src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_sse2_single.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 sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_sse2_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128i          vfitab;
 108     __m128i          ifour       = _mm_set1_epi32(4);
 109     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 110     real             *vftab;
 111     __m128           dummy_mask,cutoff_mask;
 112     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 113     __m128           one     = _mm_set1_ps(1.0);
 114     __m128           two     = _mm_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     krf              = _mm_set1_ps(fr->ic->k_rf);
 129     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 130     crf              = _mm_set1_ps(fr->ic->c_rf);
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     vftab            = kernel_data->table_vdw->data;
 136     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 137
 138     /* Setup water-specific parameters */
 139     inr              = nlist->iinr[0];
 140     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 141     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 142     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 143     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm_setzero_ps();
 179         fiy0             = _mm_setzero_ps();
 180         fiz0             = _mm_setzero_ps();
 181         fix1             = _mm_setzero_ps();
 182         fiy1             = _mm_setzero_ps();
 183         fiz1             = _mm_setzero_ps();
 184         fix2             = _mm_setzero_ps();
 185         fiy2             = _mm_setzero_ps();
 186         fiz2             = _mm_setzero_ps();
 187         fix3             = _mm_setzero_ps();
 188         fiy3             = _mm_setzero_ps();
 189         fiz3             = _mm_setzero_ps();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_ps();
 193         vvdwsum          = _mm_setzero_ps();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                               x+j_coord_offsetC,x+j_coord_offsetD,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm_sub_ps(ix0,jx0);
 216             dy00             = _mm_sub_ps(iy0,jy0);
 217             dz00             = _mm_sub_ps(iz0,jz0);
 218             dx10             = _mm_sub_ps(ix1,jx0);
 219             dy10             = _mm_sub_ps(iy1,jy0);
 220             dz10             = _mm_sub_ps(iz1,jz0);
 221             dx20             = _mm_sub_ps(ix2,jx0);
 222             dy20             = _mm_sub_ps(iy2,jy0);
 223             dz20             = _mm_sub_ps(iz2,jz0);
 224             dx30             = _mm_sub_ps(ix3,jx0);
 225             dy30             = _mm_sub_ps(iy3,jy0);
 226             dz30             = _mm_sub_ps(iz3,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 230             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 231             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 232             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 233
 234             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 235             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 236             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 237             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 238
 239             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 240             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 241             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 245                                                               charge+jnrC+0,charge+jnrD+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248             vdwjidx0C        = 2*vdwtype[jnrC+0];
 249             vdwjidx0D        = 2*vdwtype[jnrD+0];
 250
 251             fjx0             = _mm_setzero_ps();
 252             fjy0             = _mm_setzero_ps();
 253             fjz0             = _mm_setzero_ps();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             r00              = _mm_mul_ps(rsq00,rinv00);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 263                                          vdwparam+vdwioffset0+vdwjidx0B,
 264                                          vdwparam+vdwioffset0+vdwjidx0C,
 265                                          vdwparam+vdwioffset0+vdwjidx0D,
 266                                          &c6_00,&c12_00);
 267
 268             /* Calculate table index by multiplying r with table scale and truncate to integer */
 269             rt               = _mm_mul_ps(r00,vftabscale);
 270             vfitab           = _mm_cvttps_epi32(rt);
 271             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 272             vfitab           = _mm_slli_epi32(vfitab,3);
 273
 274             /* CUBIC SPLINE TABLE DISPERSION */
 275             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 277             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 278             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 279             _MM_TRANSPOSE4_PS(Y,F,G,H);
 280             Heps             = _mm_mul_ps(vfeps,H);
 281             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 282             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 283             vvdw6            = _mm_mul_ps(c6_00,VV);
 284             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 285             fvdw6            = _mm_mul_ps(c6_00,FF);
 286
 287             /* CUBIC SPLINE TABLE REPULSION */
 288             vfitab           = _mm_add_epi32(vfitab,ifour);
 289             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 290             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 291             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 292             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 293             _MM_TRANSPOSE4_PS(Y,F,G,H);
 294             Heps             = _mm_mul_ps(vfeps,H);
 295             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 296             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 297             vvdw12           = _mm_mul_ps(c12_00,VV);
 298             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 299             fvdw12           = _mm_mul_ps(c12_00,FF);
 300             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 301             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 305
 306             fscal            = fvdw;
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm_mul_ps(fscal,dx00);
 310             ty               = _mm_mul_ps(fscal,dy00);
 311             tz               = _mm_mul_ps(fscal,dz00);
 312
 313             /* Update vectorial force */
 314             fix0             = _mm_add_ps(fix0,tx);
 315             fiy0             = _mm_add_ps(fiy0,ty);
 316             fiz0             = _mm_add_ps(fiz0,tz);
 317
 318             fjx0             = _mm_add_ps(fjx0,tx);
 319             fjy0             = _mm_add_ps(fjy0,ty);
 320             fjz0             = _mm_add_ps(fjz0,tz);
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm_mul_ps(iq1,jq0);
 328
 329             /* REACTION-FIELD ELECTROSTATICS */
 330             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 331             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velecsum         = _mm_add_ps(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm_mul_ps(fscal,dx10);
 340             ty               = _mm_mul_ps(fscal,dy10);
 341             tz               = _mm_mul_ps(fscal,dz10);
 342
 343             /* Update vectorial force */
 344             fix1             = _mm_add_ps(fix1,tx);
 345             fiy1             = _mm_add_ps(fiy1,ty);
 346             fiz1             = _mm_add_ps(fiz1,tz);
 347
 348             fjx0             = _mm_add_ps(fjx0,tx);
 349             fjy0             = _mm_add_ps(fjy0,ty);
 350             fjz0             = _mm_add_ps(fjz0,tz);
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq20             = _mm_mul_ps(iq2,jq0);
 358
 359             /* REACTION-FIELD ELECTROSTATICS */
 360             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 361             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velecsum         = _mm_add_ps(velecsum,velec);
 365
 366             fscal            = felec;
 367
 368             /* Calculate temporary vectorial force */
 369             tx               = _mm_mul_ps(fscal,dx20);
 370             ty               = _mm_mul_ps(fscal,dy20);
 371             tz               = _mm_mul_ps(fscal,dz20);
 372
 373             /* Update vectorial force */
 374             fix2             = _mm_add_ps(fix2,tx);
 375             fiy2             = _mm_add_ps(fiy2,ty);
 376             fiz2             = _mm_add_ps(fiz2,tz);
 377
 378             fjx0             = _mm_add_ps(fjx0,tx);
 379             fjy0             = _mm_add_ps(fjy0,ty);
 380             fjz0             = _mm_add_ps(fjz0,tz);
 381
 382             /**************************
 383              * CALCULATE INTERACTIONS *
 384              **************************/
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq30             = _mm_mul_ps(iq3,jq0);
 388
 389             /* REACTION-FIELD ELECTROSTATICS */
 390             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 391             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx30);
 400             ty               = _mm_mul_ps(fscal,dy30);
 401             tz               = _mm_mul_ps(fscal,dz30);
 402
 403             /* Update vectorial force */
 404             fix3             = _mm_add_ps(fix3,tx);
 405             fiy3             = _mm_add_ps(fiy3,ty);
 406             fiz3             = _mm_add_ps(fiz3,tz);
 407
 408             fjx0             = _mm_add_ps(fjx0,tx);
 409             fjy0             = _mm_add_ps(fjy0,ty);
 410             fjz0             = _mm_add_ps(fjz0,tz);
 411
 412             fjptrA             = f+j_coord_offsetA;
 413             fjptrB             = f+j_coord_offsetB;
 414             fjptrC             = f+j_coord_offsetC;
 415             fjptrD             = f+j_coord_offsetD;
 416
 417             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 418
 419             /* Inner loop uses 152 flops */
 420         }
 421
 422         if(jidx<j_index_end)
 423         {
 424
 425             /* Get j neighbor index, and coordinate index */
 426             jnrlistA         = jjnr[jidx];
 427             jnrlistB         = jjnr[jidx+1];
 428             jnrlistC         = jjnr[jidx+2];
 429             jnrlistD         = jjnr[jidx+3];
 430             /* Sign of each element will be negative for non-real atoms.
 431              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 432              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 433              */
 434             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 435             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 436             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 437             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 438             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 439             j_coord_offsetA  = DIM*jnrA;
 440             j_coord_offsetB  = DIM*jnrB;
 441             j_coord_offsetC  = DIM*jnrC;
 442             j_coord_offsetD  = DIM*jnrD;
 443
 444             /* load j atom coordinates */
 445             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 446                                               x+j_coord_offsetC,x+j_coord_offsetD,
 447                                               &jx0,&jy0,&jz0);
 448
 449             /* Calculate displacement vector */
 450             dx00             = _mm_sub_ps(ix0,jx0);
 451             dy00             = _mm_sub_ps(iy0,jy0);
 452             dz00             = _mm_sub_ps(iz0,jz0);
 453             dx10             = _mm_sub_ps(ix1,jx0);
 454             dy10             = _mm_sub_ps(iy1,jy0);
 455             dz10             = _mm_sub_ps(iz1,jz0);
 456             dx20             = _mm_sub_ps(ix2,jx0);
 457             dy20             = _mm_sub_ps(iy2,jy0);
 458             dz20             = _mm_sub_ps(iz2,jz0);
 459             dx30             = _mm_sub_ps(ix3,jx0);
 460             dy30             = _mm_sub_ps(iy3,jy0);
 461             dz30             = _mm_sub_ps(iz3,jz0);
 462
 463             /* Calculate squared distance and things based on it */
 464             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 465             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 466             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 467             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 468
 469             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 470             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 471             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 472             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 473
 474             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 475             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 476             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 477
 478             /* Load parameters for j particles */
 479             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 480                                                               charge+jnrC+0,charge+jnrD+0);
 481             vdwjidx0A        = 2*vdwtype[jnrA+0];
 482             vdwjidx0B        = 2*vdwtype[jnrB+0];
 483             vdwjidx0C        = 2*vdwtype[jnrC+0];
 484             vdwjidx0D        = 2*vdwtype[jnrD+0];
 485
 486             fjx0             = _mm_setzero_ps();
 487             fjy0             = _mm_setzero_ps();
 488             fjz0             = _mm_setzero_ps();
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             r00              = _mm_mul_ps(rsq00,rinv00);
 495             r00              = _mm_andnot_ps(dummy_mask,r00);
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 499                                          vdwparam+vdwioffset0+vdwjidx0B,
 500                                          vdwparam+vdwioffset0+vdwjidx0C,
 501                                          vdwparam+vdwioffset0+vdwjidx0D,
 502                                          &c6_00,&c12_00);
 503
 504             /* Calculate table index by multiplying r with table scale and truncate to integer */
 505             rt               = _mm_mul_ps(r00,vftabscale);
 506             vfitab           = _mm_cvttps_epi32(rt);
 507             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 508             vfitab           = _mm_slli_epi32(vfitab,3);
 509
 510             /* CUBIC SPLINE TABLE DISPERSION */
 511             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 512             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 513             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 514             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 515             _MM_TRANSPOSE4_PS(Y,F,G,H);
 516             Heps             = _mm_mul_ps(vfeps,H);
 517             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 518             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 519             vvdw6            = _mm_mul_ps(c6_00,VV);
 520             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 521             fvdw6            = _mm_mul_ps(c6_00,FF);
 522
 523             /* CUBIC SPLINE TABLE REPULSION */
 524             vfitab           = _mm_add_epi32(vfitab,ifour);
 525             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 526             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 527             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 528             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 529             _MM_TRANSPOSE4_PS(Y,F,G,H);
 530             Heps             = _mm_mul_ps(vfeps,H);
 531             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 532             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 533             vvdw12           = _mm_mul_ps(c12_00,VV);
 534             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 535             fvdw12           = _mm_mul_ps(c12_00,FF);
 536             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 537             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 538
 539             /* Update potential sum for this i atom from the interaction with this j atom. */
 540             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 541             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 542
 543             fscal            = fvdw;
 544
 545             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 546
 547             /* Calculate temporary vectorial force */
 548             tx               = _mm_mul_ps(fscal,dx00);
 549             ty               = _mm_mul_ps(fscal,dy00);
 550             tz               = _mm_mul_ps(fscal,dz00);
 551
 552             /* Update vectorial force */
 553             fix0             = _mm_add_ps(fix0,tx);
 554             fiy0             = _mm_add_ps(fiy0,ty);
 555             fiz0             = _mm_add_ps(fiz0,tz);
 556
 557             fjx0             = _mm_add_ps(fjx0,tx);
 558             fjy0             = _mm_add_ps(fjy0,ty);
 559             fjz0             = _mm_add_ps(fjz0,tz);
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             qq10             = _mm_mul_ps(iq1,jq0);
 567
 568             /* REACTION-FIELD ELECTROSTATICS */
 569             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 570             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 571
 572             /* Update potential sum for this i atom from the interaction with this j atom. */
 573             velec            = _mm_andnot_ps(dummy_mask,velec);
 574             velecsum         = _mm_add_ps(velecsum,velec);
 575
 576             fscal            = felec;
 577
 578             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm_mul_ps(fscal,dx10);
 582             ty               = _mm_mul_ps(fscal,dy10);
 583             tz               = _mm_mul_ps(fscal,dz10);
 584
 585             /* Update vectorial force */
 586             fix1             = _mm_add_ps(fix1,tx);
 587             fiy1             = _mm_add_ps(fiy1,ty);
 588             fiz1             = _mm_add_ps(fiz1,tz);
 589
 590             fjx0             = _mm_add_ps(fjx0,tx);
 591             fjy0             = _mm_add_ps(fjy0,ty);
 592             fjz0             = _mm_add_ps(fjz0,tz);
 593
 594             /**************************
 595              * CALCULATE INTERACTIONS *
 596              **************************/
 597
 598             /* Compute parameters for interactions between i and j atoms */
 599             qq20             = _mm_mul_ps(iq2,jq0);
 600
 601             /* REACTION-FIELD ELECTROSTATICS */
 602             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 603             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_andnot_ps(dummy_mask,velec);
 607             velecsum         = _mm_add_ps(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_ps(fscal,dx20);
 615             ty               = _mm_mul_ps(fscal,dy20);
 616             tz               = _mm_mul_ps(fscal,dz20);
 617
 618             /* Update vectorial force */
 619             fix2             = _mm_add_ps(fix2,tx);
 620             fiy2             = _mm_add_ps(fiy2,ty);
 621             fiz2             = _mm_add_ps(fiz2,tz);
 622
 623             fjx0             = _mm_add_ps(fjx0,tx);
 624             fjy0             = _mm_add_ps(fjy0,ty);
 625             fjz0             = _mm_add_ps(fjz0,tz);
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             /* Compute parameters for interactions between i and j atoms */
 632             qq30             = _mm_mul_ps(iq3,jq0);
 633
 634             /* REACTION-FIELD ELECTROSTATICS */
 635             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 636             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 637
 638             /* Update potential sum for this i atom from the interaction with this j atom. */
 639             velec            = _mm_andnot_ps(dummy_mask,velec);
 640             velecsum         = _mm_add_ps(velecsum,velec);
 641
 642             fscal            = felec;
 643
 644             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 645
 646             /* Calculate temporary vectorial force */
 647             tx               = _mm_mul_ps(fscal,dx30);
 648             ty               = _mm_mul_ps(fscal,dy30);
 649             tz               = _mm_mul_ps(fscal,dz30);
 650
 651             /* Update vectorial force */
 652             fix3             = _mm_add_ps(fix3,tx);
 653             fiy3             = _mm_add_ps(fiy3,ty);
 654             fiz3             = _mm_add_ps(fiz3,tz);
 655
 656             fjx0             = _mm_add_ps(fjx0,tx);
 657             fjy0             = _mm_add_ps(fjy0,ty);
 658             fjz0             = _mm_add_ps(fjz0,tz);
 659
 660             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 661             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 662             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 663             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 664
 665             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 666
 667             /* Inner loop uses 153 flops */
 668         }
 669
 670         /* End of innermost loop */
 671
 672         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 673                                               f+i_coord_offset,fshift+i_shift_offset);
 674
 675         ggid                        = gid[iidx];
 676         /* Update potential energies */
 677         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 678         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 679
 680         /* Increment number of inner iterations */
 681         inneriter                  += j_index_end - j_index_start;
 682
 683         /* Outer loop uses 26 flops */
 684     }
 685
 686     /* Increment number of outer iterations */
 687     outeriter        += nri;
 688
 689     /* Update outer/inner flops */
 690
 691     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*153);
 692 }
 693 /*
 694  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
 695  * Electrostatics interaction: ReactionField
 696  * VdW interaction:            CubicSplineTable
 697  * Geometry:                   Water4-Particle
 698  * Calculate force/pot:        Force
 699  */
 700 void
 701 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
 702                     (t_nblist                    * gmx_restrict       nlist,
 703                      rvec                        * gmx_restrict          xx,
 704                      rvec                        * gmx_restrict          ff,
 705                      t_forcerec                  * gmx_restrict          fr,
 706                      t_mdatoms                   * gmx_restrict     mdatoms,
 707                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 708                      t_nrnb                      * gmx_restrict        nrnb)
 709 {
 710     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 711      * just 0 for non-waters.
 712      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 713      * jnr indices corresponding to data put in the four positions in the SIMD register.
 714      */
 715     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 716     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 717     int              jnrA,jnrB,jnrC,jnrD;
 718     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 719     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 720     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 721     real             rcutoff_scalar;
 722     real             *shiftvec,*fshift,*x,*f;
 723     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 724     real             scratch[4*DIM];
 725     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 726     int              vdwioffset0;
 727     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 728     int              vdwioffset1;
 729     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 730     int              vdwioffset2;
 731     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 732     int              vdwioffset3;
 733     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 734     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 735     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 736     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 737     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 738     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 739     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 740     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 741     real             *charge;
 742     int              nvdwtype;
 743     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 744     int              *vdwtype;
 745     real             *vdwparam;
 746     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 747     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 748     __m128i          vfitab;
 749     __m128i          ifour       = _mm_set1_epi32(4);
 750     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 751     real             *vftab;
 752     __m128           dummy_mask,cutoff_mask;
 753     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 754     __m128           one     = _mm_set1_ps(1.0);
 755     __m128           two     = _mm_set1_ps(2.0);
 756     x                = xx[0];
 757     f                = ff[0];
 758
 759     nri              = nlist->nri;
 760     iinr             = nlist->iinr;
 761     jindex           = nlist->jindex;
 762     jjnr             = nlist->jjnr;
 763     shiftidx         = nlist->shift;
 764     gid              = nlist->gid;
 765     shiftvec         = fr->shift_vec[0];
 766     fshift           = fr->fshift[0];
 767     facel            = _mm_set1_ps(fr->epsfac);
 768     charge           = mdatoms->chargeA;
 769     krf              = _mm_set1_ps(fr->ic->k_rf);
 770     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 771     crf              = _mm_set1_ps(fr->ic->c_rf);
 772     nvdwtype         = fr->ntype;
 773     vdwparam         = fr->nbfp;
 774     vdwtype          = mdatoms->typeA;
 775
 776     vftab            = kernel_data->table_vdw->data;
 777     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 778
 779     /* Setup water-specific parameters */
 780     inr              = nlist->iinr[0];
 781     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 782     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 783     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 784     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 785
 786     /* Avoid stupid compiler warnings */
 787     jnrA = jnrB = jnrC = jnrD = 0;
 788     j_coord_offsetA = 0;
 789     j_coord_offsetB = 0;
 790     j_coord_offsetC = 0;
 791     j_coord_offsetD = 0;
 792
 793     outeriter        = 0;
 794     inneriter        = 0;
 795
 796     for(iidx=0;iidx<4*DIM;iidx++)
 797     {
 798         scratch[iidx] = 0.0;
 799     }
 800
 801     /* Start outer loop over neighborlists */
 802     for(iidx=0; iidx<nri; iidx++)
 803     {
 804         /* Load shift vector for this list */
 805         i_shift_offset   = DIM*shiftidx[iidx];
 806
 807         /* Load limits for loop over neighbors */
 808         j_index_start    = jindex[iidx];
 809         j_index_end      = jindex[iidx+1];
 810
 811         /* Get outer coordinate index */
 812         inr              = iinr[iidx];
 813         i_coord_offset   = DIM*inr;
 814
 815         /* Load i particle coords and add shift vector */
 816         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 817                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 818
 819         fix0             = _mm_setzero_ps();
 820         fiy0             = _mm_setzero_ps();
 821         fiz0             = _mm_setzero_ps();
 822         fix1             = _mm_setzero_ps();
 823         fiy1             = _mm_setzero_ps();
 824         fiz1             = _mm_setzero_ps();
 825         fix2             = _mm_setzero_ps();
 826         fiy2             = _mm_setzero_ps();
 827         fiz2             = _mm_setzero_ps();
 828         fix3             = _mm_setzero_ps();
 829         fiy3             = _mm_setzero_ps();
 830         fiz3             = _mm_setzero_ps();
 831
 832         /* Start inner kernel loop */
 833         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrA             = jjnr[jidx];
 838             jnrB             = jjnr[jidx+1];
 839             jnrC             = jjnr[jidx+2];
 840             jnrD             = jjnr[jidx+3];
 841             j_coord_offsetA  = DIM*jnrA;
 842             j_coord_offsetB  = DIM*jnrB;
 843             j_coord_offsetC  = DIM*jnrC;
 844             j_coord_offsetD  = DIM*jnrD;
 845
 846             /* load j atom coordinates */
 847             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 848                                               x+j_coord_offsetC,x+j_coord_offsetD,
 849                                               &jx0,&jy0,&jz0);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm_sub_ps(ix0,jx0);
 853             dy00             = _mm_sub_ps(iy0,jy0);
 854             dz00             = _mm_sub_ps(iz0,jz0);
 855             dx10             = _mm_sub_ps(ix1,jx0);
 856             dy10             = _mm_sub_ps(iy1,jy0);
 857             dz10             = _mm_sub_ps(iz1,jz0);
 858             dx20             = _mm_sub_ps(ix2,jx0);
 859             dy20             = _mm_sub_ps(iy2,jy0);
 860             dz20             = _mm_sub_ps(iz2,jz0);
 861             dx30             = _mm_sub_ps(ix3,jx0);
 862             dy30             = _mm_sub_ps(iy3,jy0);
 863             dz30             = _mm_sub_ps(iz3,jz0);
 864
 865             /* Calculate squared distance and things based on it */
 866             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 867             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 868             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 869             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 870
 871             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 872             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 873             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 874             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 875
 876             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 877             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 878             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 879
 880             /* Load parameters for j particles */
 881             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 882                                                               charge+jnrC+0,charge+jnrD+0);
 883             vdwjidx0A        = 2*vdwtype[jnrA+0];
 884             vdwjidx0B        = 2*vdwtype[jnrB+0];
 885             vdwjidx0C        = 2*vdwtype[jnrC+0];
 886             vdwjidx0D        = 2*vdwtype[jnrD+0];
 887
 888             fjx0             = _mm_setzero_ps();
 889             fjy0             = _mm_setzero_ps();
 890             fjz0             = _mm_setzero_ps();
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             r00              = _mm_mul_ps(rsq00,rinv00);
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 900                                          vdwparam+vdwioffset0+vdwjidx0B,
 901                                          vdwparam+vdwioffset0+vdwjidx0C,
 902                                          vdwparam+vdwioffset0+vdwjidx0D,
 903                                          &c6_00,&c12_00);
 904
 905             /* Calculate table index by multiplying r with table scale and truncate to integer */
 906             rt               = _mm_mul_ps(r00,vftabscale);
 907             vfitab           = _mm_cvttps_epi32(rt);
 908             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 909             vfitab           = _mm_slli_epi32(vfitab,3);
 910
 911             /* CUBIC SPLINE TABLE DISPERSION */
 912             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 913             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 914             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 915             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 916             _MM_TRANSPOSE4_PS(Y,F,G,H);
 917             Heps             = _mm_mul_ps(vfeps,H);
 918             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 919             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 920             fvdw6            = _mm_mul_ps(c6_00,FF);
 921
 922             /* CUBIC SPLINE TABLE REPULSION */
 923             vfitab           = _mm_add_epi32(vfitab,ifour);
 924             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 925             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 926             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 927             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 928             _MM_TRANSPOSE4_PS(Y,F,G,H);
 929             Heps             = _mm_mul_ps(vfeps,H);
 930             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 931             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 932             fvdw12           = _mm_mul_ps(c12_00,FF);
 933             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 934
 935             fscal            = fvdw;
 936
 937             /* Calculate temporary vectorial force */
 938             tx               = _mm_mul_ps(fscal,dx00);
 939             ty               = _mm_mul_ps(fscal,dy00);
 940             tz               = _mm_mul_ps(fscal,dz00);
 941
 942             /* Update vectorial force */
 943             fix0             = _mm_add_ps(fix0,tx);
 944             fiy0             = _mm_add_ps(fiy0,ty);
 945             fiz0             = _mm_add_ps(fiz0,tz);
 946
 947             fjx0             = _mm_add_ps(fjx0,tx);
 948             fjy0             = _mm_add_ps(fjy0,ty);
 949             fjz0             = _mm_add_ps(fjz0,tz);
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq10             = _mm_mul_ps(iq1,jq0);
 957
 958             /* REACTION-FIELD ELECTROSTATICS */
 959             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 960
 961             fscal            = felec;
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm_mul_ps(fscal,dx10);
 965             ty               = _mm_mul_ps(fscal,dy10);
 966             tz               = _mm_mul_ps(fscal,dz10);
 967
 968             /* Update vectorial force */
 969             fix1             = _mm_add_ps(fix1,tx);
 970             fiy1             = _mm_add_ps(fiy1,ty);
 971             fiz1             = _mm_add_ps(fiz1,tz);
 972
 973             fjx0             = _mm_add_ps(fjx0,tx);
 974             fjy0             = _mm_add_ps(fjy0,ty);
 975             fjz0             = _mm_add_ps(fjz0,tz);
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             /* Compute parameters for interactions between i and j atoms */
 982             qq20             = _mm_mul_ps(iq2,jq0);
 983
 984             /* REACTION-FIELD ELECTROSTATICS */
 985             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 986
 987             fscal            = felec;
 988
 989             /* Calculate temporary vectorial force */
 990             tx               = _mm_mul_ps(fscal,dx20);
 991             ty               = _mm_mul_ps(fscal,dy20);
 992             tz               = _mm_mul_ps(fscal,dz20);
 993
 994             /* Update vectorial force */
 995             fix2             = _mm_add_ps(fix2,tx);
 996             fiy2             = _mm_add_ps(fiy2,ty);
 997             fiz2             = _mm_add_ps(fiz2,tz);
 998
 999             fjx0             = _mm_add_ps(fjx0,tx);
1000             fjy0             = _mm_add_ps(fjy0,ty);
1001             fjz0             = _mm_add_ps(fjz0,tz);
1002
1003             /**************************
1004              * CALCULATE INTERACTIONS *
1005              **************************/
1006
1007             /* Compute parameters for interactions between i and j atoms */
1008             qq30             = _mm_mul_ps(iq3,jq0);
1009
1010             /* REACTION-FIELD ELECTROSTATICS */
1011             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1012
1013             fscal            = felec;
1014
1015             /* Calculate temporary vectorial force */
1016             tx               = _mm_mul_ps(fscal,dx30);
1017             ty               = _mm_mul_ps(fscal,dy30);
1018             tz               = _mm_mul_ps(fscal,dz30);
1019
1020             /* Update vectorial force */
1021             fix3             = _mm_add_ps(fix3,tx);
1022             fiy3             = _mm_add_ps(fiy3,ty);
1023             fiz3             = _mm_add_ps(fiz3,tz);
1024
1025             fjx0             = _mm_add_ps(fjx0,tx);
1026             fjy0             = _mm_add_ps(fjy0,ty);
1027             fjz0             = _mm_add_ps(fjz0,tz);
1028
1029             fjptrA             = f+j_coord_offsetA;
1030             fjptrB             = f+j_coord_offsetB;
1031             fjptrC             = f+j_coord_offsetC;
1032             fjptrD             = f+j_coord_offsetD;
1033
1034             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1035
1036             /* Inner loop uses 129 flops */
1037         }
1038
1039         if(jidx<j_index_end)
1040         {
1041
1042             /* Get j neighbor index, and coordinate index */
1043             jnrlistA         = jjnr[jidx];
1044             jnrlistB         = jjnr[jidx+1];
1045             jnrlistC         = jjnr[jidx+2];
1046             jnrlistD         = jjnr[jidx+3];
1047             /* Sign of each element will be negative for non-real atoms.
1048              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1049              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1050              */
1051             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1052             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1053             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1054             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1055             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1056             j_coord_offsetA  = DIM*jnrA;
1057             j_coord_offsetB  = DIM*jnrB;
1058             j_coord_offsetC  = DIM*jnrC;
1059             j_coord_offsetD  = DIM*jnrD;
1060
1061             /* load j atom coordinates */
1062             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1063                                               x+j_coord_offsetC,x+j_coord_offsetD,
1064                                               &jx0,&jy0,&jz0);
1065
1066             /* Calculate displacement vector */
1067             dx00             = _mm_sub_ps(ix0,jx0);
1068             dy00             = _mm_sub_ps(iy0,jy0);
1069             dz00             = _mm_sub_ps(iz0,jz0);
1070             dx10             = _mm_sub_ps(ix1,jx0);
1071             dy10             = _mm_sub_ps(iy1,jy0);
1072             dz10             = _mm_sub_ps(iz1,jz0);
1073             dx20             = _mm_sub_ps(ix2,jx0);
1074             dy20             = _mm_sub_ps(iy2,jy0);
1075             dz20             = _mm_sub_ps(iz2,jz0);
1076             dx30             = _mm_sub_ps(ix3,jx0);
1077             dy30             = _mm_sub_ps(iy3,jy0);
1078             dz30             = _mm_sub_ps(iz3,jz0);
1079
1080             /* Calculate squared distance and things based on it */
1081             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1082             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1083             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1084             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1085
1086             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1087             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1088             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1089             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1090
1091             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1092             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1093             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1094
1095             /* Load parameters for j particles */
1096             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1097                                                               charge+jnrC+0,charge+jnrD+0);
1098             vdwjidx0A        = 2*vdwtype[jnrA+0];
1099             vdwjidx0B        = 2*vdwtype[jnrB+0];
1100             vdwjidx0C        = 2*vdwtype[jnrC+0];
1101             vdwjidx0D        = 2*vdwtype[jnrD+0];
1102
1103             fjx0             = _mm_setzero_ps();
1104             fjy0             = _mm_setzero_ps();
1105             fjz0             = _mm_setzero_ps();
1106
1107             /**************************
1108              * CALCULATE INTERACTIONS *
1109              **************************/
1110
1111             r00              = _mm_mul_ps(rsq00,rinv00);
1112             r00              = _mm_andnot_ps(dummy_mask,r00);
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1116                                          vdwparam+vdwioffset0+vdwjidx0B,
1117                                          vdwparam+vdwioffset0+vdwjidx0C,
1118                                          vdwparam+vdwioffset0+vdwjidx0D,
1119                                          &c6_00,&c12_00);
1120
1121             /* Calculate table index by multiplying r with table scale and truncate to integer */
1122             rt               = _mm_mul_ps(r00,vftabscale);
1123             vfitab           = _mm_cvttps_epi32(rt);
1124             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1125             vfitab           = _mm_slli_epi32(vfitab,3);
1126
1127             /* CUBIC SPLINE TABLE DISPERSION */
1128             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1129             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1130             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1131             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1132             _MM_TRANSPOSE4_PS(Y,F,G,H);
1133             Heps             = _mm_mul_ps(vfeps,H);
1134             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1135             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1136             fvdw6            = _mm_mul_ps(c6_00,FF);
1137
1138             /* CUBIC SPLINE TABLE REPULSION */
1139             vfitab           = _mm_add_epi32(vfitab,ifour);
1140             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1141             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1142             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1143             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1144             _MM_TRANSPOSE4_PS(Y,F,G,H);
1145             Heps             = _mm_mul_ps(vfeps,H);
1146             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1147             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1148             fvdw12           = _mm_mul_ps(c12_00,FF);
1149             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1150
1151             fscal            = fvdw;
1152
1153             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1154
1155             /* Calculate temporary vectorial force */
1156             tx               = _mm_mul_ps(fscal,dx00);
1157             ty               = _mm_mul_ps(fscal,dy00);
1158             tz               = _mm_mul_ps(fscal,dz00);
1159
1160             /* Update vectorial force */
1161             fix0             = _mm_add_ps(fix0,tx);
1162             fiy0             = _mm_add_ps(fiy0,ty);
1163             fiz0             = _mm_add_ps(fiz0,tz);
1164
1165             fjx0             = _mm_add_ps(fjx0,tx);
1166             fjy0             = _mm_add_ps(fjy0,ty);
1167             fjz0             = _mm_add_ps(fjz0,tz);
1168
1169             /**************************
1170              * CALCULATE INTERACTIONS *
1171              **************************/
1172
1173             /* Compute parameters for interactions between i and j atoms */
1174             qq10             = _mm_mul_ps(iq1,jq0);
1175
1176             /* REACTION-FIELD ELECTROSTATICS */
1177             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1178
1179             fscal            = felec;
1180
1181             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1182
1183             /* Calculate temporary vectorial force */
1184             tx               = _mm_mul_ps(fscal,dx10);
1185             ty               = _mm_mul_ps(fscal,dy10);
1186             tz               = _mm_mul_ps(fscal,dz10);
1187
1188             /* Update vectorial force */
1189             fix1             = _mm_add_ps(fix1,tx);
1190             fiy1             = _mm_add_ps(fiy1,ty);
1191             fiz1             = _mm_add_ps(fiz1,tz);
1192
1193             fjx0             = _mm_add_ps(fjx0,tx);
1194             fjy0             = _mm_add_ps(fjy0,ty);
1195             fjz0             = _mm_add_ps(fjz0,tz);
1196
1197             /**************************
1198              * CALCULATE INTERACTIONS *
1199              **************************/
1200
1201             /* Compute parameters for interactions between i and j atoms */
1202             qq20             = _mm_mul_ps(iq2,jq0);
1203
1204             /* REACTION-FIELD ELECTROSTATICS */
1205             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1206
1207             fscal            = felec;
1208
1209             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1210
1211             /* Calculate temporary vectorial force */
1212             tx               = _mm_mul_ps(fscal,dx20);
1213             ty               = _mm_mul_ps(fscal,dy20);
1214             tz               = _mm_mul_ps(fscal,dz20);
1215
1216             /* Update vectorial force */
1217             fix2             = _mm_add_ps(fix2,tx);
1218             fiy2             = _mm_add_ps(fiy2,ty);
1219             fiz2             = _mm_add_ps(fiz2,tz);
1220
1221             fjx0             = _mm_add_ps(fjx0,tx);
1222             fjy0             = _mm_add_ps(fjy0,ty);
1223             fjz0             = _mm_add_ps(fjz0,tz);
1224
1225             /**************************
1226              * CALCULATE INTERACTIONS *
1227              **************************/
1228
1229             /* Compute parameters for interactions between i and j atoms */
1230             qq30             = _mm_mul_ps(iq3,jq0);
1231
1232             /* REACTION-FIELD ELECTROSTATICS */
1233             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1234
1235             fscal            = felec;
1236
1237             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1238
1239             /* Calculate temporary vectorial force */
1240             tx               = _mm_mul_ps(fscal,dx30);
1241             ty               = _mm_mul_ps(fscal,dy30);
1242             tz               = _mm_mul_ps(fscal,dz30);
1243
1244             /* Update vectorial force */
1245             fix3             = _mm_add_ps(fix3,tx);
1246             fiy3             = _mm_add_ps(fiy3,ty);
1247             fiz3             = _mm_add_ps(fiz3,tz);
1248
1249             fjx0             = _mm_add_ps(fjx0,tx);
1250             fjy0             = _mm_add_ps(fjy0,ty);
1251             fjz0             = _mm_add_ps(fjz0,tz);
1252
1253             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1254             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1255             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1256             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1257
1258             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1259
1260             /* Inner loop uses 130 flops */
1261         }
1262
1263         /* End of innermost loop */
1264
1265         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1266                                               f+i_coord_offset,fshift+i_shift_offset);
1267
1268         /* Increment number of inner iterations */
1269         inneriter                  += j_index_end - j_index_start;
1270
1271         /* Outer loop uses 24 flops */
1272     }
1273
1274     /* Increment number of outer iterations */
1275     outeriter        += nri;
1276
1277     /* Update outer/inner flops */
1278
1279     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1280 }