src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_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_GeomW4P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_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              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 105     real             *vftab;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm_set1_pd(fr->ic->k_rf);
 124     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 125     crf              = _mm_set1_pd(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_vdw->data;
 131     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[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     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 138     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149
 150     outeriter        = 0;
 151     inneriter        = 0;
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 170
 171         fix0             = _mm_setzero_pd();
 172         fiy0             = _mm_setzero_pd();
 173         fiz0             = _mm_setzero_pd();
 174         fix1             = _mm_setzero_pd();
 175         fiy1             = _mm_setzero_pd();
 176         fiz1             = _mm_setzero_pd();
 177         fix2             = _mm_setzero_pd();
 178         fiy2             = _mm_setzero_pd();
 179         fiz2             = _mm_setzero_pd();
 180         fix3             = _mm_setzero_pd();
 181         fiy3             = _mm_setzero_pd();
 182         fiz3             = _mm_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm_setzero_pd();
 186         vvdwsum          = _mm_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_pd(ix0,jx0);
 204             dy00             = _mm_sub_pd(iy0,jy0);
 205             dz00             = _mm_sub_pd(iz0,jz0);
 206             dx10             = _mm_sub_pd(ix1,jx0);
 207             dy10             = _mm_sub_pd(iy1,jy0);
 208             dz10             = _mm_sub_pd(iz1,jz0);
 209             dx20             = _mm_sub_pd(ix2,jx0);
 210             dy20             = _mm_sub_pd(iy2,jy0);
 211             dz20             = _mm_sub_pd(iz2,jz0);
 212             dx30             = _mm_sub_pd(ix3,jx0);
 213             dy30             = _mm_sub_pd(iy3,jy0);
 214             dz30             = _mm_sub_pd(iz3,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 220             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 221
 222             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 223             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 224             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 225             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 226
 227             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 229             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235
 236             fjx0             = _mm_setzero_pd();
 237             fjy0             = _mm_setzero_pd();
 238             fjz0             = _mm_setzero_pd();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             r00              = _mm_mul_pd(rsq00,rinv00);
 245
 246             /* Compute parameters for interactions between i and j atoms */
 247             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 249
 250             /* Calculate table index by multiplying r with table scale and truncate to integer */
 251             rt               = _mm_mul_pd(r00,vftabscale);
 252             vfitab           = _mm_cvttpd_epi32(rt);
 253 #ifdef __XOP__
 254             vfeps            = _mm_frcz_pd(rt);
 255 #else
 256             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 257 #endif
 258             twovfeps         = _mm_add_pd(vfeps,vfeps);
 259             vfitab           = _mm_slli_epi32(vfitab,3);
 260
 261             /* CUBIC SPLINE TABLE DISPERSION */
 262             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 263             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 264             GMX_MM_TRANSPOSE2_PD(Y,F);
 265             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 266             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 267             GMX_MM_TRANSPOSE2_PD(G,H);
 268             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 269             VV               = _mm_macc_pd(vfeps,Fp,Y);
 270             vvdw6            = _mm_mul_pd(c6_00,VV);
 271             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 272             fvdw6            = _mm_mul_pd(c6_00,FF);
 273
 274             /* CUBIC SPLINE TABLE REPULSION */
 275             vfitab           = _mm_add_epi32(vfitab,ifour);
 276             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 277             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 278             GMX_MM_TRANSPOSE2_PD(Y,F);
 279             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 280             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 281             GMX_MM_TRANSPOSE2_PD(G,H);
 282             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 283             VV               = _mm_macc_pd(vfeps,Fp,Y);
 284             vvdw12           = _mm_mul_pd(c12_00,VV);
 285             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 286             fvdw12           = _mm_mul_pd(c12_00,FF);
 287             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 288             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 292
 293             fscal            = fvdw;
 294
 295             /* Update vectorial force */
 296             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 297             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 298             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 299
 300             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 301             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 302             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             if (gmx_mm_any_lt(rsq10,rcutoff2))
 309             {
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq10             = _mm_mul_pd(iq1,jq0);
 313
 314             /* REACTION-FIELD ELECTROSTATICS */
 315             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 316             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 317
 318             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velec            = _mm_and_pd(velec,cutoff_mask);
 322             velecsum         = _mm_add_pd(velecsum,velec);
 323
 324             fscal            = felec;
 325
 326             fscal            = _mm_and_pd(fscal,cutoff_mask);
 327
 328             /* Update vectorial force */
 329             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 330             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 331             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 332
 333             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 334             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 335             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 336
 337             }
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             if (gmx_mm_any_lt(rsq20,rcutoff2))
 344             {
 345
 346             /* Compute parameters for interactions between i and j atoms */
 347             qq20             = _mm_mul_pd(iq2,jq0);
 348
 349             /* REACTION-FIELD ELECTROSTATICS */
 350             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 351             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 352
 353             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velec            = _mm_and_pd(velec,cutoff_mask);
 357             velecsum         = _mm_add_pd(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             fscal            = _mm_and_pd(fscal,cutoff_mask);
 362
 363             /* Update vectorial force */
 364             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 365             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 366             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 367
 368             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 369             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 370             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 371
 372             }
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             if (gmx_mm_any_lt(rsq30,rcutoff2))
 379             {
 380
 381             /* Compute parameters for interactions between i and j atoms */
 382             qq30             = _mm_mul_pd(iq3,jq0);
 383
 384             /* REACTION-FIELD ELECTROSTATICS */
 385             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 386             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 387
 388             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velec            = _mm_and_pd(velec,cutoff_mask);
 392             velecsum         = _mm_add_pd(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             fscal            = _mm_and_pd(fscal,cutoff_mask);
 397
 398             /* Update vectorial force */
 399             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 400             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 401             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 402
 403             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 404             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 405             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 406
 407             }
 408
 409             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 410
 411             /* Inner loop uses 179 flops */
 412         }
 413
 414         if(jidx<j_index_end)
 415         {
 416
 417             jnrA             = jjnr[jidx];
 418             j_coord_offsetA  = DIM*jnrA;
 419
 420             /* load j atom coordinates */
 421             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 422                                               &jx0,&jy0,&jz0);
 423
 424             /* Calculate displacement vector */
 425             dx00             = _mm_sub_pd(ix0,jx0);
 426             dy00             = _mm_sub_pd(iy0,jy0);
 427             dz00             = _mm_sub_pd(iz0,jz0);
 428             dx10             = _mm_sub_pd(ix1,jx0);
 429             dy10             = _mm_sub_pd(iy1,jy0);
 430             dz10             = _mm_sub_pd(iz1,jz0);
 431             dx20             = _mm_sub_pd(ix2,jx0);
 432             dy20             = _mm_sub_pd(iy2,jy0);
 433             dz20             = _mm_sub_pd(iz2,jz0);
 434             dx30             = _mm_sub_pd(ix3,jx0);
 435             dy30             = _mm_sub_pd(iy3,jy0);
 436             dz30             = _mm_sub_pd(iz3,jz0);
 437
 438             /* Calculate squared distance and things based on it */
 439             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 440             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 441             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 442             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 443
 444             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 445             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 446             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 447             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 448
 449             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 450             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 451             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 452
 453             /* Load parameters for j particles */
 454             jq0              = _mm_load_sd(charge+jnrA+0);
 455             vdwjidx0A        = 2*vdwtype[jnrA+0];
 456
 457             fjx0             = _mm_setzero_pd();
 458             fjy0             = _mm_setzero_pd();
 459             fjz0             = _mm_setzero_pd();
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             r00              = _mm_mul_pd(rsq00,rinv00);
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 469
 470             /* Calculate table index by multiplying r with table scale and truncate to integer */
 471             rt               = _mm_mul_pd(r00,vftabscale);
 472             vfitab           = _mm_cvttpd_epi32(rt);
 473 #ifdef __XOP__
 474             vfeps            = _mm_frcz_pd(rt);
 475 #else
 476             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 477 #endif
 478             twovfeps         = _mm_add_pd(vfeps,vfeps);
 479             vfitab           = _mm_slli_epi32(vfitab,3);
 480
 481             /* CUBIC SPLINE TABLE DISPERSION */
 482             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 483             F                = _mm_setzero_pd();
 484             GMX_MM_TRANSPOSE2_PD(Y,F);
 485             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 486             H                = _mm_setzero_pd();
 487             GMX_MM_TRANSPOSE2_PD(G,H);
 488             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 489             VV               = _mm_macc_pd(vfeps,Fp,Y);
 490             vvdw6            = _mm_mul_pd(c6_00,VV);
 491             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 492             fvdw6            = _mm_mul_pd(c6_00,FF);
 493
 494             /* CUBIC SPLINE TABLE REPULSION */
 495             vfitab           = _mm_add_epi32(vfitab,ifour);
 496             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 497             F                = _mm_setzero_pd();
 498             GMX_MM_TRANSPOSE2_PD(Y,F);
 499             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 500             H                = _mm_setzero_pd();
 501             GMX_MM_TRANSPOSE2_PD(G,H);
 502             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 503             VV               = _mm_macc_pd(vfeps,Fp,Y);
 504             vvdw12           = _mm_mul_pd(c12_00,VV);
 505             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 506             fvdw12           = _mm_mul_pd(c12_00,FF);
 507             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 508             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 512             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 513
 514             fscal            = fvdw;
 515
 516             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 517
 518             /* Update vectorial force */
 519             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 520             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 521             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 522
 523             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 524             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 525             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm_any_lt(rsq10,rcutoff2))
 532             {
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq10             = _mm_mul_pd(iq1,jq0);
 536
 537             /* REACTION-FIELD ELECTROSTATICS */
 538             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 539             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 540
 541             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm_and_pd(velec,cutoff_mask);
 545             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 546             velecsum         = _mm_add_pd(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm_and_pd(fscal,cutoff_mask);
 551
 552             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 553
 554             /* Update vectorial force */
 555             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 556             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 557             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 558
 559             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 560             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 561             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 562
 563             }
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             if (gmx_mm_any_lt(rsq20,rcutoff2))
 570             {
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq20             = _mm_mul_pd(iq2,jq0);
 574
 575             /* REACTION-FIELD ELECTROSTATICS */
 576             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 577             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 578
 579             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm_and_pd(velec,cutoff_mask);
 583             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 584             velecsum         = _mm_add_pd(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm_and_pd(fscal,cutoff_mask);
 589
 590             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 591
 592             /* Update vectorial force */
 593             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 594             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 595             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 596
 597             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 598             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 599             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 600
 601             }
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             if (gmx_mm_any_lt(rsq30,rcutoff2))
 608             {
 609
 610             /* Compute parameters for interactions between i and j atoms */
 611             qq30             = _mm_mul_pd(iq3,jq0);
 612
 613             /* REACTION-FIELD ELECTROSTATICS */
 614             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 615             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 616
 617             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 618
 619             /* Update potential sum for this i atom from the interaction with this j atom. */
 620             velec            = _mm_and_pd(velec,cutoff_mask);
 621             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 622             velecsum         = _mm_add_pd(velecsum,velec);
 623
 624             fscal            = felec;
 625
 626             fscal            = _mm_and_pd(fscal,cutoff_mask);
 627
 628             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 629
 630             /* Update vectorial force */
 631             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 632             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 633             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 634
 635             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 636             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 637             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 638
 639             }
 640
 641             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 642
 643             /* Inner loop uses 179 flops */
 644         }
 645
 646         /* End of innermost loop */
 647
 648         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 649                                               f+i_coord_offset,fshift+i_shift_offset);
 650
 651         ggid                        = gid[iidx];
 652         /* Update potential energies */
 653         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 654         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 655
 656         /* Increment number of inner iterations */
 657         inneriter                  += j_index_end - j_index_start;
 658
 659         /* Outer loop uses 26 flops */
 660     }
 661
 662     /* Increment number of outer iterations */
 663     outeriter        += nri;
 664
 665     /* Update outer/inner flops */
 666
 667     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*179);
 668 }
 669 /*
 670  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 671  * Electrostatics interaction: ReactionField
 672  * VdW interaction:            CubicSplineTable
 673  * Geometry:                   Water4-Particle
 674  * Calculate force/pot:        Force
 675  */
 676 void
 677 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 678                     (t_nblist                    * gmx_restrict       nlist,
 679                      rvec                        * gmx_restrict          xx,
 680                      rvec                        * gmx_restrict          ff,
 681                      t_forcerec                  * gmx_restrict          fr,
 682                      t_mdatoms                   * gmx_restrict     mdatoms,
 683                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 684                      t_nrnb                      * gmx_restrict        nrnb)
 685 {
 686     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 687      * just 0 for non-waters.
 688      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 689      * jnr indices corresponding to data put in the four positions in the SIMD register.
 690      */
 691     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 692     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 693     int              jnrA,jnrB;
 694     int              j_coord_offsetA,j_coord_offsetB;
 695     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 696     real             rcutoff_scalar;
 697     real             *shiftvec,*fshift,*x,*f;
 698     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 699     int              vdwioffset0;
 700     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 701     int              vdwioffset1;
 702     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 703     int              vdwioffset2;
 704     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 705     int              vdwioffset3;
 706     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 707     int              vdwjidx0A,vdwjidx0B;
 708     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 709     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 710     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 711     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 712     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 713     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 714     real             *charge;
 715     int              nvdwtype;
 716     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 717     int              *vdwtype;
 718     real             *vdwparam;
 719     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 720     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 721     __m128i          vfitab;
 722     __m128i          ifour       = _mm_set1_epi32(4);
 723     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 724     real             *vftab;
 725     __m128d          dummy_mask,cutoff_mask;
 726     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 727     __m128d          one     = _mm_set1_pd(1.0);
 728     __m128d          two     = _mm_set1_pd(2.0);
 729     x                = xx[0];
 730     f                = ff[0];
 731
 732     nri              = nlist->nri;
 733     iinr             = nlist->iinr;
 734     jindex           = nlist->jindex;
 735     jjnr             = nlist->jjnr;
 736     shiftidx         = nlist->shift;
 737     gid              = nlist->gid;
 738     shiftvec         = fr->shift_vec[0];
 739     fshift           = fr->fshift[0];
 740     facel            = _mm_set1_pd(fr->epsfac);
 741     charge           = mdatoms->chargeA;
 742     krf              = _mm_set1_pd(fr->ic->k_rf);
 743     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 744     crf              = _mm_set1_pd(fr->ic->c_rf);
 745     nvdwtype         = fr->ntype;
 746     vdwparam         = fr->nbfp;
 747     vdwtype          = mdatoms->typeA;
 748
 749     vftab            = kernel_data->table_vdw->data;
 750     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 751
 752     /* Setup water-specific parameters */
 753     inr              = nlist->iinr[0];
 754     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 755     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 756     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 757     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 758
 759     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 760     rcutoff_scalar   = fr->rcoulomb;
 761     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 762     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 763
 764     /* Avoid stupid compiler warnings */
 765     jnrA = jnrB = 0;
 766     j_coord_offsetA = 0;
 767     j_coord_offsetB = 0;
 768
 769     outeriter        = 0;
 770     inneriter        = 0;
 771
 772     /* Start outer loop over neighborlists */
 773     for(iidx=0; iidx<nri; iidx++)
 774     {
 775         /* Load shift vector for this list */
 776         i_shift_offset   = DIM*shiftidx[iidx];
 777
 778         /* Load limits for loop over neighbors */
 779         j_index_start    = jindex[iidx];
 780         j_index_end      = jindex[iidx+1];
 781
 782         /* Get outer coordinate index */
 783         inr              = iinr[iidx];
 784         i_coord_offset   = DIM*inr;
 785
 786         /* Load i particle coords and add shift vector */
 787         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 788                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 789
 790         fix0             = _mm_setzero_pd();
 791         fiy0             = _mm_setzero_pd();
 792         fiz0             = _mm_setzero_pd();
 793         fix1             = _mm_setzero_pd();
 794         fiy1             = _mm_setzero_pd();
 795         fiz1             = _mm_setzero_pd();
 796         fix2             = _mm_setzero_pd();
 797         fiy2             = _mm_setzero_pd();
 798         fiz2             = _mm_setzero_pd();
 799         fix3             = _mm_setzero_pd();
 800         fiy3             = _mm_setzero_pd();
 801         fiz3             = _mm_setzero_pd();
 802
 803         /* Start inner kernel loop */
 804         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 805         {
 806
 807             /* Get j neighbor index, and coordinate index */
 808             jnrA             = jjnr[jidx];
 809             jnrB             = jjnr[jidx+1];
 810             j_coord_offsetA  = DIM*jnrA;
 811             j_coord_offsetB  = DIM*jnrB;
 812
 813             /* load j atom coordinates */
 814             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 815                                               &jx0,&jy0,&jz0);
 816
 817             /* Calculate displacement vector */
 818             dx00             = _mm_sub_pd(ix0,jx0);
 819             dy00             = _mm_sub_pd(iy0,jy0);
 820             dz00             = _mm_sub_pd(iz0,jz0);
 821             dx10             = _mm_sub_pd(ix1,jx0);
 822             dy10             = _mm_sub_pd(iy1,jy0);
 823             dz10             = _mm_sub_pd(iz1,jz0);
 824             dx20             = _mm_sub_pd(ix2,jx0);
 825             dy20             = _mm_sub_pd(iy2,jy0);
 826             dz20             = _mm_sub_pd(iz2,jz0);
 827             dx30             = _mm_sub_pd(ix3,jx0);
 828             dy30             = _mm_sub_pd(iy3,jy0);
 829             dz30             = _mm_sub_pd(iz3,jz0);
 830
 831             /* Calculate squared distance and things based on it */
 832             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 833             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 834             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 835             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 836
 837             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 838             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 839             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 840             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 841
 842             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 843             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 844             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 845
 846             /* Load parameters for j particles */
 847             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 848             vdwjidx0A        = 2*vdwtype[jnrA+0];
 849             vdwjidx0B        = 2*vdwtype[jnrB+0];
 850
 851             fjx0             = _mm_setzero_pd();
 852             fjy0             = _mm_setzero_pd();
 853             fjz0             = _mm_setzero_pd();
 854
 855             /**************************
 856              * CALCULATE INTERACTIONS *
 857              **************************/
 858
 859             r00              = _mm_mul_pd(rsq00,rinv00);
 860
 861             /* Compute parameters for interactions between i and j atoms */
 862             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 863                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 864
 865             /* Calculate table index by multiplying r with table scale and truncate to integer */
 866             rt               = _mm_mul_pd(r00,vftabscale);
 867             vfitab           = _mm_cvttpd_epi32(rt);
 868 #ifdef __XOP__
 869             vfeps            = _mm_frcz_pd(rt);
 870 #else
 871             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 872 #endif
 873             twovfeps         = _mm_add_pd(vfeps,vfeps);
 874             vfitab           = _mm_slli_epi32(vfitab,3);
 875
 876             /* CUBIC SPLINE TABLE DISPERSION */
 877             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 878             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 879             GMX_MM_TRANSPOSE2_PD(Y,F);
 880             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 881             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 882             GMX_MM_TRANSPOSE2_PD(G,H);
 883             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 884             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 885             fvdw6            = _mm_mul_pd(c6_00,FF);
 886
 887             /* CUBIC SPLINE TABLE REPULSION */
 888             vfitab           = _mm_add_epi32(vfitab,ifour);
 889             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 890             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 891             GMX_MM_TRANSPOSE2_PD(Y,F);
 892             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 893             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 894             GMX_MM_TRANSPOSE2_PD(G,H);
 895             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 896             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 897             fvdw12           = _mm_mul_pd(c12_00,FF);
 898             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 899
 900             fscal            = fvdw;
 901
 902             /* Update vectorial force */
 903             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 904             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 905             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 906
 907             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 908             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 909             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             if (gmx_mm_any_lt(rsq10,rcutoff2))
 916             {
 917
 918             /* Compute parameters for interactions between i and j atoms */
 919             qq10             = _mm_mul_pd(iq1,jq0);
 920
 921             /* REACTION-FIELD ELECTROSTATICS */
 922             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 923
 924             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 925
 926             fscal            = felec;
 927
 928             fscal            = _mm_and_pd(fscal,cutoff_mask);
 929
 930             /* Update vectorial force */
 931             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 932             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 933             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 934
 935             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 936             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 937             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 938
 939             }
 940
 941             /**************************
 942              * CALCULATE INTERACTIONS *
 943              **************************/
 944
 945             if (gmx_mm_any_lt(rsq20,rcutoff2))
 946             {
 947
 948             /* Compute parameters for interactions between i and j atoms */
 949             qq20             = _mm_mul_pd(iq2,jq0);
 950
 951             /* REACTION-FIELD ELECTROSTATICS */
 952             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 953
 954             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 955
 956             fscal            = felec;
 957
 958             fscal            = _mm_and_pd(fscal,cutoff_mask);
 959
 960             /* Update vectorial force */
 961             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 962             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 963             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 964
 965             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 966             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 967             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 968
 969             }
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             if (gmx_mm_any_lt(rsq30,rcutoff2))
 976             {
 977
 978             /* Compute parameters for interactions between i and j atoms */
 979             qq30             = _mm_mul_pd(iq3,jq0);
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 983
 984             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm_and_pd(fscal,cutoff_mask);
 989
 990             /* Update vectorial force */
 991             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 992             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 993             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 994
 995             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 996             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 997             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 998
 999             }
1000
1001             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1002
1003             /* Inner loop uses 153 flops */
1004         }
1005
1006         if(jidx<j_index_end)
1007         {
1008
1009             jnrA             = jjnr[jidx];
1010             j_coord_offsetA  = DIM*jnrA;
1011
1012             /* load j atom coordinates */
1013             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1014                                               &jx0,&jy0,&jz0);
1015
1016             /* Calculate displacement vector */
1017             dx00             = _mm_sub_pd(ix0,jx0);
1018             dy00             = _mm_sub_pd(iy0,jy0);
1019             dz00             = _mm_sub_pd(iz0,jz0);
1020             dx10             = _mm_sub_pd(ix1,jx0);
1021             dy10             = _mm_sub_pd(iy1,jy0);
1022             dz10             = _mm_sub_pd(iz1,jz0);
1023             dx20             = _mm_sub_pd(ix2,jx0);
1024             dy20             = _mm_sub_pd(iy2,jy0);
1025             dz20             = _mm_sub_pd(iz2,jz0);
1026             dx30             = _mm_sub_pd(ix3,jx0);
1027             dy30             = _mm_sub_pd(iy3,jy0);
1028             dz30             = _mm_sub_pd(iz3,jz0);
1029
1030             /* Calculate squared distance and things based on it */
1031             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1032             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1033             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1034             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1035
1036             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1037             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1038             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1039             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1040
1041             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1042             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1043             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1044
1045             /* Load parameters for j particles */
1046             jq0              = _mm_load_sd(charge+jnrA+0);
1047             vdwjidx0A        = 2*vdwtype[jnrA+0];
1048
1049             fjx0             = _mm_setzero_pd();
1050             fjy0             = _mm_setzero_pd();
1051             fjz0             = _mm_setzero_pd();
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             r00              = _mm_mul_pd(rsq00,rinv00);
1058
1059             /* Compute parameters for interactions between i and j atoms */
1060             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1061
1062             /* Calculate table index by multiplying r with table scale and truncate to integer */
1063             rt               = _mm_mul_pd(r00,vftabscale);
1064             vfitab           = _mm_cvttpd_epi32(rt);
1065 #ifdef __XOP__
1066             vfeps            = _mm_frcz_pd(rt);
1067 #else
1068             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1069 #endif
1070             twovfeps         = _mm_add_pd(vfeps,vfeps);
1071             vfitab           = _mm_slli_epi32(vfitab,3);
1072
1073             /* CUBIC SPLINE TABLE DISPERSION */
1074             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1075             F                = _mm_setzero_pd();
1076             GMX_MM_TRANSPOSE2_PD(Y,F);
1077             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1078             H                = _mm_setzero_pd();
1079             GMX_MM_TRANSPOSE2_PD(G,H);
1080             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1081             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1082             fvdw6            = _mm_mul_pd(c6_00,FF);
1083
1084             /* CUBIC SPLINE TABLE REPULSION */
1085             vfitab           = _mm_add_epi32(vfitab,ifour);
1086             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1087             F                = _mm_setzero_pd();
1088             GMX_MM_TRANSPOSE2_PD(Y,F);
1089             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1090             H                = _mm_setzero_pd();
1091             GMX_MM_TRANSPOSE2_PD(G,H);
1092             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1093             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1094             fvdw12           = _mm_mul_pd(c12_00,FF);
1095             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1096
1097             fscal            = fvdw;
1098
1099             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1100
1101             /* Update vectorial force */
1102             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1103             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1104             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1105
1106             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1107             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1108             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             if (gmx_mm_any_lt(rsq10,rcutoff2))
1115             {
1116
1117             /* Compute parameters for interactions between i and j atoms */
1118             qq10             = _mm_mul_pd(iq1,jq0);
1119
1120             /* REACTION-FIELD ELECTROSTATICS */
1121             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1122
1123             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1124
1125             fscal            = felec;
1126
1127             fscal            = _mm_and_pd(fscal,cutoff_mask);
1128
1129             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1130
1131             /* Update vectorial force */
1132             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1133             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1134             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1135
1136             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1137             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1138             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1139
1140             }
1141
1142             /**************************
1143              * CALCULATE INTERACTIONS *
1144              **************************/
1145
1146             if (gmx_mm_any_lt(rsq20,rcutoff2))
1147             {
1148
1149             /* Compute parameters for interactions between i and j atoms */
1150             qq20             = _mm_mul_pd(iq2,jq0);
1151
1152             /* REACTION-FIELD ELECTROSTATICS */
1153             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1154
1155             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1156
1157             fscal            = felec;
1158
1159             fscal            = _mm_and_pd(fscal,cutoff_mask);
1160
1161             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1162
1163             /* Update vectorial force */
1164             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1165             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1166             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1167
1168             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1169             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1170             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1171
1172             }
1173
1174             /**************************
1175              * CALCULATE INTERACTIONS *
1176              **************************/
1177
1178             if (gmx_mm_any_lt(rsq30,rcutoff2))
1179             {
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq30             = _mm_mul_pd(iq3,jq0);
1183
1184             /* REACTION-FIELD ELECTROSTATICS */
1185             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
1186
1187             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1188
1189             fscal            = felec;
1190
1191             fscal            = _mm_and_pd(fscal,cutoff_mask);
1192
1193             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1194
1195             /* Update vectorial force */
1196             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1197             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1198             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1199
1200             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1201             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1202             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1203
1204             }
1205
1206             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1207
1208             /* Inner loop uses 153 flops */
1209         }
1210
1211         /* End of innermost loop */
1212
1213         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1214                                               f+i_coord_offset,fshift+i_shift_offset);
1215
1216         /* Increment number of inner iterations */
1217         inneriter                  += j_index_end - j_index_start;
1218
1219         /* Outer loop uses 24 flops */
1220     }
1221
1222     /* Increment number of outer iterations */
1223     outeriter        += nri;
1224
1225     /* Update outer/inner flops */
1226
1227     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*153);
1228 }