src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 104     real             *vftab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     krf              = _mm_set1_pd(fr->ic->k_rf);
 123     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 124     crf              = _mm_set1_pd(fr->ic->c_rf);
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 140     rcutoff_scalar   = fr->rcoulomb;
 141     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 142     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_pd();
 171         fiy0             = _mm_setzero_pd();
 172         fiz0             = _mm_setzero_pd();
 173         fix1             = _mm_setzero_pd();
 174         fiy1             = _mm_setzero_pd();
 175         fiz1             = _mm_setzero_pd();
 176         fix2             = _mm_setzero_pd();
 177         fiy2             = _mm_setzero_pd();
 178         fiz2             = _mm_setzero_pd();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_pd();
 182         vvdwsum          = _mm_setzero_pd();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_pd(ix0,jx0);
 200             dy00             = _mm_sub_pd(iy0,jy0);
 201             dz00             = _mm_sub_pd(iz0,jz0);
 202             dx10             = _mm_sub_pd(ix1,jx0);
 203             dy10             = _mm_sub_pd(iy1,jy0);
 204             dz10             = _mm_sub_pd(iz1,jz0);
 205             dx20             = _mm_sub_pd(ix2,jx0);
 206             dy20             = _mm_sub_pd(iy2,jy0);
 207             dz20             = _mm_sub_pd(iz2,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 213
 214             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 215             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 216             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 217
 218             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 219             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226
 227             fjx0             = _mm_setzero_pd();
 228             fjy0             = _mm_setzero_pd();
 229             fjz0             = _mm_setzero_pd();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             if (gmx_mm_any_lt(rsq00,rcutoff2))
 236             {
 237
 238             r00              = _mm_mul_pd(rsq00,rinv00);
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq00             = _mm_mul_pd(iq0,jq0);
 242             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 243                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 244
 245             /* Calculate table index by multiplying r with table scale and truncate to integer */
 246             rt               = _mm_mul_pd(r00,vftabscale);
 247             vfitab           = _mm_cvttpd_epi32(rt);
 248 #ifdef __XOP__
 249             vfeps            = _mm_frcz_pd(rt);
 250 #else
 251             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 252 #endif
 253             twovfeps         = _mm_add_pd(vfeps,vfeps);
 254             vfitab           = _mm_slli_epi32(vfitab,3);
 255
 256             /* REACTION-FIELD ELECTROSTATICS */
 257             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 258             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 259
 260             /* CUBIC SPLINE TABLE DISPERSION */
 261             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 262             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 263             GMX_MM_TRANSPOSE2_PD(Y,F);
 264             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 265             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 266             GMX_MM_TRANSPOSE2_PD(G,H);
 267             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 268             VV               = _mm_macc_pd(vfeps,Fp,Y);
 269             vvdw6            = _mm_mul_pd(c6_00,VV);
 270             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 271             fvdw6            = _mm_mul_pd(c6_00,FF);
 272
 273             /* CUBIC SPLINE TABLE REPULSION */
 274             vfitab           = _mm_add_epi32(vfitab,ifour);
 275             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 277             GMX_MM_TRANSPOSE2_PD(Y,F);
 278             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 279             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 280             GMX_MM_TRANSPOSE2_PD(G,H);
 281             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 282             VV               = _mm_macc_pd(vfeps,Fp,Y);
 283             vvdw12           = _mm_mul_pd(c12_00,VV);
 284             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 285             fvdw12           = _mm_mul_pd(c12_00,FF);
 286             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 287             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 288
 289             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 290
 291             /* Update potential sum for this i atom from the interaction with this j atom. */
 292             velec            = _mm_and_pd(velec,cutoff_mask);
 293             velecsum         = _mm_add_pd(velecsum,velec);
 294             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 295             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 296
 297             fscal            = _mm_add_pd(felec,fvdw);
 298
 299             fscal            = _mm_and_pd(fscal,cutoff_mask);
 300
 301             /* Update vectorial force */
 302             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 303             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 304             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 305
 306             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 307             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 308             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 309
 310             }
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm_any_lt(rsq10,rcutoff2))
 317             {
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_pd(iq1,jq0);
 321
 322             /* REACTION-FIELD ELECTROSTATICS */
 323             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 324             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 325
 326             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velec            = _mm_and_pd(velec,cutoff_mask);
 330             velecsum         = _mm_add_pd(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334             fscal            = _mm_and_pd(fscal,cutoff_mask);
 335
 336             /* Update vectorial force */
 337             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 338             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 339             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 340
 341             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 342             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 343             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 344
 345             }
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             if (gmx_mm_any_lt(rsq20,rcutoff2))
 352             {
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq20             = _mm_mul_pd(iq2,jq0);
 356
 357             /* REACTION-FIELD ELECTROSTATICS */
 358             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 359             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 360
 361             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velec            = _mm_and_pd(velec,cutoff_mask);
 365             velecsum         = _mm_add_pd(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             fscal            = _mm_and_pd(fscal,cutoff_mask);
 370
 371             /* Update vectorial force */
 372             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 373             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 374             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 375
 376             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 377             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 378             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 379
 380             }
 381
 382             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 383
 384             /* Inner loop uses 156 flops */
 385         }
 386
 387         if(jidx<j_index_end)
 388         {
 389
 390             jnrA             = jjnr[jidx];
 391             j_coord_offsetA  = DIM*jnrA;
 392
 393             /* load j atom coordinates */
 394             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 395                                               &jx0,&jy0,&jz0);
 396
 397             /* Calculate displacement vector */
 398             dx00             = _mm_sub_pd(ix0,jx0);
 399             dy00             = _mm_sub_pd(iy0,jy0);
 400             dz00             = _mm_sub_pd(iz0,jz0);
 401             dx10             = _mm_sub_pd(ix1,jx0);
 402             dy10             = _mm_sub_pd(iy1,jy0);
 403             dz10             = _mm_sub_pd(iz1,jz0);
 404             dx20             = _mm_sub_pd(ix2,jx0);
 405             dy20             = _mm_sub_pd(iy2,jy0);
 406             dz20             = _mm_sub_pd(iz2,jz0);
 407
 408             /* Calculate squared distance and things based on it */
 409             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 410             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 411             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 412
 413             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 414             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 415             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 416
 417             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 418             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 419             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 420
 421             /* Load parameters for j particles */
 422             jq0              = _mm_load_sd(charge+jnrA+0);
 423             vdwjidx0A        = 2*vdwtype[jnrA+0];
 424
 425             fjx0             = _mm_setzero_pd();
 426             fjy0             = _mm_setzero_pd();
 427             fjz0             = _mm_setzero_pd();
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             if (gmx_mm_any_lt(rsq00,rcutoff2))
 434             {
 435
 436             r00              = _mm_mul_pd(rsq00,rinv00);
 437
 438             /* Compute parameters for interactions between i and j atoms */
 439             qq00             = _mm_mul_pd(iq0,jq0);
 440             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 441
 442             /* Calculate table index by multiplying r with table scale and truncate to integer */
 443             rt               = _mm_mul_pd(r00,vftabscale);
 444             vfitab           = _mm_cvttpd_epi32(rt);
 445 #ifdef __XOP__
 446             vfeps            = _mm_frcz_pd(rt);
 447 #else
 448             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 449 #endif
 450             twovfeps         = _mm_add_pd(vfeps,vfeps);
 451             vfitab           = _mm_slli_epi32(vfitab,3);
 452
 453             /* REACTION-FIELD ELECTROSTATICS */
 454             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 455             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 456
 457             /* CUBIC SPLINE TABLE DISPERSION */
 458             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 459             F                = _mm_setzero_pd();
 460             GMX_MM_TRANSPOSE2_PD(Y,F);
 461             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 462             H                = _mm_setzero_pd();
 463             GMX_MM_TRANSPOSE2_PD(G,H);
 464             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 465             VV               = _mm_macc_pd(vfeps,Fp,Y);
 466             vvdw6            = _mm_mul_pd(c6_00,VV);
 467             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 468             fvdw6            = _mm_mul_pd(c6_00,FF);
 469
 470             /* CUBIC SPLINE TABLE REPULSION */
 471             vfitab           = _mm_add_epi32(vfitab,ifour);
 472             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 473             F                = _mm_setzero_pd();
 474             GMX_MM_TRANSPOSE2_PD(Y,F);
 475             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 476             H                = _mm_setzero_pd();
 477             GMX_MM_TRANSPOSE2_PD(G,H);
 478             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 479             VV               = _mm_macc_pd(vfeps,Fp,Y);
 480             vvdw12           = _mm_mul_pd(c12_00,VV);
 481             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 482             fvdw12           = _mm_mul_pd(c12_00,FF);
 483             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 484             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 485
 486             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velec            = _mm_and_pd(velec,cutoff_mask);
 490             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 491             velecsum         = _mm_add_pd(velecsum,velec);
 492             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 493             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 494             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 495
 496             fscal            = _mm_add_pd(felec,fvdw);
 497
 498             fscal            = _mm_and_pd(fscal,cutoff_mask);
 499
 500             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 501
 502             /* Update vectorial force */
 503             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 504             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 505             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 506
 507             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 508             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 509             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 510
 511             }
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             if (gmx_mm_any_lt(rsq10,rcutoff2))
 518             {
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq10             = _mm_mul_pd(iq1,jq0);
 522
 523             /* REACTION-FIELD ELECTROSTATICS */
 524             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 525             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 526
 527             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm_and_pd(velec,cutoff_mask);
 531             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 532             velecsum         = _mm_add_pd(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm_and_pd(fscal,cutoff_mask);
 537
 538             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 539
 540             /* Update vectorial force */
 541             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 542             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 543             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 544
 545             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 546             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 547             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 548
 549             }
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             if (gmx_mm_any_lt(rsq20,rcutoff2))
 556             {
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq20             = _mm_mul_pd(iq2,jq0);
 560
 561             /* REACTION-FIELD ELECTROSTATICS */
 562             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 563             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 564
 565             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _mm_and_pd(velec,cutoff_mask);
 569             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 570             velecsum         = _mm_add_pd(velecsum,velec);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm_and_pd(fscal,cutoff_mask);
 575
 576             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 577
 578             /* Update vectorial force */
 579             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 580             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 581             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 582
 583             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 584             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 585             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 586
 587             }
 588
 589             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 590
 591             /* Inner loop uses 156 flops */
 592         }
 593
 594         /* End of innermost loop */
 595
 596         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 597                                               f+i_coord_offset,fshift+i_shift_offset);
 598
 599         ggid                        = gid[iidx];
 600         /* Update potential energies */
 601         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 602         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 603
 604         /* Increment number of inner iterations */
 605         inneriter                  += j_index_end - j_index_start;
 606
 607         /* Outer loop uses 20 flops */
 608     }
 609
 610     /* Increment number of outer iterations */
 611     outeriter        += nri;
 612
 613     /* Update outer/inner flops */
 614
 615     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*156);
 616 }
 617 /*
 618  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 619  * Electrostatics interaction: ReactionField
 620  * VdW interaction:            CubicSplineTable
 621  * Geometry:                   Water3-Particle
 622  * Calculate force/pot:        Force
 623  */
 624 void
 625 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 626                     (t_nblist                    * gmx_restrict       nlist,
 627                      rvec                        * gmx_restrict          xx,
 628                      rvec                        * gmx_restrict          ff,
 629                      t_forcerec                  * gmx_restrict          fr,
 630                      t_mdatoms                   * gmx_restrict     mdatoms,
 631                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 632                      t_nrnb                      * gmx_restrict        nrnb)
 633 {
 634     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 635      * just 0 for non-waters.
 636      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 637      * jnr indices corresponding to data put in the four positions in the SIMD register.
 638      */
 639     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 640     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 641     int              jnrA,jnrB;
 642     int              j_coord_offsetA,j_coord_offsetB;
 643     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 644     real             rcutoff_scalar;
 645     real             *shiftvec,*fshift,*x,*f;
 646     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 647     int              vdwioffset0;
 648     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 649     int              vdwioffset1;
 650     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 651     int              vdwioffset2;
 652     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 653     int              vdwjidx0A,vdwjidx0B;
 654     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 655     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 656     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 657     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 658     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 659     real             *charge;
 660     int              nvdwtype;
 661     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 662     int              *vdwtype;
 663     real             *vdwparam;
 664     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 665     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 666     __m128i          vfitab;
 667     __m128i          ifour       = _mm_set1_epi32(4);
 668     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 669     real             *vftab;
 670     __m128d          dummy_mask,cutoff_mask;
 671     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 672     __m128d          one     = _mm_set1_pd(1.0);
 673     __m128d          two     = _mm_set1_pd(2.0);
 674     x                = xx[0];
 675     f                = ff[0];
 676
 677     nri              = nlist->nri;
 678     iinr             = nlist->iinr;
 679     jindex           = nlist->jindex;
 680     jjnr             = nlist->jjnr;
 681     shiftidx         = nlist->shift;
 682     gid              = nlist->gid;
 683     shiftvec         = fr->shift_vec[0];
 684     fshift           = fr->fshift[0];
 685     facel            = _mm_set1_pd(fr->epsfac);
 686     charge           = mdatoms->chargeA;
 687     krf              = _mm_set1_pd(fr->ic->k_rf);
 688     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 689     crf              = _mm_set1_pd(fr->ic->c_rf);
 690     nvdwtype         = fr->ntype;
 691     vdwparam         = fr->nbfp;
 692     vdwtype          = mdatoms->typeA;
 693
 694     vftab            = kernel_data->table_vdw->data;
 695     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 696
 697     /* Setup water-specific parameters */
 698     inr              = nlist->iinr[0];
 699     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 700     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 701     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 702     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 703
 704     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 705     rcutoff_scalar   = fr->rcoulomb;
 706     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 707     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 708
 709     /* Avoid stupid compiler warnings */
 710     jnrA = jnrB = 0;
 711     j_coord_offsetA = 0;
 712     j_coord_offsetB = 0;
 713
 714     outeriter        = 0;
 715     inneriter        = 0;
 716
 717     /* Start outer loop over neighborlists */
 718     for(iidx=0; iidx<nri; iidx++)
 719     {
 720         /* Load shift vector for this list */
 721         i_shift_offset   = DIM*shiftidx[iidx];
 722
 723         /* Load limits for loop over neighbors */
 724         j_index_start    = jindex[iidx];
 725         j_index_end      = jindex[iidx+1];
 726
 727         /* Get outer coordinate index */
 728         inr              = iinr[iidx];
 729         i_coord_offset   = DIM*inr;
 730
 731         /* Load i particle coords and add shift vector */
 732         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 733                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 734
 735         fix0             = _mm_setzero_pd();
 736         fiy0             = _mm_setzero_pd();
 737         fiz0             = _mm_setzero_pd();
 738         fix1             = _mm_setzero_pd();
 739         fiy1             = _mm_setzero_pd();
 740         fiz1             = _mm_setzero_pd();
 741         fix2             = _mm_setzero_pd();
 742         fiy2             = _mm_setzero_pd();
 743         fiz2             = _mm_setzero_pd();
 744
 745         /* Start inner kernel loop */
 746         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 747         {
 748
 749             /* Get j neighbor index, and coordinate index */
 750             jnrA             = jjnr[jidx];
 751             jnrB             = jjnr[jidx+1];
 752             j_coord_offsetA  = DIM*jnrA;
 753             j_coord_offsetB  = DIM*jnrB;
 754
 755             /* load j atom coordinates */
 756             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 757                                               &jx0,&jy0,&jz0);
 758
 759             /* Calculate displacement vector */
 760             dx00             = _mm_sub_pd(ix0,jx0);
 761             dy00             = _mm_sub_pd(iy0,jy0);
 762             dz00             = _mm_sub_pd(iz0,jz0);
 763             dx10             = _mm_sub_pd(ix1,jx0);
 764             dy10             = _mm_sub_pd(iy1,jy0);
 765             dz10             = _mm_sub_pd(iz1,jz0);
 766             dx20             = _mm_sub_pd(ix2,jx0);
 767             dy20             = _mm_sub_pd(iy2,jy0);
 768             dz20             = _mm_sub_pd(iz2,jz0);
 769
 770             /* Calculate squared distance and things based on it */
 771             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 772             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 773             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 774
 775             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 776             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 777             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 778
 779             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 780             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 781             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 782
 783             /* Load parameters for j particles */
 784             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 785             vdwjidx0A        = 2*vdwtype[jnrA+0];
 786             vdwjidx0B        = 2*vdwtype[jnrB+0];
 787
 788             fjx0             = _mm_setzero_pd();
 789             fjy0             = _mm_setzero_pd();
 790             fjz0             = _mm_setzero_pd();
 791
 792             /**************************
 793              * CALCULATE INTERACTIONS *
 794              **************************/
 795
 796             if (gmx_mm_any_lt(rsq00,rcutoff2))
 797             {
 798
 799             r00              = _mm_mul_pd(rsq00,rinv00);
 800
 801             /* Compute parameters for interactions between i and j atoms */
 802             qq00             = _mm_mul_pd(iq0,jq0);
 803             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 804                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 805
 806             /* Calculate table index by multiplying r with table scale and truncate to integer */
 807             rt               = _mm_mul_pd(r00,vftabscale);
 808             vfitab           = _mm_cvttpd_epi32(rt);
 809 #ifdef __XOP__
 810             vfeps            = _mm_frcz_pd(rt);
 811 #else
 812             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 813 #endif
 814             twovfeps         = _mm_add_pd(vfeps,vfeps);
 815             vfitab           = _mm_slli_epi32(vfitab,3);
 816
 817             /* REACTION-FIELD ELECTROSTATICS */
 818             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 819
 820             /* CUBIC SPLINE TABLE DISPERSION */
 821             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 822             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 823             GMX_MM_TRANSPOSE2_PD(Y,F);
 824             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 825             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 826             GMX_MM_TRANSPOSE2_PD(G,H);
 827             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 828             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 829             fvdw6            = _mm_mul_pd(c6_00,FF);
 830
 831             /* CUBIC SPLINE TABLE REPULSION */
 832             vfitab           = _mm_add_epi32(vfitab,ifour);
 833             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 834             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 835             GMX_MM_TRANSPOSE2_PD(Y,F);
 836             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 837             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 838             GMX_MM_TRANSPOSE2_PD(G,H);
 839             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 840             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 841             fvdw12           = _mm_mul_pd(c12_00,FF);
 842             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 843
 844             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 845
 846             fscal            = _mm_add_pd(felec,fvdw);
 847
 848             fscal            = _mm_and_pd(fscal,cutoff_mask);
 849
 850             /* Update vectorial force */
 851             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 852             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 853             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 854
 855             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 856             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 857             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 858
 859             }
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             if (gmx_mm_any_lt(rsq10,rcutoff2))
 866             {
 867
 868             /* Compute parameters for interactions between i and j atoms */
 869             qq10             = _mm_mul_pd(iq1,jq0);
 870
 871             /* REACTION-FIELD ELECTROSTATICS */
 872             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 873
 874             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 875
 876             fscal            = felec;
 877
 878             fscal            = _mm_and_pd(fscal,cutoff_mask);
 879
 880             /* Update vectorial force */
 881             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 882             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 883             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 884
 885             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 886             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 887             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 888
 889             }
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             if (gmx_mm_any_lt(rsq20,rcutoff2))
 896             {
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             qq20             = _mm_mul_pd(iq2,jq0);
 900
 901             /* REACTION-FIELD ELECTROSTATICS */
 902             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 903
 904             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 905
 906             fscal            = felec;
 907
 908             fscal            = _mm_and_pd(fscal,cutoff_mask);
 909
 910             /* Update vectorial force */
 911             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 912             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 913             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 914
 915             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 916             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 917             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 918
 919             }
 920
 921             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 922
 923             /* Inner loop uses 129 flops */
 924         }
 925
 926         if(jidx<j_index_end)
 927         {
 928
 929             jnrA             = jjnr[jidx];
 930             j_coord_offsetA  = DIM*jnrA;
 931
 932             /* load j atom coordinates */
 933             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 934                                               &jx0,&jy0,&jz0);
 935
 936             /* Calculate displacement vector */
 937             dx00             = _mm_sub_pd(ix0,jx0);
 938             dy00             = _mm_sub_pd(iy0,jy0);
 939             dz00             = _mm_sub_pd(iz0,jz0);
 940             dx10             = _mm_sub_pd(ix1,jx0);
 941             dy10             = _mm_sub_pd(iy1,jy0);
 942             dz10             = _mm_sub_pd(iz1,jz0);
 943             dx20             = _mm_sub_pd(ix2,jx0);
 944             dy20             = _mm_sub_pd(iy2,jy0);
 945             dz20             = _mm_sub_pd(iz2,jz0);
 946
 947             /* Calculate squared distance and things based on it */
 948             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 949             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 950             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 951
 952             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 953             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 954             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 955
 956             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 957             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 958             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 959
 960             /* Load parameters for j particles */
 961             jq0              = _mm_load_sd(charge+jnrA+0);
 962             vdwjidx0A        = 2*vdwtype[jnrA+0];
 963
 964             fjx0             = _mm_setzero_pd();
 965             fjy0             = _mm_setzero_pd();
 966             fjz0             = _mm_setzero_pd();
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             if (gmx_mm_any_lt(rsq00,rcutoff2))
 973             {
 974
 975             r00              = _mm_mul_pd(rsq00,rinv00);
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             qq00             = _mm_mul_pd(iq0,jq0);
 979             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 980
 981             /* Calculate table index by multiplying r with table scale and truncate to integer */
 982             rt               = _mm_mul_pd(r00,vftabscale);
 983             vfitab           = _mm_cvttpd_epi32(rt);
 984 #ifdef __XOP__
 985             vfeps            = _mm_frcz_pd(rt);
 986 #else
 987             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 988 #endif
 989             twovfeps         = _mm_add_pd(vfeps,vfeps);
 990             vfitab           = _mm_slli_epi32(vfitab,3);
 991
 992             /* REACTION-FIELD ELECTROSTATICS */
 993             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 994
 995             /* CUBIC SPLINE TABLE DISPERSION */
 996             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 997             F                = _mm_setzero_pd();
 998             GMX_MM_TRANSPOSE2_PD(Y,F);
 999             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1000             H                = _mm_setzero_pd();
1001             GMX_MM_TRANSPOSE2_PD(G,H);
1002             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1003             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1004             fvdw6            = _mm_mul_pd(c6_00,FF);
1005
1006             /* CUBIC SPLINE TABLE REPULSION */
1007             vfitab           = _mm_add_epi32(vfitab,ifour);
1008             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1009             F                = _mm_setzero_pd();
1010             GMX_MM_TRANSPOSE2_PD(Y,F);
1011             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1012             H                = _mm_setzero_pd();
1013             GMX_MM_TRANSPOSE2_PD(G,H);
1014             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1015             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1016             fvdw12           = _mm_mul_pd(c12_00,FF);
1017             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1018
1019             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1020
1021             fscal            = _mm_add_pd(felec,fvdw);
1022
1023             fscal            = _mm_and_pd(fscal,cutoff_mask);
1024
1025             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1026
1027             /* Update vectorial force */
1028             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1029             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1030             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1031
1032             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1033             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1034             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1035
1036             }
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             if (gmx_mm_any_lt(rsq10,rcutoff2))
1043             {
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq10             = _mm_mul_pd(iq1,jq0);
1047
1048             /* REACTION-FIELD ELECTROSTATICS */
1049             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1050
1051             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1052
1053             fscal            = felec;
1054
1055             fscal            = _mm_and_pd(fscal,cutoff_mask);
1056
1057             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1058
1059             /* Update vectorial force */
1060             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1061             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1062             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1063
1064             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1065             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1066             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1067
1068             }
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             if (gmx_mm_any_lt(rsq20,rcutoff2))
1075             {
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             qq20             = _mm_mul_pd(iq2,jq0);
1079
1080             /* REACTION-FIELD ELECTROSTATICS */
1081             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1082
1083             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1084
1085             fscal            = felec;
1086
1087             fscal            = _mm_and_pd(fscal,cutoff_mask);
1088
1089             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1090
1091             /* Update vectorial force */
1092             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1093             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1094             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1095
1096             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1097             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1098             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1099
1100             }
1101
1102             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1103
1104             /* Inner loop uses 129 flops */
1105         }
1106
1107         /* End of innermost loop */
1108
1109         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1110                                               f+i_coord_offset,fshift+i_shift_offset);
1111
1112         /* Increment number of inner iterations */
1113         inneriter                  += j_index_end - j_index_start;
1114
1115         /* Outer loop uses 18 flops */
1116     }
1117
1118     /* Increment number of outer iterations */
1119     outeriter        += nri;
1120
1121     /* Update outer/inner flops */
1122
1123     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*129);
1124 }