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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 102     real             *vftab;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     vftab            = kernel_data->table_elec_vdw->data;
 125     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm_setzero_pd();
 161         fiy0             = _mm_setzero_pd();
 162         fiz0             = _mm_setzero_pd();
 163         fix1             = _mm_setzero_pd();
 164         fiy1             = _mm_setzero_pd();
 165         fiz1             = _mm_setzero_pd();
 166         fix2             = _mm_setzero_pd();
 167         fiy2             = _mm_setzero_pd();
 168         fiz2             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172         vvdwsum          = _mm_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_pd(ix0,jx0);
 190             dy00             = _mm_sub_pd(iy0,jy0);
 191             dz00             = _mm_sub_pd(iz0,jz0);
 192             dx10             = _mm_sub_pd(ix1,jx0);
 193             dy10             = _mm_sub_pd(iy1,jy0);
 194             dz10             = _mm_sub_pd(iz1,jz0);
 195             dx20             = _mm_sub_pd(ix2,jx0);
 196             dy20             = _mm_sub_pd(iy2,jy0);
 197             dz20             = _mm_sub_pd(iz2,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 201             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 202             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 203
 204             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 205             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 206             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 210             vdwjidx0A        = 2*vdwtype[jnrA+0];
 211             vdwjidx0B        = 2*vdwtype[jnrB+0];
 212
 213             fjx0             = _mm_setzero_pd();
 214             fjy0             = _mm_setzero_pd();
 215             fjz0             = _mm_setzero_pd();
 216
 217             /**************************
 218              * CALCULATE INTERACTIONS *
 219              **************************/
 220
 221             r00              = _mm_mul_pd(rsq00,rinv00);
 222
 223             /* Compute parameters for interactions between i and j atoms */
 224             qq00             = _mm_mul_pd(iq0,jq0);
 225             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 226                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 227
 228             /* Calculate table index by multiplying r with table scale and truncate to integer */
 229             rt               = _mm_mul_pd(r00,vftabscale);
 230             vfitab           = _mm_cvttpd_epi32(rt);
 231 #ifdef __XOP__
 232             vfeps            = _mm_frcz_pd(rt);
 233 #else
 234             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 235 #endif
 236             twovfeps         = _mm_add_pd(vfeps,vfeps);
 237             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 238
 239             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 240             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 241             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 242             GMX_MM_TRANSPOSE2_PD(Y,F);
 243             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 244             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 245             GMX_MM_TRANSPOSE2_PD(G,H);
 246             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 247             VV               = _mm_macc_pd(vfeps,Fp,Y);
 248             velec            = _mm_mul_pd(qq00,VV);
 249             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 250             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 251
 252             /* CUBIC SPLINE TABLE DISPERSION */
 253             vfitab           = _mm_add_epi32(vfitab,ifour);
 254             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 255             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 256             GMX_MM_TRANSPOSE2_PD(Y,F);
 257             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 258             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 259             GMX_MM_TRANSPOSE2_PD(G,H);
 260             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 261             VV               = _mm_macc_pd(vfeps,Fp,Y);
 262             vvdw6            = _mm_mul_pd(c6_00,VV);
 263             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 264             fvdw6            = _mm_mul_pd(c6_00,FF);
 265
 266             /* CUBIC SPLINE TABLE REPULSION */
 267             vfitab           = _mm_add_epi32(vfitab,ifour);
 268             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 269             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 270             GMX_MM_TRANSPOSE2_PD(Y,F);
 271             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 272             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 273             GMX_MM_TRANSPOSE2_PD(G,H);
 274             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 275             VV               = _mm_macc_pd(vfeps,Fp,Y);
 276             vvdw12           = _mm_mul_pd(c12_00,VV);
 277             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 278             fvdw12           = _mm_mul_pd(c12_00,FF);
 279             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 280             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _mm_add_pd(velecsum,velec);
 284             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 285
 286             fscal            = _mm_add_pd(felec,fvdw);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 290             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 291             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 292
 293             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 294             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 295             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             r10              = _mm_mul_pd(rsq10,rinv10);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_pd(iq1,jq0);
 305
 306             /* Calculate table index by multiplying r with table scale and truncate to integer */
 307             rt               = _mm_mul_pd(r10,vftabscale);
 308             vfitab           = _mm_cvttpd_epi32(rt);
 309 #ifdef __XOP__
 310             vfeps            = _mm_frcz_pd(rt);
 311 #else
 312             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 313 #endif
 314             twovfeps         = _mm_add_pd(vfeps,vfeps);
 315             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 316
 317             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 318             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 319             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 320             GMX_MM_TRANSPOSE2_PD(Y,F);
 321             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 322             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 323             GMX_MM_TRANSPOSE2_PD(G,H);
 324             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 325             VV               = _mm_macc_pd(vfeps,Fp,Y);
 326             velec            = _mm_mul_pd(qq10,VV);
 327             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 328             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm_add_pd(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Update vectorial force */
 336             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 337             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 338             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 339
 340             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 341             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 342             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             r20              = _mm_mul_pd(rsq20,rinv20);
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm_mul_pd(iq2,jq0);
 352
 353             /* Calculate table index by multiplying r with table scale and truncate to integer */
 354             rt               = _mm_mul_pd(r20,vftabscale);
 355             vfitab           = _mm_cvttpd_epi32(rt);
 356 #ifdef __XOP__
 357             vfeps            = _mm_frcz_pd(rt);
 358 #else
 359             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 360 #endif
 361             twovfeps         = _mm_add_pd(vfeps,vfeps);
 362             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 363
 364             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 365             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 366             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 367             GMX_MM_TRANSPOSE2_PD(Y,F);
 368             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 369             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 370             GMX_MM_TRANSPOSE2_PD(G,H);
 371             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 372             VV               = _mm_macc_pd(vfeps,Fp,Y);
 373             velec            = _mm_mul_pd(qq20,VV);
 374             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 375             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm_add_pd(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             /* Update vectorial force */
 383             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 384             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 385             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 386
 387             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 388             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 389             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 390
 391             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 392
 393             /* Inner loop uses 171 flops */
 394         }
 395
 396         if(jidx<j_index_end)
 397         {
 398
 399             jnrA             = jjnr[jidx];
 400             j_coord_offsetA  = DIM*jnrA;
 401
 402             /* load j atom coordinates */
 403             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 404                                               &jx0,&jy0,&jz0);
 405
 406             /* Calculate displacement vector */
 407             dx00             = _mm_sub_pd(ix0,jx0);
 408             dy00             = _mm_sub_pd(iy0,jy0);
 409             dz00             = _mm_sub_pd(iz0,jz0);
 410             dx10             = _mm_sub_pd(ix1,jx0);
 411             dy10             = _mm_sub_pd(iy1,jy0);
 412             dz10             = _mm_sub_pd(iz1,jz0);
 413             dx20             = _mm_sub_pd(ix2,jx0);
 414             dy20             = _mm_sub_pd(iy2,jy0);
 415             dz20             = _mm_sub_pd(iz2,jz0);
 416
 417             /* Calculate squared distance and things based on it */
 418             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 419             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 420             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 421
 422             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 423             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 424             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 425
 426             /* Load parameters for j particles */
 427             jq0              = _mm_load_sd(charge+jnrA+0);
 428             vdwjidx0A        = 2*vdwtype[jnrA+0];
 429
 430             fjx0             = _mm_setzero_pd();
 431             fjy0             = _mm_setzero_pd();
 432             fjz0             = _mm_setzero_pd();
 433
 434             /**************************
 435              * CALCULATE INTERACTIONS *
 436              **************************/
 437
 438             r00              = _mm_mul_pd(rsq00,rinv00);
 439
 440             /* Compute parameters for interactions between i and j atoms */
 441             qq00             = _mm_mul_pd(iq0,jq0);
 442             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 443
 444             /* Calculate table index by multiplying r with table scale and truncate to integer */
 445             rt               = _mm_mul_pd(r00,vftabscale);
 446             vfitab           = _mm_cvttpd_epi32(rt);
 447 #ifdef __XOP__
 448             vfeps            = _mm_frcz_pd(rt);
 449 #else
 450             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 451 #endif
 452             twovfeps         = _mm_add_pd(vfeps,vfeps);
 453             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 454
 455             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 456             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 457             F                = _mm_setzero_pd();
 458             GMX_MM_TRANSPOSE2_PD(Y,F);
 459             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 460             H                = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(G,H);
 462             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 463             VV               = _mm_macc_pd(vfeps,Fp,Y);
 464             velec            = _mm_mul_pd(qq00,VV);
 465             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 466             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 467
 468             /* CUBIC SPLINE TABLE DISPERSION */
 469             vfitab           = _mm_add_epi32(vfitab,ifour);
 470             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 471             F                = _mm_setzero_pd();
 472             GMX_MM_TRANSPOSE2_PD(Y,F);
 473             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 474             H                = _mm_setzero_pd();
 475             GMX_MM_TRANSPOSE2_PD(G,H);
 476             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 477             VV               = _mm_macc_pd(vfeps,Fp,Y);
 478             vvdw6            = _mm_mul_pd(c6_00,VV);
 479             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 480             fvdw6            = _mm_mul_pd(c6_00,FF);
 481
 482             /* CUBIC SPLINE TABLE REPULSION */
 483             vfitab           = _mm_add_epi32(vfitab,ifour);
 484             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 485             F                = _mm_setzero_pd();
 486             GMX_MM_TRANSPOSE2_PD(Y,F);
 487             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 488             H                = _mm_setzero_pd();
 489             GMX_MM_TRANSPOSE2_PD(G,H);
 490             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 491             VV               = _mm_macc_pd(vfeps,Fp,Y);
 492             vvdw12           = _mm_mul_pd(c12_00,VV);
 493             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 494             fvdw12           = _mm_mul_pd(c12_00,FF);
 495             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 496             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 497
 498             /* Update potential sum for this i atom from the interaction with this j atom. */
 499             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 500             velecsum         = _mm_add_pd(velecsum,velec);
 501             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 502             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 503
 504             fscal            = _mm_add_pd(felec,fvdw);
 505
 506             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 507
 508             /* Update vectorial force */
 509             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 510             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 511             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 512
 513             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 514             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 515             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             r10              = _mm_mul_pd(rsq10,rinv10);
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq10             = _mm_mul_pd(iq1,jq0);
 525
 526             /* Calculate table index by multiplying r with table scale and truncate to integer */
 527             rt               = _mm_mul_pd(r10,vftabscale);
 528             vfitab           = _mm_cvttpd_epi32(rt);
 529 #ifdef __XOP__
 530             vfeps            = _mm_frcz_pd(rt);
 531 #else
 532             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 533 #endif
 534             twovfeps         = _mm_add_pd(vfeps,vfeps);
 535             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 536
 537             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 538             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 539             F                = _mm_setzero_pd();
 540             GMX_MM_TRANSPOSE2_PD(Y,F);
 541             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 542             H                = _mm_setzero_pd();
 543             GMX_MM_TRANSPOSE2_PD(G,H);
 544             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 545             VV               = _mm_macc_pd(vfeps,Fp,Y);
 546             velec            = _mm_mul_pd(qq10,VV);
 547             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 548             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 552             velecsum         = _mm_add_pd(velecsum,velec);
 553
 554             fscal            = felec;
 555
 556             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 557
 558             /* Update vectorial force */
 559             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 560             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 561             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 562
 563             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 564             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 565             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 566
 567             /**************************
 568              * CALCULATE INTERACTIONS *
 569              **************************/
 570
 571             r20              = _mm_mul_pd(rsq20,rinv20);
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq20             = _mm_mul_pd(iq2,jq0);
 575
 576             /* Calculate table index by multiplying r with table scale and truncate to integer */
 577             rt               = _mm_mul_pd(r20,vftabscale);
 578             vfitab           = _mm_cvttpd_epi32(rt);
 579 #ifdef __XOP__
 580             vfeps            = _mm_frcz_pd(rt);
 581 #else
 582             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 583 #endif
 584             twovfeps         = _mm_add_pd(vfeps,vfeps);
 585             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 586
 587             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 588             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 589             F                = _mm_setzero_pd();
 590             GMX_MM_TRANSPOSE2_PD(Y,F);
 591             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 592             H                = _mm_setzero_pd();
 593             GMX_MM_TRANSPOSE2_PD(G,H);
 594             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 595             VV               = _mm_macc_pd(vfeps,Fp,Y);
 596             velec            = _mm_mul_pd(qq20,VV);
 597             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 598             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 602             velecsum         = _mm_add_pd(velecsum,velec);
 603
 604             fscal            = felec;
 605
 606             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 607
 608             /* Update vectorial force */
 609             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 610             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 611             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 612
 613             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 614             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 615             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 616
 617             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 618
 619             /* Inner loop uses 171 flops */
 620         }
 621
 622         /* End of innermost loop */
 623
 624         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 625                                               f+i_coord_offset,fshift+i_shift_offset);
 626
 627         ggid                        = gid[iidx];
 628         /* Update potential energies */
 629         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 630         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 631
 632         /* Increment number of inner iterations */
 633         inneriter                  += j_index_end - j_index_start;
 634
 635         /* Outer loop uses 20 flops */
 636     }
 637
 638     /* Increment number of outer iterations */
 639     outeriter        += nri;
 640
 641     /* Update outer/inner flops */
 642
 643     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*171);
 644 }
 645 /*
 646  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 647  * Electrostatics interaction: CubicSplineTable
 648  * VdW interaction:            CubicSplineTable
 649  * Geometry:                   Water3-Particle
 650  * Calculate force/pot:        Force
 651  */
 652 void
 653 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 654                     (t_nblist                    * gmx_restrict       nlist,
 655                      rvec                        * gmx_restrict          xx,
 656                      rvec                        * gmx_restrict          ff,
 657                      t_forcerec                  * gmx_restrict          fr,
 658                      t_mdatoms                   * gmx_restrict     mdatoms,
 659                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 660                      t_nrnb                      * gmx_restrict        nrnb)
 661 {
 662     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 663      * just 0 for non-waters.
 664      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 665      * jnr indices corresponding to data put in the four positions in the SIMD register.
 666      */
 667     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 668     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 669     int              jnrA,jnrB;
 670     int              j_coord_offsetA,j_coord_offsetB;
 671     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 672     real             rcutoff_scalar;
 673     real             *shiftvec,*fshift,*x,*f;
 674     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 675     int              vdwioffset0;
 676     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 677     int              vdwioffset1;
 678     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 679     int              vdwioffset2;
 680     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 681     int              vdwjidx0A,vdwjidx0B;
 682     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 683     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 684     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 685     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 686     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 687     real             *charge;
 688     int              nvdwtype;
 689     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 690     int              *vdwtype;
 691     real             *vdwparam;
 692     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 693     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 694     __m128i          vfitab;
 695     __m128i          ifour       = _mm_set1_epi32(4);
 696     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 697     real             *vftab;
 698     __m128d          dummy_mask,cutoff_mask;
 699     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 700     __m128d          one     = _mm_set1_pd(1.0);
 701     __m128d          two     = _mm_set1_pd(2.0);
 702     x                = xx[0];
 703     f                = ff[0];
 704
 705     nri              = nlist->nri;
 706     iinr             = nlist->iinr;
 707     jindex           = nlist->jindex;
 708     jjnr             = nlist->jjnr;
 709     shiftidx         = nlist->shift;
 710     gid              = nlist->gid;
 711     shiftvec         = fr->shift_vec[0];
 712     fshift           = fr->fshift[0];
 713     facel            = _mm_set1_pd(fr->epsfac);
 714     charge           = mdatoms->chargeA;
 715     nvdwtype         = fr->ntype;
 716     vdwparam         = fr->nbfp;
 717     vdwtype          = mdatoms->typeA;
 718
 719     vftab            = kernel_data->table_elec_vdw->data;
 720     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 721
 722     /* Setup water-specific parameters */
 723     inr              = nlist->iinr[0];
 724     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 725     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 726     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 727     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 728
 729     /* Avoid stupid compiler warnings */
 730     jnrA = jnrB = 0;
 731     j_coord_offsetA = 0;
 732     j_coord_offsetB = 0;
 733
 734     outeriter        = 0;
 735     inneriter        = 0;
 736
 737     /* Start outer loop over neighborlists */
 738     for(iidx=0; iidx<nri; iidx++)
 739     {
 740         /* Load shift vector for this list */
 741         i_shift_offset   = DIM*shiftidx[iidx];
 742
 743         /* Load limits for loop over neighbors */
 744         j_index_start    = jindex[iidx];
 745         j_index_end      = jindex[iidx+1];
 746
 747         /* Get outer coordinate index */
 748         inr              = iinr[iidx];
 749         i_coord_offset   = DIM*inr;
 750
 751         /* Load i particle coords and add shift vector */
 752         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 753                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 754
 755         fix0             = _mm_setzero_pd();
 756         fiy0             = _mm_setzero_pd();
 757         fiz0             = _mm_setzero_pd();
 758         fix1             = _mm_setzero_pd();
 759         fiy1             = _mm_setzero_pd();
 760         fiz1             = _mm_setzero_pd();
 761         fix2             = _mm_setzero_pd();
 762         fiy2             = _mm_setzero_pd();
 763         fiz2             = _mm_setzero_pd();
 764
 765         /* Start inner kernel loop */
 766         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 767         {
 768
 769             /* Get j neighbor index, and coordinate index */
 770             jnrA             = jjnr[jidx];
 771             jnrB             = jjnr[jidx+1];
 772             j_coord_offsetA  = DIM*jnrA;
 773             j_coord_offsetB  = DIM*jnrB;
 774
 775             /* load j atom coordinates */
 776             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 777                                               &jx0,&jy0,&jz0);
 778
 779             /* Calculate displacement vector */
 780             dx00             = _mm_sub_pd(ix0,jx0);
 781             dy00             = _mm_sub_pd(iy0,jy0);
 782             dz00             = _mm_sub_pd(iz0,jz0);
 783             dx10             = _mm_sub_pd(ix1,jx0);
 784             dy10             = _mm_sub_pd(iy1,jy0);
 785             dz10             = _mm_sub_pd(iz1,jz0);
 786             dx20             = _mm_sub_pd(ix2,jx0);
 787             dy20             = _mm_sub_pd(iy2,jy0);
 788             dz20             = _mm_sub_pd(iz2,jz0);
 789
 790             /* Calculate squared distance and things based on it */
 791             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 792             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 793             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 794
 795             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 796             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 797             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 798
 799             /* Load parameters for j particles */
 800             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 801             vdwjidx0A        = 2*vdwtype[jnrA+0];
 802             vdwjidx0B        = 2*vdwtype[jnrB+0];
 803
 804             fjx0             = _mm_setzero_pd();
 805             fjy0             = _mm_setzero_pd();
 806             fjz0             = _mm_setzero_pd();
 807
 808             /**************************
 809              * CALCULATE INTERACTIONS *
 810              **************************/
 811
 812             r00              = _mm_mul_pd(rsq00,rinv00);
 813
 814             /* Compute parameters for interactions between i and j atoms */
 815             qq00             = _mm_mul_pd(iq0,jq0);
 816             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 817                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 818
 819             /* Calculate table index by multiplying r with table scale and truncate to integer */
 820             rt               = _mm_mul_pd(r00,vftabscale);
 821             vfitab           = _mm_cvttpd_epi32(rt);
 822 #ifdef __XOP__
 823             vfeps            = _mm_frcz_pd(rt);
 824 #else
 825             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 826 #endif
 827             twovfeps         = _mm_add_pd(vfeps,vfeps);
 828             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 829
 830             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 831             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 832             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 833             GMX_MM_TRANSPOSE2_PD(Y,F);
 834             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 835             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 836             GMX_MM_TRANSPOSE2_PD(G,H);
 837             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 838             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 839             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 840
 841             /* CUBIC SPLINE TABLE DISPERSION */
 842             vfitab           = _mm_add_epi32(vfitab,ifour);
 843             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 844             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 845             GMX_MM_TRANSPOSE2_PD(Y,F);
 846             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 847             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 848             GMX_MM_TRANSPOSE2_PD(G,H);
 849             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 850             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 851             fvdw6            = _mm_mul_pd(c6_00,FF);
 852
 853             /* CUBIC SPLINE TABLE REPULSION */
 854             vfitab           = _mm_add_epi32(vfitab,ifour);
 855             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 856             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 857             GMX_MM_TRANSPOSE2_PD(Y,F);
 858             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 859             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 860             GMX_MM_TRANSPOSE2_PD(G,H);
 861             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 862             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 863             fvdw12           = _mm_mul_pd(c12_00,FF);
 864             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 865
 866             fscal            = _mm_add_pd(felec,fvdw);
 867
 868             /* Update vectorial force */
 869             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 870             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 871             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 872
 873             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 874             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 875             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 876
 877             /**************************
 878              * CALCULATE INTERACTIONS *
 879              **************************/
 880
 881             r10              = _mm_mul_pd(rsq10,rinv10);
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq10             = _mm_mul_pd(iq1,jq0);
 885
 886             /* Calculate table index by multiplying r with table scale and truncate to integer */
 887             rt               = _mm_mul_pd(r10,vftabscale);
 888             vfitab           = _mm_cvttpd_epi32(rt);
 889 #ifdef __XOP__
 890             vfeps            = _mm_frcz_pd(rt);
 891 #else
 892             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 893 #endif
 894             twovfeps         = _mm_add_pd(vfeps,vfeps);
 895             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 896
 897             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 898             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 899             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 900             GMX_MM_TRANSPOSE2_PD(Y,F);
 901             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 902             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 903             GMX_MM_TRANSPOSE2_PD(G,H);
 904             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 905             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 906             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 907
 908             fscal            = felec;
 909
 910             /* Update vectorial force */
 911             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 912             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 913             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 914
 915             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 916             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 917             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 918
 919             /**************************
 920              * CALCULATE INTERACTIONS *
 921              **************************/
 922
 923             r20              = _mm_mul_pd(rsq20,rinv20);
 924
 925             /* Compute parameters for interactions between i and j atoms */
 926             qq20             = _mm_mul_pd(iq2,jq0);
 927
 928             /* Calculate table index by multiplying r with table scale and truncate to integer */
 929             rt               = _mm_mul_pd(r20,vftabscale);
 930             vfitab           = _mm_cvttpd_epi32(rt);
 931 #ifdef __XOP__
 932             vfeps            = _mm_frcz_pd(rt);
 933 #else
 934             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 935 #endif
 936             twovfeps         = _mm_add_pd(vfeps,vfeps);
 937             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 938
 939             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 940             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 941             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 942             GMX_MM_TRANSPOSE2_PD(Y,F);
 943             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 944             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 945             GMX_MM_TRANSPOSE2_PD(G,H);
 946             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 947             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 948             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 949
 950             fscal            = felec;
 951
 952             /* Update vectorial force */
 953             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 954             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 955             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 956
 957             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 958             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 959             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 960
 961             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 962
 963             /* Inner loop uses 151 flops */
 964         }
 965
 966         if(jidx<j_index_end)
 967         {
 968
 969             jnrA             = jjnr[jidx];
 970             j_coord_offsetA  = DIM*jnrA;
 971
 972             /* load j atom coordinates */
 973             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 974                                               &jx0,&jy0,&jz0);
 975
 976             /* Calculate displacement vector */
 977             dx00             = _mm_sub_pd(ix0,jx0);
 978             dy00             = _mm_sub_pd(iy0,jy0);
 979             dz00             = _mm_sub_pd(iz0,jz0);
 980             dx10             = _mm_sub_pd(ix1,jx0);
 981             dy10             = _mm_sub_pd(iy1,jy0);
 982             dz10             = _mm_sub_pd(iz1,jz0);
 983             dx20             = _mm_sub_pd(ix2,jx0);
 984             dy20             = _mm_sub_pd(iy2,jy0);
 985             dz20             = _mm_sub_pd(iz2,jz0);
 986
 987             /* Calculate squared distance and things based on it */
 988             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 989             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 990             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 991
 992             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 993             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 994             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 995
 996             /* Load parameters for j particles */
 997             jq0              = _mm_load_sd(charge+jnrA+0);
 998             vdwjidx0A        = 2*vdwtype[jnrA+0];
 999
1000             fjx0             = _mm_setzero_pd();
1001             fjy0             = _mm_setzero_pd();
1002             fjz0             = _mm_setzero_pd();
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             r00              = _mm_mul_pd(rsq00,rinv00);
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq00             = _mm_mul_pd(iq0,jq0);
1012             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1013
1014             /* Calculate table index by multiplying r with table scale and truncate to integer */
1015             rt               = _mm_mul_pd(r00,vftabscale);
1016             vfitab           = _mm_cvttpd_epi32(rt);
1017 #ifdef __XOP__
1018             vfeps            = _mm_frcz_pd(rt);
1019 #else
1020             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1021 #endif
1022             twovfeps         = _mm_add_pd(vfeps,vfeps);
1023             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1024
1025             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1026             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1027             F                = _mm_setzero_pd();
1028             GMX_MM_TRANSPOSE2_PD(Y,F);
1029             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1030             H                = _mm_setzero_pd();
1031             GMX_MM_TRANSPOSE2_PD(G,H);
1032             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1033             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1034             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1035
1036             /* CUBIC SPLINE TABLE DISPERSION */
1037             vfitab           = _mm_add_epi32(vfitab,ifour);
1038             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1039             F                = _mm_setzero_pd();
1040             GMX_MM_TRANSPOSE2_PD(Y,F);
1041             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1042             H                = _mm_setzero_pd();
1043             GMX_MM_TRANSPOSE2_PD(G,H);
1044             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1045             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1046             fvdw6            = _mm_mul_pd(c6_00,FF);
1047
1048             /* CUBIC SPLINE TABLE REPULSION */
1049             vfitab           = _mm_add_epi32(vfitab,ifour);
1050             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1051             F                = _mm_setzero_pd();
1052             GMX_MM_TRANSPOSE2_PD(Y,F);
1053             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1054             H                = _mm_setzero_pd();
1055             GMX_MM_TRANSPOSE2_PD(G,H);
1056             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1057             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1058             fvdw12           = _mm_mul_pd(c12_00,FF);
1059             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1060
1061             fscal            = _mm_add_pd(felec,fvdw);
1062
1063             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1064
1065             /* Update vectorial force */
1066             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1067             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1068             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1069
1070             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1071             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1072             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             r10              = _mm_mul_pd(rsq10,rinv10);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq10             = _mm_mul_pd(iq1,jq0);
1082
1083             /* Calculate table index by multiplying r with table scale and truncate to integer */
1084             rt               = _mm_mul_pd(r10,vftabscale);
1085             vfitab           = _mm_cvttpd_epi32(rt);
1086 #ifdef __XOP__
1087             vfeps            = _mm_frcz_pd(rt);
1088 #else
1089             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1090 #endif
1091             twovfeps         = _mm_add_pd(vfeps,vfeps);
1092             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1093
1094             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1095             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1096             F                = _mm_setzero_pd();
1097             GMX_MM_TRANSPOSE2_PD(Y,F);
1098             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1099             H                = _mm_setzero_pd();
1100             GMX_MM_TRANSPOSE2_PD(G,H);
1101             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1102             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1103             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1108
1109             /* Update vectorial force */
1110             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1111             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1112             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1113
1114             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1115             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1116             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1117
1118             /**************************
1119              * CALCULATE INTERACTIONS *
1120              **************************/
1121
1122             r20              = _mm_mul_pd(rsq20,rinv20);
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             qq20             = _mm_mul_pd(iq2,jq0);
1126
1127             /* Calculate table index by multiplying r with table scale and truncate to integer */
1128             rt               = _mm_mul_pd(r20,vftabscale);
1129             vfitab           = _mm_cvttpd_epi32(rt);
1130 #ifdef __XOP__
1131             vfeps            = _mm_frcz_pd(rt);
1132 #else
1133             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1134 #endif
1135             twovfeps         = _mm_add_pd(vfeps,vfeps);
1136             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1137
1138             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1139             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1140             F                = _mm_setzero_pd();
1141             GMX_MM_TRANSPOSE2_PD(Y,F);
1142             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1143             H                = _mm_setzero_pd();
1144             GMX_MM_TRANSPOSE2_PD(G,H);
1145             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1146             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1147             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1148
1149             fscal            = felec;
1150
1151             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1152
1153             /* Update vectorial force */
1154             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1155             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1156             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1157
1158             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1159             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1160             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1161
1162             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1163
1164             /* Inner loop uses 151 flops */
1165         }
1166
1167         /* End of innermost loop */
1168
1169         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1170                                               f+i_coord_offset,fshift+i_shift_offset);
1171
1172         /* Increment number of inner iterations */
1173         inneriter                  += j_index_end - j_index_start;
1174
1175         /* Outer loop uses 18 flops */
1176     }
1177
1178     /* Increment number of outer iterations */
1179     outeriter        += nri;
1180
1181     /* Update outer/inner flops */
1182
1183     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);
1184 }