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 609
 610             /* Update vectorial force */
 611             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 612             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 613             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 614
 615             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 616             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 617             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 618
 619             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 620
 621             /* Inner loop uses 171 flops */
 622         }
 623
 624         /* End of innermost loop */
 625
 626         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 627                                               f+i_coord_offset,fshift+i_shift_offset);
 628
 629         ggid                        = gid[iidx];
 630         /* Update potential energies */
 631         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 632         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 633
 634         /* Increment number of inner iterations */
 635         inneriter                  += j_index_end - j_index_start;
 636
 637         /* Outer loop uses 20 flops */
 638     }
 639
 640     /* Increment number of outer iterations */
 641     outeriter        += nri;
 642
 643     /* Update outer/inner flops */
 644
 645     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*171);
 646 }
 647 /*
 648  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 649  * Electrostatics interaction: CubicSplineTable
 650  * VdW interaction:            CubicSplineTable
 651  * Geometry:                   Water3-Particle
 652  * Calculate force/pot:        Force
 653  */
 654 void
 655 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 656                     (t_nblist                    * gmx_restrict       nlist,
 657                      rvec                        * gmx_restrict          xx,
 658                      rvec                        * gmx_restrict          ff,
 659                      t_forcerec                  * gmx_restrict          fr,
 660                      t_mdatoms                   * gmx_restrict     mdatoms,
 661                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 662                      t_nrnb                      * gmx_restrict        nrnb)
 663 {
 664     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 665      * just 0 for non-waters.
 666      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 667      * jnr indices corresponding to data put in the four positions in the SIMD register.
 668      */
 669     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 670     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 671     int              jnrA,jnrB;
 672     int              j_coord_offsetA,j_coord_offsetB;
 673     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 674     real             rcutoff_scalar;
 675     real             *shiftvec,*fshift,*x,*f;
 676     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 677     int              vdwioffset0;
 678     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 679     int              vdwioffset1;
 680     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 681     int              vdwioffset2;
 682     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 683     int              vdwjidx0A,vdwjidx0B;
 684     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 685     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 686     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 687     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 688     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 689     real             *charge;
 690     int              nvdwtype;
 691     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 692     int              *vdwtype;
 693     real             *vdwparam;
 694     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 695     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 696     __m128i          vfitab;
 697     __m128i          ifour       = _mm_set1_epi32(4);
 698     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 699     real             *vftab;
 700     __m128d          dummy_mask,cutoff_mask;
 701     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 702     __m128d          one     = _mm_set1_pd(1.0);
 703     __m128d          two     = _mm_set1_pd(2.0);
 704     x                = xx[0];
 705     f                = ff[0];
 706
 707     nri              = nlist->nri;
 708     iinr             = nlist->iinr;
 709     jindex           = nlist->jindex;
 710     jjnr             = nlist->jjnr;
 711     shiftidx         = nlist->shift;
 712     gid              = nlist->gid;
 713     shiftvec         = fr->shift_vec[0];
 714     fshift           = fr->fshift[0];
 715     facel            = _mm_set1_pd(fr->epsfac);
 716     charge           = mdatoms->chargeA;
 717     nvdwtype         = fr->ntype;
 718     vdwparam         = fr->nbfp;
 719     vdwtype          = mdatoms->typeA;
 720
 721     vftab            = kernel_data->table_elec_vdw->data;
 722     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 723
 724     /* Setup water-specific parameters */
 725     inr              = nlist->iinr[0];
 726     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 727     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 728     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 729     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 730
 731     /* Avoid stupid compiler warnings */
 732     jnrA = jnrB = 0;
 733     j_coord_offsetA = 0;
 734     j_coord_offsetB = 0;
 735
 736     outeriter        = 0;
 737     inneriter        = 0;
 738
 739     /* Start outer loop over neighborlists */
 740     for(iidx=0; iidx<nri; iidx++)
 741     {
 742         /* Load shift vector for this list */
 743         i_shift_offset   = DIM*shiftidx[iidx];
 744
 745         /* Load limits for loop over neighbors */
 746         j_index_start    = jindex[iidx];
 747         j_index_end      = jindex[iidx+1];
 748
 749         /* Get outer coordinate index */
 750         inr              = iinr[iidx];
 751         i_coord_offset   = DIM*inr;
 752
 753         /* Load i particle coords and add shift vector */
 754         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 755                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 756
 757         fix0             = _mm_setzero_pd();
 758         fiy0             = _mm_setzero_pd();
 759         fiz0             = _mm_setzero_pd();
 760         fix1             = _mm_setzero_pd();
 761         fiy1             = _mm_setzero_pd();
 762         fiz1             = _mm_setzero_pd();
 763         fix2             = _mm_setzero_pd();
 764         fiy2             = _mm_setzero_pd();
 765         fiz2             = _mm_setzero_pd();
 766
 767         /* Start inner kernel loop */
 768         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 769         {
 770
 771             /* Get j neighbor index, and coordinate index */
 772             jnrA             = jjnr[jidx];
 773             jnrB             = jjnr[jidx+1];
 774             j_coord_offsetA  = DIM*jnrA;
 775             j_coord_offsetB  = DIM*jnrB;
 776
 777             /* load j atom coordinates */
 778             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 779                                               &jx0,&jy0,&jz0);
 780
 781             /* Calculate displacement vector */
 782             dx00             = _mm_sub_pd(ix0,jx0);
 783             dy00             = _mm_sub_pd(iy0,jy0);
 784             dz00             = _mm_sub_pd(iz0,jz0);
 785             dx10             = _mm_sub_pd(ix1,jx0);
 786             dy10             = _mm_sub_pd(iy1,jy0);
 787             dz10             = _mm_sub_pd(iz1,jz0);
 788             dx20             = _mm_sub_pd(ix2,jx0);
 789             dy20             = _mm_sub_pd(iy2,jy0);
 790             dz20             = _mm_sub_pd(iz2,jz0);
 791
 792             /* Calculate squared distance and things based on it */
 793             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 794             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 795             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 796
 797             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 798             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 799             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 800
 801             /* Load parameters for j particles */
 802             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 803             vdwjidx0A        = 2*vdwtype[jnrA+0];
 804             vdwjidx0B        = 2*vdwtype[jnrB+0];
 805
 806             fjx0             = _mm_setzero_pd();
 807             fjy0             = _mm_setzero_pd();
 808             fjz0             = _mm_setzero_pd();
 809
 810             /**************************
 811              * CALCULATE INTERACTIONS *
 812              **************************/
 813
 814             r00              = _mm_mul_pd(rsq00,rinv00);
 815
 816             /* Compute parameters for interactions between i and j atoms */
 817             qq00             = _mm_mul_pd(iq0,jq0);
 818             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 819                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 820
 821             /* Calculate table index by multiplying r with table scale and truncate to integer */
 822             rt               = _mm_mul_pd(r00,vftabscale);
 823             vfitab           = _mm_cvttpd_epi32(rt);
 824 #ifdef __XOP__
 825             vfeps            = _mm_frcz_pd(rt);
 826 #else
 827             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 828 #endif
 829             twovfeps         = _mm_add_pd(vfeps,vfeps);
 830             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 831
 832             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 833             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 834             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 835             GMX_MM_TRANSPOSE2_PD(Y,F);
 836             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 837             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 838             GMX_MM_TRANSPOSE2_PD(G,H);
 839             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 840             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 841             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 842
 843             /* CUBIC SPLINE TABLE DISPERSION */
 844             vfitab           = _mm_add_epi32(vfitab,ifour);
 845             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 846             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 847             GMX_MM_TRANSPOSE2_PD(Y,F);
 848             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 849             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 850             GMX_MM_TRANSPOSE2_PD(G,H);
 851             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 852             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 853             fvdw6            = _mm_mul_pd(c6_00,FF);
 854
 855             /* CUBIC SPLINE TABLE REPULSION */
 856             vfitab           = _mm_add_epi32(vfitab,ifour);
 857             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 858             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 859             GMX_MM_TRANSPOSE2_PD(Y,F);
 860             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 861             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 862             GMX_MM_TRANSPOSE2_PD(G,H);
 863             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 864             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 865             fvdw12           = _mm_mul_pd(c12_00,FF);
 866             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 867
 868             fscal            = _mm_add_pd(felec,fvdw);
 869
 870             /* Update vectorial force */
 871             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 872             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 873             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 874
 875             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 876             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 877             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             r10              = _mm_mul_pd(rsq10,rinv10);
 884
 885             /* Compute parameters for interactions between i and j atoms */
 886             qq10             = _mm_mul_pd(iq1,jq0);
 887
 888             /* Calculate table index by multiplying r with table scale and truncate to integer */
 889             rt               = _mm_mul_pd(r10,vftabscale);
 890             vfitab           = _mm_cvttpd_epi32(rt);
 891 #ifdef __XOP__
 892             vfeps            = _mm_frcz_pd(rt);
 893 #else
 894             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 895 #endif
 896             twovfeps         = _mm_add_pd(vfeps,vfeps);
 897             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 898
 899             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 900             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 901             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 902             GMX_MM_TRANSPOSE2_PD(Y,F);
 903             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 904             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 905             GMX_MM_TRANSPOSE2_PD(G,H);
 906             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 907             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 908             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 909
 910             fscal            = felec;
 911
 912             /* Update vectorial force */
 913             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 914             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 915             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 916
 917             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 918             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 919             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             r20              = _mm_mul_pd(rsq20,rinv20);
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             qq20             = _mm_mul_pd(iq2,jq0);
 929
 930             /* Calculate table index by multiplying r with table scale and truncate to integer */
 931             rt               = _mm_mul_pd(r20,vftabscale);
 932             vfitab           = _mm_cvttpd_epi32(rt);
 933 #ifdef __XOP__
 934             vfeps            = _mm_frcz_pd(rt);
 935 #else
 936             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 937 #endif
 938             twovfeps         = _mm_add_pd(vfeps,vfeps);
 939             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 940
 941             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 942             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 943             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 944             GMX_MM_TRANSPOSE2_PD(Y,F);
 945             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 946             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 947             GMX_MM_TRANSPOSE2_PD(G,H);
 948             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 949             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 950             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 951
 952             fscal            = felec;
 953
 954             /* Update vectorial force */
 955             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 956             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 957             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 958
 959             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 960             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 961             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 962
 963             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 964
 965             /* Inner loop uses 151 flops */
 966         }
 967
 968         if(jidx<j_index_end)
 969         {
 970
 971             jnrA             = jjnr[jidx];
 972             j_coord_offsetA  = DIM*jnrA;
 973
 974             /* load j atom coordinates */
 975             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 976                                               &jx0,&jy0,&jz0);
 977
 978             /* Calculate displacement vector */
 979             dx00             = _mm_sub_pd(ix0,jx0);
 980             dy00             = _mm_sub_pd(iy0,jy0);
 981             dz00             = _mm_sub_pd(iz0,jz0);
 982             dx10             = _mm_sub_pd(ix1,jx0);
 983             dy10             = _mm_sub_pd(iy1,jy0);
 984             dz10             = _mm_sub_pd(iz1,jz0);
 985             dx20             = _mm_sub_pd(ix2,jx0);
 986             dy20             = _mm_sub_pd(iy2,jy0);
 987             dz20             = _mm_sub_pd(iz2,jz0);
 988
 989             /* Calculate squared distance and things based on it */
 990             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 991             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 992             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 993
 994             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 995             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 996             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 997
 998             /* Load parameters for j particles */
 999             jq0              = _mm_load_sd(charge+jnrA+0);
1000             vdwjidx0A        = 2*vdwtype[jnrA+0];
1001
1002             fjx0             = _mm_setzero_pd();
1003             fjy0             = _mm_setzero_pd();
1004             fjz0             = _mm_setzero_pd();
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             r00              = _mm_mul_pd(rsq00,rinv00);
1011
1012             /* Compute parameters for interactions between i and j atoms */
1013             qq00             = _mm_mul_pd(iq0,jq0);
1014             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1015
1016             /* Calculate table index by multiplying r with table scale and truncate to integer */
1017             rt               = _mm_mul_pd(r00,vftabscale);
1018             vfitab           = _mm_cvttpd_epi32(rt);
1019 #ifdef __XOP__
1020             vfeps            = _mm_frcz_pd(rt);
1021 #else
1022             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1023 #endif
1024             twovfeps         = _mm_add_pd(vfeps,vfeps);
1025             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1026
1027             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1028             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1029             F                = _mm_setzero_pd();
1030             GMX_MM_TRANSPOSE2_PD(Y,F);
1031             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1032             H                = _mm_setzero_pd();
1033             GMX_MM_TRANSPOSE2_PD(G,H);
1034             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1035             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1036             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1037
1038             /* CUBIC SPLINE TABLE DISPERSION */
1039             vfitab           = _mm_add_epi32(vfitab,ifour);
1040             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1041             F                = _mm_setzero_pd();
1042             GMX_MM_TRANSPOSE2_PD(Y,F);
1043             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1044             H                = _mm_setzero_pd();
1045             GMX_MM_TRANSPOSE2_PD(G,H);
1046             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1047             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1048             fvdw6            = _mm_mul_pd(c6_00,FF);
1049
1050             /* CUBIC SPLINE TABLE REPULSION */
1051             vfitab           = _mm_add_epi32(vfitab,ifour);
1052             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1053             F                = _mm_setzero_pd();
1054             GMX_MM_TRANSPOSE2_PD(Y,F);
1055             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1056             H                = _mm_setzero_pd();
1057             GMX_MM_TRANSPOSE2_PD(G,H);
1058             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1059             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1060             fvdw12           = _mm_mul_pd(c12_00,FF);
1061             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1062
1063             fscal            = _mm_add_pd(felec,fvdw);
1064
1065             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1066
1067             /* Update vectorial force */
1068             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1069             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1070             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1071
1072             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1073             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1074             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             r10              = _mm_mul_pd(rsq10,rinv10);
1081
1082             /* Compute parameters for interactions between i and j atoms */
1083             qq10             = _mm_mul_pd(iq1,jq0);
1084
1085             /* Calculate table index by multiplying r with table scale and truncate to integer */
1086             rt               = _mm_mul_pd(r10,vftabscale);
1087             vfitab           = _mm_cvttpd_epi32(rt);
1088 #ifdef __XOP__
1089             vfeps            = _mm_frcz_pd(rt);
1090 #else
1091             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1092 #endif
1093             twovfeps         = _mm_add_pd(vfeps,vfeps);
1094             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1095
1096             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1097             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1098             F                = _mm_setzero_pd();
1099             GMX_MM_TRANSPOSE2_PD(Y,F);
1100             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1101             H                = _mm_setzero_pd();
1102             GMX_MM_TRANSPOSE2_PD(G,H);
1103             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1104             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1105             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1110
1111             /* Update vectorial force */
1112             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1113             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1114             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1115
1116             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1117             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1118             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1119
1120             /**************************
1121              * CALCULATE INTERACTIONS *
1122              **************************/
1123
1124             r20              = _mm_mul_pd(rsq20,rinv20);
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             qq20             = _mm_mul_pd(iq2,jq0);
1128
1129             /* Calculate table index by multiplying r with table scale and truncate to integer */
1130             rt               = _mm_mul_pd(r20,vftabscale);
1131             vfitab           = _mm_cvttpd_epi32(rt);
1132 #ifdef __XOP__
1133             vfeps            = _mm_frcz_pd(rt);
1134 #else
1135             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1136 #endif
1137             twovfeps         = _mm_add_pd(vfeps,vfeps);
1138             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1139
1140             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1141             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1142             F                = _mm_setzero_pd();
1143             GMX_MM_TRANSPOSE2_PD(Y,F);
1144             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1145             H                = _mm_setzero_pd();
1146             GMX_MM_TRANSPOSE2_PD(G,H);
1147             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1148             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1149             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1150
1151             fscal            = felec;
1152
1153             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1154
1155             /* Update vectorial force */
1156             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1157             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1158             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1159
1160             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1161             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1162             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1163
1164             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1165
1166             /* Inner loop uses 151 flops */
1167         }
1168
1169         /* End of innermost loop */
1170
1171         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1172                                               f+i_coord_offset,fshift+i_shift_offset);
1173
1174         /* Increment number of inner iterations */
1175         inneriter                  += j_index_end - j_index_start;
1176
1177         /* Outer loop uses 18 flops */
1178     }
1179
1180     /* Increment number of outer iterations */
1181     outeriter        += nri;
1182
1183     /* Update outer/inner flops */
1184
1185     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);
1186 }