src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 sse4_1_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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_sse4_1_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  85     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  94     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_ps(fr->ic->epsfac);
 115     charge           = mdatoms->chargeA;
 116     krf              = _mm_set1_ps(fr->ic->k_rf);
 117     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 118     crf              = _mm_set1_ps(fr->ic->c_rf);
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->ic->rcoulomb;
 128     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 162
 163         fix0             = _mm_setzero_ps();
 164         fiy0             = _mm_setzero_ps();
 165         fiz0             = _mm_setzero_ps();
 166
 167         /* Load parameters for i particles */
 168         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 169         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_ps();
 173         vvdwsum          = _mm_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186             j_coord_offsetC  = DIM*jnrC;
 187             j_coord_offsetD  = DIM*jnrD;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               x+j_coord_offsetC,x+j_coord_offsetD,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_ps(ix0,jx0);
 196             dy00             = _mm_sub_ps(iy0,jy0);
 197             dz00             = _mm_sub_ps(iz0,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 201
 202             rinv00           = sse41_invsqrt_f(rsq00);
 203
 204             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 208                                                               charge+jnrC+0,charge+jnrD+0);
 209             vdwjidx0A        = 2*vdwtype[jnrA+0];
 210             vdwjidx0B        = 2*vdwtype[jnrB+0];
 211             vdwjidx0C        = 2*vdwtype[jnrC+0];
 212             vdwjidx0D        = 2*vdwtype[jnrD+0];
 213
 214             /**************************
 215              * CALCULATE INTERACTIONS *
 216              **************************/
 217
 218             if (gmx_mm_any_lt(rsq00,rcutoff2))
 219             {
 220
 221             r00              = _mm_mul_ps(rsq00,rinv00);
 222
 223             /* Compute parameters for interactions between i and j atoms */
 224             qq00             = _mm_mul_ps(iq0,jq0);
 225             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 226                                          vdwparam+vdwioffset0+vdwjidx0B,
 227                                          vdwparam+vdwioffset0+vdwjidx0C,
 228                                          vdwparam+vdwioffset0+vdwjidx0D,
 229                                          &c6_00,&c12_00);
 230
 231             /* Calculate table index by multiplying r with table scale and truncate to integer */
 232             rt               = _mm_mul_ps(r00,vftabscale);
 233             vfitab           = _mm_cvttps_epi32(rt);
 234             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 235             vfitab           = _mm_slli_epi32(vfitab,3);
 236
 237             /* REACTION-FIELD ELECTROSTATICS */
 238             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 239             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 240
 241             /* CUBIC SPLINE TABLE DISPERSION */
 242             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 243             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 244             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 245             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 246             _MM_TRANSPOSE4_PS(Y,F,G,H);
 247             Heps             = _mm_mul_ps(vfeps,H);
 248             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 249             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 250             vvdw6            = _mm_mul_ps(c6_00,VV);
 251             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 252             fvdw6            = _mm_mul_ps(c6_00,FF);
 253
 254             /* CUBIC SPLINE TABLE REPULSION */
 255             vfitab           = _mm_add_epi32(vfitab,ifour);
 256             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 257             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 258             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 259             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 260             _MM_TRANSPOSE4_PS(Y,F,G,H);
 261             Heps             = _mm_mul_ps(vfeps,H);
 262             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 263             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 264             vvdw12           = _mm_mul_ps(c12_00,VV);
 265             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 266             fvdw12           = _mm_mul_ps(c12_00,FF);
 267             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 268             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 269
 270             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velec            = _mm_and_ps(velec,cutoff_mask);
 274             velecsum         = _mm_add_ps(velecsum,velec);
 275             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 276             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 277
 278             fscal            = _mm_add_ps(felec,fvdw);
 279
 280             fscal            = _mm_and_ps(fscal,cutoff_mask);
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_ps(fscal,dx00);
 284             ty               = _mm_mul_ps(fscal,dy00);
 285             tz               = _mm_mul_ps(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm_add_ps(fix0,tx);
 289             fiy0             = _mm_add_ps(fiy0,ty);
 290             fiz0             = _mm_add_ps(fiz0,tz);
 291
 292             fjptrA             = f+j_coord_offsetA;
 293             fjptrB             = f+j_coord_offsetB;
 294             fjptrC             = f+j_coord_offsetC;
 295             fjptrD             = f+j_coord_offsetD;
 296             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 297
 298             }
 299
 300             /* Inner loop uses 72 flops */
 301         }
 302
 303         if(jidx<j_index_end)
 304         {
 305
 306             /* Get j neighbor index, and coordinate index */
 307             jnrlistA         = jjnr[jidx];
 308             jnrlistB         = jjnr[jidx+1];
 309             jnrlistC         = jjnr[jidx+2];
 310             jnrlistD         = jjnr[jidx+3];
 311             /* Sign of each element will be negative for non-real atoms.
 312              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 313              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 314              */
 315             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 316             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 317             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 318             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 319             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 320             j_coord_offsetA  = DIM*jnrA;
 321             j_coord_offsetB  = DIM*jnrB;
 322             j_coord_offsetC  = DIM*jnrC;
 323             j_coord_offsetD  = DIM*jnrD;
 324
 325             /* load j atom coordinates */
 326             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 327                                               x+j_coord_offsetC,x+j_coord_offsetD,
 328                                               &jx0,&jy0,&jz0);
 329
 330             /* Calculate displacement vector */
 331             dx00             = _mm_sub_ps(ix0,jx0);
 332             dy00             = _mm_sub_ps(iy0,jy0);
 333             dz00             = _mm_sub_ps(iz0,jz0);
 334
 335             /* Calculate squared distance and things based on it */
 336             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 337
 338             rinv00           = sse41_invsqrt_f(rsq00);
 339
 340             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 341
 342             /* Load parameters for j particles */
 343             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 344                                                               charge+jnrC+0,charge+jnrD+0);
 345             vdwjidx0A        = 2*vdwtype[jnrA+0];
 346             vdwjidx0B        = 2*vdwtype[jnrB+0];
 347             vdwjidx0C        = 2*vdwtype[jnrC+0];
 348             vdwjidx0D        = 2*vdwtype[jnrD+0];
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             if (gmx_mm_any_lt(rsq00,rcutoff2))
 355             {
 356
 357             r00              = _mm_mul_ps(rsq00,rinv00);
 358             r00              = _mm_andnot_ps(dummy_mask,r00);
 359
 360             /* Compute parameters for interactions between i and j atoms */
 361             qq00             = _mm_mul_ps(iq0,jq0);
 362             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 363                                          vdwparam+vdwioffset0+vdwjidx0B,
 364                                          vdwparam+vdwioffset0+vdwjidx0C,
 365                                          vdwparam+vdwioffset0+vdwjidx0D,
 366                                          &c6_00,&c12_00);
 367
 368             /* Calculate table index by multiplying r with table scale and truncate to integer */
 369             rt               = _mm_mul_ps(r00,vftabscale);
 370             vfitab           = _mm_cvttps_epi32(rt);
 371             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 372             vfitab           = _mm_slli_epi32(vfitab,3);
 373
 374             /* REACTION-FIELD ELECTROSTATICS */
 375             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 376             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 377
 378             /* CUBIC SPLINE TABLE DISPERSION */
 379             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 380             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 381             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 382             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 383             _MM_TRANSPOSE4_PS(Y,F,G,H);
 384             Heps             = _mm_mul_ps(vfeps,H);
 385             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 386             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 387             vvdw6            = _mm_mul_ps(c6_00,VV);
 388             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 389             fvdw6            = _mm_mul_ps(c6_00,FF);
 390
 391             /* CUBIC SPLINE TABLE REPULSION */
 392             vfitab           = _mm_add_epi32(vfitab,ifour);
 393             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 394             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 395             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 396             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 397             _MM_TRANSPOSE4_PS(Y,F,G,H);
 398             Heps             = _mm_mul_ps(vfeps,H);
 399             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 400             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 401             vvdw12           = _mm_mul_ps(c12_00,VV);
 402             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 403             fvdw12           = _mm_mul_ps(c12_00,FF);
 404             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 405             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 406
 407             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm_and_ps(velec,cutoff_mask);
 411             velec            = _mm_andnot_ps(dummy_mask,velec);
 412             velecsum         = _mm_add_ps(velecsum,velec);
 413             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 414             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 415             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 416
 417             fscal            = _mm_add_ps(felec,fvdw);
 418
 419             fscal            = _mm_and_ps(fscal,cutoff_mask);
 420
 421             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 422
 423             /* Calculate temporary vectorial force */
 424             tx               = _mm_mul_ps(fscal,dx00);
 425             ty               = _mm_mul_ps(fscal,dy00);
 426             tz               = _mm_mul_ps(fscal,dz00);
 427
 428             /* Update vectorial force */
 429             fix0             = _mm_add_ps(fix0,tx);
 430             fiy0             = _mm_add_ps(fiy0,ty);
 431             fiz0             = _mm_add_ps(fiz0,tz);
 432
 433             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 434             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 435             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 436             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 437             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 438
 439             }
 440
 441             /* Inner loop uses 73 flops */
 442         }
 443
 444         /* End of innermost loop */
 445
 446         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 447                                               f+i_coord_offset,fshift+i_shift_offset);
 448
 449         ggid                        = gid[iidx];
 450         /* Update potential energies */
 451         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 452         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 453
 454         /* Increment number of inner iterations */
 455         inneriter                  += j_index_end - j_index_start;
 456
 457         /* Outer loop uses 9 flops */
 458     }
 459
 460     /* Increment number of outer iterations */
 461     outeriter        += nri;
 462
 463     /* Update outer/inner flops */
 464
 465     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 466 }
 467 /*
 468  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse4_1_single
 469  * Electrostatics interaction: ReactionField
 470  * VdW interaction:            CubicSplineTable
 471  * Geometry:                   Particle-Particle
 472  * Calculate force/pot:        Force
 473  */
 474 void
 475 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse4_1_single
 476                     (t_nblist                    * gmx_restrict       nlist,
 477                      rvec                        * gmx_restrict          xx,
 478                      rvec                        * gmx_restrict          ff,
 479                      struct t_forcerec           * gmx_restrict          fr,
 480                      t_mdatoms                   * gmx_restrict     mdatoms,
 481                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 482                      t_nrnb                      * gmx_restrict        nrnb)
 483 {
 484     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 485      * just 0 for non-waters.
 486      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 487      * jnr indices corresponding to data put in the four positions in the SIMD register.
 488      */
 489     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 490     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 491     int              jnrA,jnrB,jnrC,jnrD;
 492     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 493     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 494     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 495     real             rcutoff_scalar;
 496     real             *shiftvec,*fshift,*x,*f;
 497     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 498     real             scratch[4*DIM];
 499     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 500     int              vdwioffset0;
 501     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 502     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 503     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 504     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 505     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 506     real             *charge;
 507     int              nvdwtype;
 508     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 509     int              *vdwtype;
 510     real             *vdwparam;
 511     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 512     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 513     __m128i          vfitab;
 514     __m128i          ifour       = _mm_set1_epi32(4);
 515     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 516     real             *vftab;
 517     __m128           dummy_mask,cutoff_mask;
 518     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 519     __m128           one     = _mm_set1_ps(1.0);
 520     __m128           two     = _mm_set1_ps(2.0);
 521     x                = xx[0];
 522     f                = ff[0];
 523
 524     nri              = nlist->nri;
 525     iinr             = nlist->iinr;
 526     jindex           = nlist->jindex;
 527     jjnr             = nlist->jjnr;
 528     shiftidx         = nlist->shift;
 529     gid              = nlist->gid;
 530     shiftvec         = fr->shift_vec[0];
 531     fshift           = fr->fshift[0];
 532     facel            = _mm_set1_ps(fr->ic->epsfac);
 533     charge           = mdatoms->chargeA;
 534     krf              = _mm_set1_ps(fr->ic->k_rf);
 535     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 536     crf              = _mm_set1_ps(fr->ic->c_rf);
 537     nvdwtype         = fr->ntype;
 538     vdwparam         = fr->nbfp;
 539     vdwtype          = mdatoms->typeA;
 540
 541     vftab            = kernel_data->table_vdw->data;
 542     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 543
 544     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 545     rcutoff_scalar   = fr->ic->rcoulomb;
 546     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 547     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 548
 549     /* Avoid stupid compiler warnings */
 550     jnrA = jnrB = jnrC = jnrD = 0;
 551     j_coord_offsetA = 0;
 552     j_coord_offsetB = 0;
 553     j_coord_offsetC = 0;
 554     j_coord_offsetD = 0;
 555
 556     outeriter        = 0;
 557     inneriter        = 0;
 558
 559     for(iidx=0;iidx<4*DIM;iidx++)
 560     {
 561         scratch[iidx] = 0.0;
 562     }
 563
 564     /* Start outer loop over neighborlists */
 565     for(iidx=0; iidx<nri; iidx++)
 566     {
 567         /* Load shift vector for this list */
 568         i_shift_offset   = DIM*shiftidx[iidx];
 569
 570         /* Load limits for loop over neighbors */
 571         j_index_start    = jindex[iidx];
 572         j_index_end      = jindex[iidx+1];
 573
 574         /* Get outer coordinate index */
 575         inr              = iinr[iidx];
 576         i_coord_offset   = DIM*inr;
 577
 578         /* Load i particle coords and add shift vector */
 579         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 580
 581         fix0             = _mm_setzero_ps();
 582         fiy0             = _mm_setzero_ps();
 583         fiz0             = _mm_setzero_ps();
 584
 585         /* Load parameters for i particles */
 586         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 587         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 588
 589         /* Start inner kernel loop */
 590         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 591         {
 592
 593             /* Get j neighbor index, and coordinate index */
 594             jnrA             = jjnr[jidx];
 595             jnrB             = jjnr[jidx+1];
 596             jnrC             = jjnr[jidx+2];
 597             jnrD             = jjnr[jidx+3];
 598             j_coord_offsetA  = DIM*jnrA;
 599             j_coord_offsetB  = DIM*jnrB;
 600             j_coord_offsetC  = DIM*jnrC;
 601             j_coord_offsetD  = DIM*jnrD;
 602
 603             /* load j atom coordinates */
 604             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 605                                               x+j_coord_offsetC,x+j_coord_offsetD,
 606                                               &jx0,&jy0,&jz0);
 607
 608             /* Calculate displacement vector */
 609             dx00             = _mm_sub_ps(ix0,jx0);
 610             dy00             = _mm_sub_ps(iy0,jy0);
 611             dz00             = _mm_sub_ps(iz0,jz0);
 612
 613             /* Calculate squared distance and things based on it */
 614             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 615
 616             rinv00           = sse41_invsqrt_f(rsq00);
 617
 618             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 619
 620             /* Load parameters for j particles */
 621             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 622                                                               charge+jnrC+0,charge+jnrD+0);
 623             vdwjidx0A        = 2*vdwtype[jnrA+0];
 624             vdwjidx0B        = 2*vdwtype[jnrB+0];
 625             vdwjidx0C        = 2*vdwtype[jnrC+0];
 626             vdwjidx0D        = 2*vdwtype[jnrD+0];
 627
 628             /**************************
 629              * CALCULATE INTERACTIONS *
 630              **************************/
 631
 632             if (gmx_mm_any_lt(rsq00,rcutoff2))
 633             {
 634
 635             r00              = _mm_mul_ps(rsq00,rinv00);
 636
 637             /* Compute parameters for interactions between i and j atoms */
 638             qq00             = _mm_mul_ps(iq0,jq0);
 639             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 640                                          vdwparam+vdwioffset0+vdwjidx0B,
 641                                          vdwparam+vdwioffset0+vdwjidx0C,
 642                                          vdwparam+vdwioffset0+vdwjidx0D,
 643                                          &c6_00,&c12_00);
 644
 645             /* Calculate table index by multiplying r with table scale and truncate to integer */
 646             rt               = _mm_mul_ps(r00,vftabscale);
 647             vfitab           = _mm_cvttps_epi32(rt);
 648             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 649             vfitab           = _mm_slli_epi32(vfitab,3);
 650
 651             /* REACTION-FIELD ELECTROSTATICS */
 652             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 653
 654             /* CUBIC SPLINE TABLE DISPERSION */
 655             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 656             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 657             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 658             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 659             _MM_TRANSPOSE4_PS(Y,F,G,H);
 660             Heps             = _mm_mul_ps(vfeps,H);
 661             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 662             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 663             fvdw6            = _mm_mul_ps(c6_00,FF);
 664
 665             /* CUBIC SPLINE TABLE REPULSION */
 666             vfitab           = _mm_add_epi32(vfitab,ifour);
 667             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 668             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 669             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 670             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 671             _MM_TRANSPOSE4_PS(Y,F,G,H);
 672             Heps             = _mm_mul_ps(vfeps,H);
 673             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 674             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 675             fvdw12           = _mm_mul_ps(c12_00,FF);
 676             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 677
 678             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 679
 680             fscal            = _mm_add_ps(felec,fvdw);
 681
 682             fscal            = _mm_and_ps(fscal,cutoff_mask);
 683
 684             /* Calculate temporary vectorial force */
 685             tx               = _mm_mul_ps(fscal,dx00);
 686             ty               = _mm_mul_ps(fscal,dy00);
 687             tz               = _mm_mul_ps(fscal,dz00);
 688
 689             /* Update vectorial force */
 690             fix0             = _mm_add_ps(fix0,tx);
 691             fiy0             = _mm_add_ps(fiy0,ty);
 692             fiz0             = _mm_add_ps(fiz0,tz);
 693
 694             fjptrA             = f+j_coord_offsetA;
 695             fjptrB             = f+j_coord_offsetB;
 696             fjptrC             = f+j_coord_offsetC;
 697             fjptrD             = f+j_coord_offsetD;
 698             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 699
 700             }
 701
 702             /* Inner loop uses 57 flops */
 703         }
 704
 705         if(jidx<j_index_end)
 706         {
 707
 708             /* Get j neighbor index, and coordinate index */
 709             jnrlistA         = jjnr[jidx];
 710             jnrlistB         = jjnr[jidx+1];
 711             jnrlistC         = jjnr[jidx+2];
 712             jnrlistD         = jjnr[jidx+3];
 713             /* Sign of each element will be negative for non-real atoms.
 714              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 715              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 716              */
 717             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 718             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 719             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 720             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 721             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 722             j_coord_offsetA  = DIM*jnrA;
 723             j_coord_offsetB  = DIM*jnrB;
 724             j_coord_offsetC  = DIM*jnrC;
 725             j_coord_offsetD  = DIM*jnrD;
 726
 727             /* load j atom coordinates */
 728             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 729                                               x+j_coord_offsetC,x+j_coord_offsetD,
 730                                               &jx0,&jy0,&jz0);
 731
 732             /* Calculate displacement vector */
 733             dx00             = _mm_sub_ps(ix0,jx0);
 734             dy00             = _mm_sub_ps(iy0,jy0);
 735             dz00             = _mm_sub_ps(iz0,jz0);
 736
 737             /* Calculate squared distance and things based on it */
 738             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 739
 740             rinv00           = sse41_invsqrt_f(rsq00);
 741
 742             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 743
 744             /* Load parameters for j particles */
 745             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 746                                                               charge+jnrC+0,charge+jnrD+0);
 747             vdwjidx0A        = 2*vdwtype[jnrA+0];
 748             vdwjidx0B        = 2*vdwtype[jnrB+0];
 749             vdwjidx0C        = 2*vdwtype[jnrC+0];
 750             vdwjidx0D        = 2*vdwtype[jnrD+0];
 751
 752             /**************************
 753              * CALCULATE INTERACTIONS *
 754              **************************/
 755
 756             if (gmx_mm_any_lt(rsq00,rcutoff2))
 757             {
 758
 759             r00              = _mm_mul_ps(rsq00,rinv00);
 760             r00              = _mm_andnot_ps(dummy_mask,r00);
 761
 762             /* Compute parameters for interactions between i and j atoms */
 763             qq00             = _mm_mul_ps(iq0,jq0);
 764             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 765                                          vdwparam+vdwioffset0+vdwjidx0B,
 766                                          vdwparam+vdwioffset0+vdwjidx0C,
 767                                          vdwparam+vdwioffset0+vdwjidx0D,
 768                                          &c6_00,&c12_00);
 769
 770             /* Calculate table index by multiplying r with table scale and truncate to integer */
 771             rt               = _mm_mul_ps(r00,vftabscale);
 772             vfitab           = _mm_cvttps_epi32(rt);
 773             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 774             vfitab           = _mm_slli_epi32(vfitab,3);
 775
 776             /* REACTION-FIELD ELECTROSTATICS */
 777             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 778
 779             /* CUBIC SPLINE TABLE DISPERSION */
 780             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 781             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 782             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 783             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 784             _MM_TRANSPOSE4_PS(Y,F,G,H);
 785             Heps             = _mm_mul_ps(vfeps,H);
 786             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 787             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 788             fvdw6            = _mm_mul_ps(c6_00,FF);
 789
 790             /* CUBIC SPLINE TABLE REPULSION */
 791             vfitab           = _mm_add_epi32(vfitab,ifour);
 792             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 793             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 794             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 795             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 796             _MM_TRANSPOSE4_PS(Y,F,G,H);
 797             Heps             = _mm_mul_ps(vfeps,H);
 798             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 799             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 800             fvdw12           = _mm_mul_ps(c12_00,FF);
 801             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 802
 803             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 804
 805             fscal            = _mm_add_ps(felec,fvdw);
 806
 807             fscal            = _mm_and_ps(fscal,cutoff_mask);
 808
 809             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 810
 811             /* Calculate temporary vectorial force */
 812             tx               = _mm_mul_ps(fscal,dx00);
 813             ty               = _mm_mul_ps(fscal,dy00);
 814             tz               = _mm_mul_ps(fscal,dz00);
 815
 816             /* Update vectorial force */
 817             fix0             = _mm_add_ps(fix0,tx);
 818             fiy0             = _mm_add_ps(fiy0,ty);
 819             fiz0             = _mm_add_ps(fiz0,tz);
 820
 821             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 822             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 823             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 824             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 825             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 826
 827             }
 828
 829             /* Inner loop uses 58 flops */
 830         }
 831
 832         /* End of innermost loop */
 833
 834         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 835                                               f+i_coord_offset,fshift+i_shift_offset);
 836
 837         /* Increment number of inner iterations */
 838         inneriter                  += j_index_end - j_index_start;
 839
 840         /* Outer loop uses 7 flops */
 841     }
 842
 843     /* Increment number of outer iterations */
 844     outeriter        += nri;
 845
 846     /* Update outer/inner flops */
 847
 848     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*58);
 849 }