src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  78     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  79     __m128i          ewitab;
  80     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  82     __m128d          dummy_mask,cutoff_mask;
  83     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  84     __m128d          one     = _mm_set1_pd(1.0);
  85     __m128d          two     = _mm_set1_pd(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_pd(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99     nvdwtype         = fr->ntype;
 100     vdwparam         = fr->nbfp;
 101     vdwtype          = mdatoms->typeA;
 102
 103     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 104     ewtab            = fr->ic->tabq_coul_FDV0;
 105     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 106     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 107
 108     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 109     rcutoff_scalar   = fr->rcoulomb;
 110     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 111     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 112
 113     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 114     rvdw             = _mm_set1_pd(fr->rvdw);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 140
 141         fix0             = _mm_setzero_pd();
 142         fiy0             = _mm_setzero_pd();
 143         fiz0             = _mm_setzero_pd();
 144
 145         /* Load parameters for i particles */
 146         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 147         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 148
 149         /* Reset potential sums */
 150         velecsum         = _mm_setzero_pd();
 151         vvdwsum          = _mm_setzero_pd();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162
 163             /* load j atom coordinates */
 164             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_pd(ix0,jx0);
 169             dy00             = _mm_sub_pd(iy0,jy0);
 170             dz00             = _mm_sub_pd(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 174
 175             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 176
 177             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 178
 179             /* Load parameters for j particles */
 180             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 181             vdwjidx0A        = 2*vdwtype[jnrA+0];
 182             vdwjidx0B        = 2*vdwtype[jnrB+0];
 183
 184             /**************************
 185              * CALCULATE INTERACTIONS *
 186              **************************/
 187
 188             if (gmx_mm_any_lt(rsq00,rcutoff2))
 189             {
 190
 191             r00              = _mm_mul_pd(rsq00,rinv00);
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm_mul_pd(iq0,jq0);
 195             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 196                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 197
 198             /* EWALD ELECTROSTATICS */
 199
 200             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 201             ewrt             = _mm_mul_pd(r00,ewtabscale);
 202             ewitab           = _mm_cvttpd_epi32(ewrt);
 203 #ifdef __XOP__
 204             eweps            = _mm_frcz_pd(ewrt);
 205 #else
 206             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 207 #endif
 208             twoeweps         = _mm_add_pd(eweps,eweps);
 209             ewitab           = _mm_slli_epi32(ewitab,2);
 210             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 211             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 212             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 213             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 214             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 215             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 216             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 217             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 218             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 219             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 220
 221             /* LENNARD-JONES DISPERSION/REPULSION */
 222
 223             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 224             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 225             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 226             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 227                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 228             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 229
 230             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             velec            = _mm_and_pd(velec,cutoff_mask);
 234             velecsum         = _mm_add_pd(velecsum,velec);
 235             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 236             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 237
 238             fscal            = _mm_add_pd(felec,fvdw);
 239
 240             fscal            = _mm_and_pd(fscal,cutoff_mask);
 241
 242             /* Update vectorial force */
 243             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 244             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 245             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 246
 247             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 248                                                    _mm_mul_pd(dx00,fscal),
 249                                                    _mm_mul_pd(dy00,fscal),
 250                                                    _mm_mul_pd(dz00,fscal));
 251
 252             }
 253
 254             /* Inner loop uses 67 flops */
 255         }
 256
 257         if(jidx<j_index_end)
 258         {
 259
 260             jnrA             = jjnr[jidx];
 261             j_coord_offsetA  = DIM*jnrA;
 262
 263             /* load j atom coordinates */
 264             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 265                                               &jx0,&jy0,&jz0);
 266
 267             /* Calculate displacement vector */
 268             dx00             = _mm_sub_pd(ix0,jx0);
 269             dy00             = _mm_sub_pd(iy0,jy0);
 270             dz00             = _mm_sub_pd(iz0,jz0);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 274
 275             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 276
 277             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 278
 279             /* Load parameters for j particles */
 280             jq0              = _mm_load_sd(charge+jnrA+0);
 281             vdwjidx0A        = 2*vdwtype[jnrA+0];
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             if (gmx_mm_any_lt(rsq00,rcutoff2))
 288             {
 289
 290             r00              = _mm_mul_pd(rsq00,rinv00);
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq00             = _mm_mul_pd(iq0,jq0);
 294             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 295
 296             /* EWALD ELECTROSTATICS */
 297
 298             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 299             ewrt             = _mm_mul_pd(r00,ewtabscale);
 300             ewitab           = _mm_cvttpd_epi32(ewrt);
 301 #ifdef __XOP__
 302             eweps            = _mm_frcz_pd(ewrt);
 303 #else
 304             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 305 #endif
 306             twoeweps         = _mm_add_pd(eweps,eweps);
 307             ewitab           = _mm_slli_epi32(ewitab,2);
 308             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 309             ewtabD           = _mm_setzero_pd();
 310             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 311             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 312             ewtabFn          = _mm_setzero_pd();
 313             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 314             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 315             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 316             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 317             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 318
 319             /* LENNARD-JONES DISPERSION/REPULSION */
 320
 321             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 322             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 323             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 324             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 325                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 326             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 327
 328             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velec            = _mm_and_pd(velec,cutoff_mask);
 332             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 333             velecsum         = _mm_add_pd(velecsum,velec);
 334             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 335             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 336             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 337
 338             fscal            = _mm_add_pd(felec,fvdw);
 339
 340             fscal            = _mm_and_pd(fscal,cutoff_mask);
 341
 342             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 343
 344             /* Update vectorial force */
 345             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 346             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 347             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 348
 349             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 350                                                    _mm_mul_pd(dx00,fscal),
 351                                                    _mm_mul_pd(dy00,fscal),
 352                                                    _mm_mul_pd(dz00,fscal));
 353
 354             }
 355
 356             /* Inner loop uses 67 flops */
 357         }
 358
 359         /* End of innermost loop */
 360
 361         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 362                                               f+i_coord_offset,fshift+i_shift_offset);
 363
 364         ggid                        = gid[iidx];
 365         /* Update potential energies */
 366         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 367         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 368
 369         /* Increment number of inner iterations */
 370         inneriter                  += j_index_end - j_index_start;
 371
 372         /* Outer loop uses 9 flops */
 373     }
 374
 375     /* Increment number of outer iterations */
 376     outeriter        += nri;
 377
 378     /* Update outer/inner flops */
 379
 380     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*67);
 381 }
 382 /*
 383  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 384  * Electrostatics interaction: Ewald
 385  * VdW interaction:            LennardJones
 386  * Geometry:                   Particle-Particle
 387  * Calculate force/pot:        Force
 388  */
 389 void
 390 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 391                     (t_nblist * gmx_restrict                nlist,
 392                      rvec * gmx_restrict                    xx,
 393                      rvec * gmx_restrict                    ff,
 394                      t_forcerec * gmx_restrict              fr,
 395                      t_mdatoms * gmx_restrict               mdatoms,
 396                      nb_kernel_data_t * gmx_restrict        kernel_data,
 397                      t_nrnb * gmx_restrict                  nrnb)
 398 {
 399     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 400      * just 0 for non-waters.
 401      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 402      * jnr indices corresponding to data put in the four positions in the SIMD register.
 403      */
 404     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 405     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 406     int              jnrA,jnrB;
 407     int              j_coord_offsetA,j_coord_offsetB;
 408     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 409     real             rcutoff_scalar;
 410     real             *shiftvec,*fshift,*x,*f;
 411     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 412     int              vdwioffset0;
 413     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 414     int              vdwjidx0A,vdwjidx0B;
 415     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 416     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 417     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 418     real             *charge;
 419     int              nvdwtype;
 420     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 421     int              *vdwtype;
 422     real             *vdwparam;
 423     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 424     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 425     __m128i          ewitab;
 426     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 427     real             *ewtab;
 428     __m128d          dummy_mask,cutoff_mask;
 429     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 430     __m128d          one     = _mm_set1_pd(1.0);
 431     __m128d          two     = _mm_set1_pd(2.0);
 432     x                = xx[0];
 433     f                = ff[0];
 434
 435     nri              = nlist->nri;
 436     iinr             = nlist->iinr;
 437     jindex           = nlist->jindex;
 438     jjnr             = nlist->jjnr;
 439     shiftidx         = nlist->shift;
 440     gid              = nlist->gid;
 441     shiftvec         = fr->shift_vec[0];
 442     fshift           = fr->fshift[0];
 443     facel            = _mm_set1_pd(fr->epsfac);
 444     charge           = mdatoms->chargeA;
 445     nvdwtype         = fr->ntype;
 446     vdwparam         = fr->nbfp;
 447     vdwtype          = mdatoms->typeA;
 448
 449     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 450     ewtab            = fr->ic->tabq_coul_F;
 451     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 452     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 453
 454     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 455     rcutoff_scalar   = fr->rcoulomb;
 456     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 457     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 458
 459     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 460     rvdw             = _mm_set1_pd(fr->rvdw);
 461
 462     /* Avoid stupid compiler warnings */
 463     jnrA = jnrB = 0;
 464     j_coord_offsetA = 0;
 465     j_coord_offsetB = 0;
 466
 467     outeriter        = 0;
 468     inneriter        = 0;
 469
 470     /* Start outer loop over neighborlists */
 471     for(iidx=0; iidx<nri; iidx++)
 472     {
 473         /* Load shift vector for this list */
 474         i_shift_offset   = DIM*shiftidx[iidx];
 475
 476         /* Load limits for loop over neighbors */
 477         j_index_start    = jindex[iidx];
 478         j_index_end      = jindex[iidx+1];
 479
 480         /* Get outer coordinate index */
 481         inr              = iinr[iidx];
 482         i_coord_offset   = DIM*inr;
 483
 484         /* Load i particle coords and add shift vector */
 485         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 486
 487         fix0             = _mm_setzero_pd();
 488         fiy0             = _mm_setzero_pd();
 489         fiz0             = _mm_setzero_pd();
 490
 491         /* Load parameters for i particles */
 492         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 493         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 494
 495         /* Start inner kernel loop */
 496         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 497         {
 498
 499             /* Get j neighbor index, and coordinate index */
 500             jnrA             = jjnr[jidx];
 501             jnrB             = jjnr[jidx+1];
 502             j_coord_offsetA  = DIM*jnrA;
 503             j_coord_offsetB  = DIM*jnrB;
 504
 505             /* load j atom coordinates */
 506             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 507                                               &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _mm_sub_pd(ix0,jx0);
 511             dy00             = _mm_sub_pd(iy0,jy0);
 512             dz00             = _mm_sub_pd(iz0,jz0);
 513
 514             /* Calculate squared distance and things based on it */
 515             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 516
 517             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 518
 519             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 520
 521             /* Load parameters for j particles */
 522             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 523             vdwjidx0A        = 2*vdwtype[jnrA+0];
 524             vdwjidx0B        = 2*vdwtype[jnrB+0];
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             if (gmx_mm_any_lt(rsq00,rcutoff2))
 531             {
 532
 533             r00              = _mm_mul_pd(rsq00,rinv00);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq00             = _mm_mul_pd(iq0,jq0);
 537             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 538                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 539
 540             /* EWALD ELECTROSTATICS */
 541
 542             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 543             ewrt             = _mm_mul_pd(r00,ewtabscale);
 544             ewitab           = _mm_cvttpd_epi32(ewrt);
 545 #ifdef __XOP__
 546             eweps            = _mm_frcz_pd(ewrt);
 547 #else
 548             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 549 #endif
 550             twoeweps         = _mm_add_pd(eweps,eweps);
 551             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 552                                          &ewtabF,&ewtabFn);
 553             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 554             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 555
 556             /* LENNARD-JONES DISPERSION/REPULSION */
 557
 558             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 559             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 560
 561             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 562
 563             fscal            = _mm_add_pd(felec,fvdw);
 564
 565             fscal            = _mm_and_pd(fscal,cutoff_mask);
 566
 567             /* Update vectorial force */
 568             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 569             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 570             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 571
 572             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 573                                                    _mm_mul_pd(dx00,fscal),
 574                                                    _mm_mul_pd(dy00,fscal),
 575                                                    _mm_mul_pd(dz00,fscal));
 576
 577             }
 578
 579             /* Inner loop uses 49 flops */
 580         }
 581
 582         if(jidx<j_index_end)
 583         {
 584
 585             jnrA             = jjnr[jidx];
 586             j_coord_offsetA  = DIM*jnrA;
 587
 588             /* load j atom coordinates */
 589             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 590                                               &jx0,&jy0,&jz0);
 591
 592             /* Calculate displacement vector */
 593             dx00             = _mm_sub_pd(ix0,jx0);
 594             dy00             = _mm_sub_pd(iy0,jy0);
 595             dz00             = _mm_sub_pd(iz0,jz0);
 596
 597             /* Calculate squared distance and things based on it */
 598             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 599
 600             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 601
 602             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 603
 604             /* Load parameters for j particles */
 605             jq0              = _mm_load_sd(charge+jnrA+0);
 606             vdwjidx0A        = 2*vdwtype[jnrA+0];
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             if (gmx_mm_any_lt(rsq00,rcutoff2))
 613             {
 614
 615             r00              = _mm_mul_pd(rsq00,rinv00);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq00             = _mm_mul_pd(iq0,jq0);
 619             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 620
 621             /* EWALD ELECTROSTATICS */
 622
 623             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 624             ewrt             = _mm_mul_pd(r00,ewtabscale);
 625             ewitab           = _mm_cvttpd_epi32(ewrt);
 626 #ifdef __XOP__
 627             eweps            = _mm_frcz_pd(ewrt);
 628 #else
 629             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 630 #endif
 631             twoeweps         = _mm_add_pd(eweps,eweps);
 632             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 633             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 634             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 635
 636             /* LENNARD-JONES DISPERSION/REPULSION */
 637
 638             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 639             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 640
 641             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 642
 643             fscal            = _mm_add_pd(felec,fvdw);
 644
 645             fscal            = _mm_and_pd(fscal,cutoff_mask);
 646
 647             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 648
 649             /* Update vectorial force */
 650             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 651             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 652             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 653
 654             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 655                                                    _mm_mul_pd(dx00,fscal),
 656                                                    _mm_mul_pd(dy00,fscal),
 657                                                    _mm_mul_pd(dz00,fscal));
 658
 659             }
 660
 661             /* Inner loop uses 49 flops */
 662         }
 663
 664         /* End of innermost loop */
 665
 666         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 667                                               f+i_coord_offset,fshift+i_shift_offset);
 668
 669         /* Increment number of inner iterations */
 670         inneriter                  += j_index_end - j_index_start;
 671
 672         /* Outer loop uses 7 flops */
 673     }
 674
 675     /* Increment number of outer iterations */
 676     outeriter        += nri;
 677
 678     /* Update outer/inner flops */
 679
 680     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*49);
 681 }