src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_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              vdwjidx0A,vdwjidx0B;
  83     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  84     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  85     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  86     real             *charge;
  87     int              nvdwtype;
  88     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  89     int              *vdwtype;
  90     real             *vdwparam;
  91     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  92     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  93     __m128i          vfitab;
  94     __m128i          ifour       = _mm_set1_epi32(4);
  95     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  96     real             *vftab;
  97     __m128d          dummy_mask,cutoff_mask;
  98     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  99     __m128d          one     = _mm_set1_pd(1.0);
 100     __m128d          two     = _mm_set1_pd(2.0);
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     facel            = _mm_set1_pd(fr->epsfac);
 113     charge           = mdatoms->chargeA;
 114     krf              = _mm_set1_pd(fr->ic->k_rf);
 115     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 116     crf              = _mm_set1_pd(fr->ic->c_rf);
 117     nvdwtype         = fr->ntype;
 118     vdwparam         = fr->nbfp;
 119     vdwtype          = mdatoms->typeA;
 120
 121     vftab            = kernel_data->table_vdw->data;
 122     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 127     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _mm_setzero_pd();
 155         fiy0             = _mm_setzero_pd();
 156         fiz0             = _mm_setzero_pd();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 187
 188             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 189
 190             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 191
 192             /* Load parameters for j particles */
 193             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 194             vdwjidx0A        = 2*vdwtype[jnrA+0];
 195             vdwjidx0B        = 2*vdwtype[jnrB+0];
 196
 197             /**************************
 198              * CALCULATE INTERACTIONS *
 199              **************************/
 200
 201             if (gmx_mm_any_lt(rsq00,rcutoff2))
 202             {
 203
 204             r00              = _mm_mul_pd(rsq00,rinv00);
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq00             = _mm_mul_pd(iq0,jq0);
 208             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 209                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 210
 211             /* Calculate table index by multiplying r with table scale and truncate to integer */
 212             rt               = _mm_mul_pd(r00,vftabscale);
 213             vfitab           = _mm_cvttpd_epi32(rt);
 214 #ifdef __XOP__
 215             vfeps            = _mm_frcz_pd(rt);
 216 #else
 217             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 218 #endif
 219             twovfeps         = _mm_add_pd(vfeps,vfeps);
 220             vfitab           = _mm_slli_epi32(vfitab,3);
 221
 222             /* REACTION-FIELD ELECTROSTATICS */
 223             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 224             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 225
 226             /* CUBIC SPLINE TABLE DISPERSION */
 227             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 228             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 229             GMX_MM_TRANSPOSE2_PD(Y,F);
 230             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 231             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 232             GMX_MM_TRANSPOSE2_PD(G,H);
 233             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 234             VV               = _mm_macc_pd(vfeps,Fp,Y);
 235             vvdw6            = _mm_mul_pd(c6_00,VV);
 236             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 237             fvdw6            = _mm_mul_pd(c6_00,FF);
 238
 239             /* CUBIC SPLINE TABLE REPULSION */
 240             vfitab           = _mm_add_epi32(vfitab,ifour);
 241             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 242             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 243             GMX_MM_TRANSPOSE2_PD(Y,F);
 244             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 245             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 246             GMX_MM_TRANSPOSE2_PD(G,H);
 247             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 248             VV               = _mm_macc_pd(vfeps,Fp,Y);
 249             vvdw12           = _mm_mul_pd(c12_00,VV);
 250             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 251             fvdw12           = _mm_mul_pd(c12_00,FF);
 252             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 253             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 254
 255             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             velec            = _mm_and_pd(velec,cutoff_mask);
 259             velecsum         = _mm_add_pd(velecsum,velec);
 260             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 261             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 262
 263             fscal            = _mm_add_pd(felec,fvdw);
 264
 265             fscal            = _mm_and_pd(fscal,cutoff_mask);
 266
 267             /* Update vectorial force */
 268             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 269             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 270             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 271
 272             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 273                                                    _mm_mul_pd(dx00,fscal),
 274                                                    _mm_mul_pd(dy00,fscal),
 275                                                    _mm_mul_pd(dz00,fscal));
 276
 277             }
 278
 279             /* Inner loop uses 75 flops */
 280         }
 281
 282         if(jidx<j_index_end)
 283         {
 284
 285             jnrA             = jjnr[jidx];
 286             j_coord_offsetA  = DIM*jnrA;
 287
 288             /* load j atom coordinates */
 289             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 290                                               &jx0,&jy0,&jz0);
 291
 292             /* Calculate displacement vector */
 293             dx00             = _mm_sub_pd(ix0,jx0);
 294             dy00             = _mm_sub_pd(iy0,jy0);
 295             dz00             = _mm_sub_pd(iz0,jz0);
 296
 297             /* Calculate squared distance and things based on it */
 298             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 299
 300             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 301
 302             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 303
 304             /* Load parameters for j particles */
 305             jq0              = _mm_load_sd(charge+jnrA+0);
 306             vdwjidx0A        = 2*vdwtype[jnrA+0];
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             if (gmx_mm_any_lt(rsq00,rcutoff2))
 313             {
 314
 315             r00              = _mm_mul_pd(rsq00,rinv00);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq00             = _mm_mul_pd(iq0,jq0);
 319             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 320
 321             /* Calculate table index by multiplying r with table scale and truncate to integer */
 322             rt               = _mm_mul_pd(r00,vftabscale);
 323             vfitab           = _mm_cvttpd_epi32(rt);
 324 #ifdef __XOP__
 325             vfeps            = _mm_frcz_pd(rt);
 326 #else
 327             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 328 #endif
 329             twovfeps         = _mm_add_pd(vfeps,vfeps);
 330             vfitab           = _mm_slli_epi32(vfitab,3);
 331
 332             /* REACTION-FIELD ELECTROSTATICS */
 333             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 334             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 335
 336             /* CUBIC SPLINE TABLE DISPERSION */
 337             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 338             F                = _mm_setzero_pd();
 339             GMX_MM_TRANSPOSE2_PD(Y,F);
 340             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 341             H                = _mm_setzero_pd();
 342             GMX_MM_TRANSPOSE2_PD(G,H);
 343             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 344             VV               = _mm_macc_pd(vfeps,Fp,Y);
 345             vvdw6            = _mm_mul_pd(c6_00,VV);
 346             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 347             fvdw6            = _mm_mul_pd(c6_00,FF);
 348
 349             /* CUBIC SPLINE TABLE REPULSION */
 350             vfitab           = _mm_add_epi32(vfitab,ifour);
 351             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 352             F                = _mm_setzero_pd();
 353             GMX_MM_TRANSPOSE2_PD(Y,F);
 354             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 355             H                = _mm_setzero_pd();
 356             GMX_MM_TRANSPOSE2_PD(G,H);
 357             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 358             VV               = _mm_macc_pd(vfeps,Fp,Y);
 359             vvdw12           = _mm_mul_pd(c12_00,VV);
 360             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 361             fvdw12           = _mm_mul_pd(c12_00,FF);
 362             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 363             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 364
 365             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velec            = _mm_and_pd(velec,cutoff_mask);
 369             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 370             velecsum         = _mm_add_pd(velecsum,velec);
 371             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 372             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 373             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 374
 375             fscal            = _mm_add_pd(felec,fvdw);
 376
 377             fscal            = _mm_and_pd(fscal,cutoff_mask);
 378
 379             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 380
 381             /* Update vectorial force */
 382             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 383             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 384             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 385
 386             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 387                                                    _mm_mul_pd(dx00,fscal),
 388                                                    _mm_mul_pd(dy00,fscal),
 389                                                    _mm_mul_pd(dz00,fscal));
 390
 391             }
 392
 393             /* Inner loop uses 75 flops */
 394         }
 395
 396         /* End of innermost loop */
 397
 398         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 399                                               f+i_coord_offset,fshift+i_shift_offset);
 400
 401         ggid                        = gid[iidx];
 402         /* Update potential energies */
 403         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 404         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 405
 406         /* Increment number of inner iterations */
 407         inneriter                  += j_index_end - j_index_start;
 408
 409         /* Outer loop uses 9 flops */
 410     }
 411
 412     /* Increment number of outer iterations */
 413     outeriter        += nri;
 414
 415     /* Update outer/inner flops */
 416
 417     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*75);
 418 }
 419 /*
 420  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 421  * Electrostatics interaction: ReactionField
 422  * VdW interaction:            CubicSplineTable
 423  * Geometry:                   Particle-Particle
 424  * Calculate force/pot:        Force
 425  */
 426 void
 427 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 428                     (t_nblist                    * gmx_restrict       nlist,
 429                      rvec                        * gmx_restrict          xx,
 430                      rvec                        * gmx_restrict          ff,
 431                      t_forcerec                  * gmx_restrict          fr,
 432                      t_mdatoms                   * gmx_restrict     mdatoms,
 433                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 434                      t_nrnb                      * gmx_restrict        nrnb)
 435 {
 436     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 437      * just 0 for non-waters.
 438      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 439      * jnr indices corresponding to data put in the four positions in the SIMD register.
 440      */
 441     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 442     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 443     int              jnrA,jnrB;
 444     int              j_coord_offsetA,j_coord_offsetB;
 445     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 446     real             rcutoff_scalar;
 447     real             *shiftvec,*fshift,*x,*f;
 448     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 449     int              vdwioffset0;
 450     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 451     int              vdwjidx0A,vdwjidx0B;
 452     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 453     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 454     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 455     real             *charge;
 456     int              nvdwtype;
 457     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 458     int              *vdwtype;
 459     real             *vdwparam;
 460     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 461     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 462     __m128i          vfitab;
 463     __m128i          ifour       = _mm_set1_epi32(4);
 464     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 465     real             *vftab;
 466     __m128d          dummy_mask,cutoff_mask;
 467     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 468     __m128d          one     = _mm_set1_pd(1.0);
 469     __m128d          two     = _mm_set1_pd(2.0);
 470     x                = xx[0];
 471     f                = ff[0];
 472
 473     nri              = nlist->nri;
 474     iinr             = nlist->iinr;
 475     jindex           = nlist->jindex;
 476     jjnr             = nlist->jjnr;
 477     shiftidx         = nlist->shift;
 478     gid              = nlist->gid;
 479     shiftvec         = fr->shift_vec[0];
 480     fshift           = fr->fshift[0];
 481     facel            = _mm_set1_pd(fr->epsfac);
 482     charge           = mdatoms->chargeA;
 483     krf              = _mm_set1_pd(fr->ic->k_rf);
 484     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 485     crf              = _mm_set1_pd(fr->ic->c_rf);
 486     nvdwtype         = fr->ntype;
 487     vdwparam         = fr->nbfp;
 488     vdwtype          = mdatoms->typeA;
 489
 490     vftab            = kernel_data->table_vdw->data;
 491     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 492
 493     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 494     rcutoff_scalar   = fr->rcoulomb;
 495     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 496     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 497
 498     /* Avoid stupid compiler warnings */
 499     jnrA = jnrB = 0;
 500     j_coord_offsetA = 0;
 501     j_coord_offsetB = 0;
 502
 503     outeriter        = 0;
 504     inneriter        = 0;
 505
 506     /* Start outer loop over neighborlists */
 507     for(iidx=0; iidx<nri; iidx++)
 508     {
 509         /* Load shift vector for this list */
 510         i_shift_offset   = DIM*shiftidx[iidx];
 511
 512         /* Load limits for loop over neighbors */
 513         j_index_start    = jindex[iidx];
 514         j_index_end      = jindex[iidx+1];
 515
 516         /* Get outer coordinate index */
 517         inr              = iinr[iidx];
 518         i_coord_offset   = DIM*inr;
 519
 520         /* Load i particle coords and add shift vector */
 521         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 522
 523         fix0             = _mm_setzero_pd();
 524         fiy0             = _mm_setzero_pd();
 525         fiz0             = _mm_setzero_pd();
 526
 527         /* Load parameters for i particles */
 528         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 529         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 530
 531         /* Start inner kernel loop */
 532         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 533         {
 534
 535             /* Get j neighbor index, and coordinate index */
 536             jnrA             = jjnr[jidx];
 537             jnrB             = jjnr[jidx+1];
 538             j_coord_offsetA  = DIM*jnrA;
 539             j_coord_offsetB  = DIM*jnrB;
 540
 541             /* load j atom coordinates */
 542             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 543                                               &jx0,&jy0,&jz0);
 544
 545             /* Calculate displacement vector */
 546             dx00             = _mm_sub_pd(ix0,jx0);
 547             dy00             = _mm_sub_pd(iy0,jy0);
 548             dz00             = _mm_sub_pd(iz0,jz0);
 549
 550             /* Calculate squared distance and things based on it */
 551             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 552
 553             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 554
 555             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 556
 557             /* Load parameters for j particles */
 558             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 559             vdwjidx0A        = 2*vdwtype[jnrA+0];
 560             vdwjidx0B        = 2*vdwtype[jnrB+0];
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             if (gmx_mm_any_lt(rsq00,rcutoff2))
 567             {
 568
 569             r00              = _mm_mul_pd(rsq00,rinv00);
 570
 571             /* Compute parameters for interactions between i and j atoms */
 572             qq00             = _mm_mul_pd(iq0,jq0);
 573             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 574                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 575
 576             /* Calculate table index by multiplying r with table scale and truncate to integer */
 577             rt               = _mm_mul_pd(r00,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(vfitab,3);
 586
 587             /* REACTION-FIELD ELECTROSTATICS */
 588             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 589
 590             /* CUBIC SPLINE TABLE DISPERSION */
 591             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 592             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 593             GMX_MM_TRANSPOSE2_PD(Y,F);
 594             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 595             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 596             GMX_MM_TRANSPOSE2_PD(G,H);
 597             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 598             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 599             fvdw6            = _mm_mul_pd(c6_00,FF);
 600
 601             /* CUBIC SPLINE TABLE REPULSION */
 602             vfitab           = _mm_add_epi32(vfitab,ifour);
 603             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 604             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 605             GMX_MM_TRANSPOSE2_PD(Y,F);
 606             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 607             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 608             GMX_MM_TRANSPOSE2_PD(G,H);
 609             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 610             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 611             fvdw12           = _mm_mul_pd(c12_00,FF);
 612             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 613
 614             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 615
 616             fscal            = _mm_add_pd(felec,fvdw);
 617
 618             fscal            = _mm_and_pd(fscal,cutoff_mask);
 619
 620             /* Update vectorial force */
 621             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 622             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 623             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 624
 625             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 626                                                    _mm_mul_pd(dx00,fscal),
 627                                                    _mm_mul_pd(dy00,fscal),
 628                                                    _mm_mul_pd(dz00,fscal));
 629
 630             }
 631
 632             /* Inner loop uses 60 flops */
 633         }
 634
 635         if(jidx<j_index_end)
 636         {
 637
 638             jnrA             = jjnr[jidx];
 639             j_coord_offsetA  = DIM*jnrA;
 640
 641             /* load j atom coordinates */
 642             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 643                                               &jx0,&jy0,&jz0);
 644
 645             /* Calculate displacement vector */
 646             dx00             = _mm_sub_pd(ix0,jx0);
 647             dy00             = _mm_sub_pd(iy0,jy0);
 648             dz00             = _mm_sub_pd(iz0,jz0);
 649
 650             /* Calculate squared distance and things based on it */
 651             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 652
 653             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 654
 655             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 656
 657             /* Load parameters for j particles */
 658             jq0              = _mm_load_sd(charge+jnrA+0);
 659             vdwjidx0A        = 2*vdwtype[jnrA+0];
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             if (gmx_mm_any_lt(rsq00,rcutoff2))
 666             {
 667
 668             r00              = _mm_mul_pd(rsq00,rinv00);
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             qq00             = _mm_mul_pd(iq0,jq0);
 672             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 673
 674             /* Calculate table index by multiplying r with table scale and truncate to integer */
 675             rt               = _mm_mul_pd(r00,vftabscale);
 676             vfitab           = _mm_cvttpd_epi32(rt);
 677 #ifdef __XOP__
 678             vfeps            = _mm_frcz_pd(rt);
 679 #else
 680             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 681 #endif
 682             twovfeps         = _mm_add_pd(vfeps,vfeps);
 683             vfitab           = _mm_slli_epi32(vfitab,3);
 684
 685             /* REACTION-FIELD ELECTROSTATICS */
 686             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 687
 688             /* CUBIC SPLINE TABLE DISPERSION */
 689             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 690             F                = _mm_setzero_pd();
 691             GMX_MM_TRANSPOSE2_PD(Y,F);
 692             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 693             H                = _mm_setzero_pd();
 694             GMX_MM_TRANSPOSE2_PD(G,H);
 695             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 696             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 697             fvdw6            = _mm_mul_pd(c6_00,FF);
 698
 699             /* CUBIC SPLINE TABLE REPULSION */
 700             vfitab           = _mm_add_epi32(vfitab,ifour);
 701             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 702             F                = _mm_setzero_pd();
 703             GMX_MM_TRANSPOSE2_PD(Y,F);
 704             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 705             H                = _mm_setzero_pd();
 706             GMX_MM_TRANSPOSE2_PD(G,H);
 707             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 708             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 709             fvdw12           = _mm_mul_pd(c12_00,FF);
 710             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 711
 712             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 713
 714             fscal            = _mm_add_pd(felec,fvdw);
 715
 716             fscal            = _mm_and_pd(fscal,cutoff_mask);
 717
 718             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 719
 720             /* Update vectorial force */
 721             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 722             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 723             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 724
 725             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 726                                                    _mm_mul_pd(dx00,fscal),
 727                                                    _mm_mul_pd(dy00,fscal),
 728                                                    _mm_mul_pd(dz00,fscal));
 729
 730             }
 731
 732             /* Inner loop uses 60 flops */
 733         }
 734
 735         /* End of innermost loop */
 736
 737         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 738                                               f+i_coord_offset,fshift+i_shift_offset);
 739
 740         /* Increment number of inner iterations */
 741         inneriter                  += j_index_end - j_index_start;
 742
 743         /* Outer loop uses 7 flops */
 744     }
 745
 746     /* Increment number of outer iterations */
 747     outeriter        += nri;
 748
 749     /* Update outer/inner flops */
 750
 751     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 752 }