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