src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse4_1_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse4_1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 101     real             *vftab;
 102     __m128           dummy_mask,cutoff_mask;
 103     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 104     __m128           one     = _mm_set1_ps(1.0);
 105     __m128           two     = _mm_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119     krf              = _mm_set1_ps(fr->ic->k_rf);
 120     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 121     crf              = _mm_set1_ps(fr->ic->c_rf);
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 160
 161         fix0             = _mm_setzero_ps();
 162         fiy0             = _mm_setzero_ps();
 163         fiz0             = _mm_setzero_ps();
 164
 165         /* Load parameters for i particles */
 166         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 167         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_ps();
 171         vvdwsum          = _mm_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184             j_coord_offsetC  = DIM*jnrC;
 185             j_coord_offsetD  = DIM*jnrD;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               x+j_coord_offsetC,x+j_coord_offsetD,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_ps(ix0,jx0);
 194             dy00             = _mm_sub_ps(iy0,jy0);
 195             dz00             = _mm_sub_ps(iz0,jz0);
 196
 197             /* Calculate squared distance and things based on it */
 198             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 199
 200             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 201
 202             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 203
 204             /* Load parameters for j particles */
 205             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 206                                                               charge+jnrC+0,charge+jnrD+0);
 207             vdwjidx0A        = 2*vdwtype[jnrA+0];
 208             vdwjidx0B        = 2*vdwtype[jnrB+0];
 209             vdwjidx0C        = 2*vdwtype[jnrC+0];
 210             vdwjidx0D        = 2*vdwtype[jnrD+0];
 211
 212             /**************************
 213              * CALCULATE INTERACTIONS *
 214              **************************/
 215
 216             r00              = _mm_mul_ps(rsq00,rinv00);
 217
 218             /* Compute parameters for interactions between i and j atoms */
 219             qq00             = _mm_mul_ps(iq0,jq0);
 220             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 221                                          vdwparam+vdwioffset0+vdwjidx0B,
 222                                          vdwparam+vdwioffset0+vdwjidx0C,
 223                                          vdwparam+vdwioffset0+vdwjidx0D,
 224                                          &c6_00,&c12_00);
 225
 226             /* Calculate table index by multiplying r with table scale and truncate to integer */
 227             rt               = _mm_mul_ps(r00,vftabscale);
 228             vfitab           = _mm_cvttps_epi32(rt);
 229             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 230             vfitab           = _mm_slli_epi32(vfitab,3);
 231
 232             /* REACTION-FIELD ELECTROSTATICS */
 233             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 234             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 235
 236             /* CUBIC SPLINE TABLE DISPERSION */
 237             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 238             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 239             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 240             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 241             _MM_TRANSPOSE4_PS(Y,F,G,H);
 242             Heps             = _mm_mul_ps(vfeps,H);
 243             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 244             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 245             vvdw6            = _mm_mul_ps(c6_00,VV);
 246             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 247             fvdw6            = _mm_mul_ps(c6_00,FF);
 248
 249             /* CUBIC SPLINE TABLE REPULSION */
 250             vfitab           = _mm_add_epi32(vfitab,ifour);
 251             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 252             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 253             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 254             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 255             _MM_TRANSPOSE4_PS(Y,F,G,H);
 256             Heps             = _mm_mul_ps(vfeps,H);
 257             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 258             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 259             vvdw12           = _mm_mul_ps(c12_00,VV);
 260             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 261             fvdw12           = _mm_mul_ps(c12_00,FF);
 262             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 263             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             velecsum         = _mm_add_ps(velecsum,velec);
 267             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = _mm_add_ps(felec,fvdw);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_ps(fscal,dx00);
 273             ty               = _mm_mul_ps(fscal,dy00);
 274             tz               = _mm_mul_ps(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_ps(fix0,tx);
 278             fiy0             = _mm_add_ps(fiy0,ty);
 279             fiz0             = _mm_add_ps(fiz0,tz);
 280
 281             fjptrA             = f+j_coord_offsetA;
 282             fjptrB             = f+j_coord_offsetB;
 283             fjptrC             = f+j_coord_offsetC;
 284             fjptrD             = f+j_coord_offsetD;
 285             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 286
 287             /* Inner loop uses 67 flops */
 288         }
 289
 290         if(jidx<j_index_end)
 291         {
 292
 293             /* Get j neighbor index, and coordinate index */
 294             jnrlistA         = jjnr[jidx];
 295             jnrlistB         = jjnr[jidx+1];
 296             jnrlistC         = jjnr[jidx+2];
 297             jnrlistD         = jjnr[jidx+3];
 298             /* Sign of each element will be negative for non-real atoms.
 299              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 300              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 301              */
 302             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 303             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 304             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 305             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 306             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 307             j_coord_offsetA  = DIM*jnrA;
 308             j_coord_offsetB  = DIM*jnrB;
 309             j_coord_offsetC  = DIM*jnrC;
 310             j_coord_offsetD  = DIM*jnrD;
 311
 312             /* load j atom coordinates */
 313             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 314                                               x+j_coord_offsetC,x+j_coord_offsetD,
 315                                               &jx0,&jy0,&jz0);
 316
 317             /* Calculate displacement vector */
 318             dx00             = _mm_sub_ps(ix0,jx0);
 319             dy00             = _mm_sub_ps(iy0,jy0);
 320             dz00             = _mm_sub_ps(iz0,jz0);
 321
 322             /* Calculate squared distance and things based on it */
 323             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 324
 325             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 326
 327             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 328
 329             /* Load parameters for j particles */
 330             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 331                                                               charge+jnrC+0,charge+jnrD+0);
 332             vdwjidx0A        = 2*vdwtype[jnrA+0];
 333             vdwjidx0B        = 2*vdwtype[jnrB+0];
 334             vdwjidx0C        = 2*vdwtype[jnrC+0];
 335             vdwjidx0D        = 2*vdwtype[jnrD+0];
 336
 337             /**************************
 338              * CALCULATE INTERACTIONS *
 339              **************************/
 340
 341             r00              = _mm_mul_ps(rsq00,rinv00);
 342             r00              = _mm_andnot_ps(dummy_mask,r00);
 343
 344             /* Compute parameters for interactions between i and j atoms */
 345             qq00             = _mm_mul_ps(iq0,jq0);
 346             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 347                                          vdwparam+vdwioffset0+vdwjidx0B,
 348                                          vdwparam+vdwioffset0+vdwjidx0C,
 349                                          vdwparam+vdwioffset0+vdwjidx0D,
 350                                          &c6_00,&c12_00);
 351
 352             /* Calculate table index by multiplying r with table scale and truncate to integer */
 353             rt               = _mm_mul_ps(r00,vftabscale);
 354             vfitab           = _mm_cvttps_epi32(rt);
 355             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 356             vfitab           = _mm_slli_epi32(vfitab,3);
 357
 358             /* REACTION-FIELD ELECTROSTATICS */
 359             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 360             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 361
 362             /* CUBIC SPLINE TABLE DISPERSION */
 363             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 364             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 365             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 366             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 367             _MM_TRANSPOSE4_PS(Y,F,G,H);
 368             Heps             = _mm_mul_ps(vfeps,H);
 369             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 370             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 371             vvdw6            = _mm_mul_ps(c6_00,VV);
 372             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 373             fvdw6            = _mm_mul_ps(c6_00,FF);
 374
 375             /* CUBIC SPLINE TABLE REPULSION */
 376             vfitab           = _mm_add_epi32(vfitab,ifour);
 377             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 378             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 379             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 380             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 381             _MM_TRANSPOSE4_PS(Y,F,G,H);
 382             Heps             = _mm_mul_ps(vfeps,H);
 383             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 384             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 385             vvdw12           = _mm_mul_ps(c12_00,VV);
 386             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 387             fvdw12           = _mm_mul_ps(c12_00,FF);
 388             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 389             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velec            = _mm_andnot_ps(dummy_mask,velec);
 393             velecsum         = _mm_add_ps(velecsum,velec);
 394             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 395             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 396
 397             fscal            = _mm_add_ps(felec,fvdw);
 398
 399             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 400
 401             /* Calculate temporary vectorial force */
 402             tx               = _mm_mul_ps(fscal,dx00);
 403             ty               = _mm_mul_ps(fscal,dy00);
 404             tz               = _mm_mul_ps(fscal,dz00);
 405
 406             /* Update vectorial force */
 407             fix0             = _mm_add_ps(fix0,tx);
 408             fiy0             = _mm_add_ps(fiy0,ty);
 409             fiz0             = _mm_add_ps(fiz0,tz);
 410
 411             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 412             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 413             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 414             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 415             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 416
 417             /* Inner loop uses 68 flops */
 418         }
 419
 420         /* End of innermost loop */
 421
 422         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 423                                               f+i_coord_offset,fshift+i_shift_offset);
 424
 425         ggid                        = gid[iidx];
 426         /* Update potential energies */
 427         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 428         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 429
 430         /* Increment number of inner iterations */
 431         inneriter                  += j_index_end - j_index_start;
 432
 433         /* Outer loop uses 9 flops */
 434     }
 435
 436     /* Increment number of outer iterations */
 437     outeriter        += nri;
 438
 439     /* Update outer/inner flops */
 440
 441     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 442 }
 443 /*
 444  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse4_1_single
 445  * Electrostatics interaction: ReactionField
 446  * VdW interaction:            CubicSplineTable
 447  * Geometry:                   Particle-Particle
 448  * Calculate force/pot:        Force
 449  */
 450 void
 451 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse4_1_single
 452                     (t_nblist                    * gmx_restrict       nlist,
 453                      rvec                        * gmx_restrict          xx,
 454                      rvec                        * gmx_restrict          ff,
 455                      t_forcerec                  * gmx_restrict          fr,
 456                      t_mdatoms                   * gmx_restrict     mdatoms,
 457                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 458                      t_nrnb                      * gmx_restrict        nrnb)
 459 {
 460     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 461      * just 0 for non-waters.
 462      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 463      * jnr indices corresponding to data put in the four positions in the SIMD register.
 464      */
 465     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 466     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 467     int              jnrA,jnrB,jnrC,jnrD;
 468     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 469     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 470     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 471     real             rcutoff_scalar;
 472     real             *shiftvec,*fshift,*x,*f;
 473     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 474     real             scratch[4*DIM];
 475     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 476     int              vdwioffset0;
 477     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 478     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 479     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 480     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 481     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 482     real             *charge;
 483     int              nvdwtype;
 484     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 485     int              *vdwtype;
 486     real             *vdwparam;
 487     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 488     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 489     __m128i          vfitab;
 490     __m128i          ifour       = _mm_set1_epi32(4);
 491     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 492     real             *vftab;
 493     __m128           dummy_mask,cutoff_mask;
 494     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 495     __m128           one     = _mm_set1_ps(1.0);
 496     __m128           two     = _mm_set1_ps(2.0);
 497     x                = xx[0];
 498     f                = ff[0];
 499
 500     nri              = nlist->nri;
 501     iinr             = nlist->iinr;
 502     jindex           = nlist->jindex;
 503     jjnr             = nlist->jjnr;
 504     shiftidx         = nlist->shift;
 505     gid              = nlist->gid;
 506     shiftvec         = fr->shift_vec[0];
 507     fshift           = fr->fshift[0];
 508     facel            = _mm_set1_ps(fr->epsfac);
 509     charge           = mdatoms->chargeA;
 510     krf              = _mm_set1_ps(fr->ic->k_rf);
 511     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 512     crf              = _mm_set1_ps(fr->ic->c_rf);
 513     nvdwtype         = fr->ntype;
 514     vdwparam         = fr->nbfp;
 515     vdwtype          = mdatoms->typeA;
 516
 517     vftab            = kernel_data->table_vdw->data;
 518     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 519
 520     /* Avoid stupid compiler warnings */
 521     jnrA = jnrB = jnrC = jnrD = 0;
 522     j_coord_offsetA = 0;
 523     j_coord_offsetB = 0;
 524     j_coord_offsetC = 0;
 525     j_coord_offsetD = 0;
 526
 527     outeriter        = 0;
 528     inneriter        = 0;
 529
 530     for(iidx=0;iidx<4*DIM;iidx++)
 531     {
 532         scratch[iidx] = 0.0;
 533     }
 534
 535     /* Start outer loop over neighborlists */
 536     for(iidx=0; iidx<nri; iidx++)
 537     {
 538         /* Load shift vector for this list */
 539         i_shift_offset   = DIM*shiftidx[iidx];
 540
 541         /* Load limits for loop over neighbors */
 542         j_index_start    = jindex[iidx];
 543         j_index_end      = jindex[iidx+1];
 544
 545         /* Get outer coordinate index */
 546         inr              = iinr[iidx];
 547         i_coord_offset   = DIM*inr;
 548
 549         /* Load i particle coords and add shift vector */
 550         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 551
 552         fix0             = _mm_setzero_ps();
 553         fiy0             = _mm_setzero_ps();
 554         fiz0             = _mm_setzero_ps();
 555
 556         /* Load parameters for i particles */
 557         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 558         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 559
 560         /* Start inner kernel loop */
 561         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 562         {
 563
 564             /* Get j neighbor index, and coordinate index */
 565             jnrA             = jjnr[jidx];
 566             jnrB             = jjnr[jidx+1];
 567             jnrC             = jjnr[jidx+2];
 568             jnrD             = jjnr[jidx+3];
 569             j_coord_offsetA  = DIM*jnrA;
 570             j_coord_offsetB  = DIM*jnrB;
 571             j_coord_offsetC  = DIM*jnrC;
 572             j_coord_offsetD  = DIM*jnrD;
 573
 574             /* load j atom coordinates */
 575             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 576                                               x+j_coord_offsetC,x+j_coord_offsetD,
 577                                               &jx0,&jy0,&jz0);
 578
 579             /* Calculate displacement vector */
 580             dx00             = _mm_sub_ps(ix0,jx0);
 581             dy00             = _mm_sub_ps(iy0,jy0);
 582             dz00             = _mm_sub_ps(iz0,jz0);
 583
 584             /* Calculate squared distance and things based on it */
 585             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 586
 587             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 588
 589             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 590
 591             /* Load parameters for j particles */
 592             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 593                                                               charge+jnrC+0,charge+jnrD+0);
 594             vdwjidx0A        = 2*vdwtype[jnrA+0];
 595             vdwjidx0B        = 2*vdwtype[jnrB+0];
 596             vdwjidx0C        = 2*vdwtype[jnrC+0];
 597             vdwjidx0D        = 2*vdwtype[jnrD+0];
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             r00              = _mm_mul_ps(rsq00,rinv00);
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq00             = _mm_mul_ps(iq0,jq0);
 607             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 608                                          vdwparam+vdwioffset0+vdwjidx0B,
 609                                          vdwparam+vdwioffset0+vdwjidx0C,
 610                                          vdwparam+vdwioffset0+vdwjidx0D,
 611                                          &c6_00,&c12_00);
 612
 613             /* Calculate table index by multiplying r with table scale and truncate to integer */
 614             rt               = _mm_mul_ps(r00,vftabscale);
 615             vfitab           = _mm_cvttps_epi32(rt);
 616             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 617             vfitab           = _mm_slli_epi32(vfitab,3);
 618
 619             /* REACTION-FIELD ELECTROSTATICS */
 620             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 621
 622             /* CUBIC SPLINE TABLE DISPERSION */
 623             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 624             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 625             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 626             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 627             _MM_TRANSPOSE4_PS(Y,F,G,H);
 628             Heps             = _mm_mul_ps(vfeps,H);
 629             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 630             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 631             fvdw6            = _mm_mul_ps(c6_00,FF);
 632
 633             /* CUBIC SPLINE TABLE REPULSION */
 634             vfitab           = _mm_add_epi32(vfitab,ifour);
 635             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 636             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 637             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 638             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 639             _MM_TRANSPOSE4_PS(Y,F,G,H);
 640             Heps             = _mm_mul_ps(vfeps,H);
 641             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 642             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 643             fvdw12           = _mm_mul_ps(c12_00,FF);
 644             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 645
 646             fscal            = _mm_add_ps(felec,fvdw);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm_mul_ps(fscal,dx00);
 650             ty               = _mm_mul_ps(fscal,dy00);
 651             tz               = _mm_mul_ps(fscal,dz00);
 652
 653             /* Update vectorial force */
 654             fix0             = _mm_add_ps(fix0,tx);
 655             fiy0             = _mm_add_ps(fiy0,ty);
 656             fiz0             = _mm_add_ps(fiz0,tz);
 657
 658             fjptrA             = f+j_coord_offsetA;
 659             fjptrB             = f+j_coord_offsetB;
 660             fjptrC             = f+j_coord_offsetC;
 661             fjptrD             = f+j_coord_offsetD;
 662             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 663
 664             /* Inner loop uses 54 flops */
 665         }
 666
 667         if(jidx<j_index_end)
 668         {
 669
 670             /* Get j neighbor index, and coordinate index */
 671             jnrlistA         = jjnr[jidx];
 672             jnrlistB         = jjnr[jidx+1];
 673             jnrlistC         = jjnr[jidx+2];
 674             jnrlistD         = jjnr[jidx+3];
 675             /* Sign of each element will be negative for non-real atoms.
 676              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 677              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 678              */
 679             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 680             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 681             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 682             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 683             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 684             j_coord_offsetA  = DIM*jnrA;
 685             j_coord_offsetB  = DIM*jnrB;
 686             j_coord_offsetC  = DIM*jnrC;
 687             j_coord_offsetD  = DIM*jnrD;
 688
 689             /* load j atom coordinates */
 690             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 691                                               x+j_coord_offsetC,x+j_coord_offsetD,
 692                                               &jx0,&jy0,&jz0);
 693
 694             /* Calculate displacement vector */
 695             dx00             = _mm_sub_ps(ix0,jx0);
 696             dy00             = _mm_sub_ps(iy0,jy0);
 697             dz00             = _mm_sub_ps(iz0,jz0);
 698
 699             /* Calculate squared distance and things based on it */
 700             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 701
 702             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 703
 704             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 705
 706             /* Load parameters for j particles */
 707             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 708                                                               charge+jnrC+0,charge+jnrD+0);
 709             vdwjidx0A        = 2*vdwtype[jnrA+0];
 710             vdwjidx0B        = 2*vdwtype[jnrB+0];
 711             vdwjidx0C        = 2*vdwtype[jnrC+0];
 712             vdwjidx0D        = 2*vdwtype[jnrD+0];
 713
 714             /**************************
 715              * CALCULATE INTERACTIONS *
 716              **************************/
 717
 718             r00              = _mm_mul_ps(rsq00,rinv00);
 719             r00              = _mm_andnot_ps(dummy_mask,r00);
 720
 721             /* Compute parameters for interactions between i and j atoms */
 722             qq00             = _mm_mul_ps(iq0,jq0);
 723             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 724                                          vdwparam+vdwioffset0+vdwjidx0B,
 725                                          vdwparam+vdwioffset0+vdwjidx0C,
 726                                          vdwparam+vdwioffset0+vdwjidx0D,
 727                                          &c6_00,&c12_00);
 728
 729             /* Calculate table index by multiplying r with table scale and truncate to integer */
 730             rt               = _mm_mul_ps(r00,vftabscale);
 731             vfitab           = _mm_cvttps_epi32(rt);
 732             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 733             vfitab           = _mm_slli_epi32(vfitab,3);
 734
 735             /* REACTION-FIELD ELECTROSTATICS */
 736             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 737
 738             /* CUBIC SPLINE TABLE DISPERSION */
 739             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 740             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 741             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 742             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 743             _MM_TRANSPOSE4_PS(Y,F,G,H);
 744             Heps             = _mm_mul_ps(vfeps,H);
 745             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 746             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 747             fvdw6            = _mm_mul_ps(c6_00,FF);
 748
 749             /* CUBIC SPLINE TABLE REPULSION */
 750             vfitab           = _mm_add_epi32(vfitab,ifour);
 751             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 752             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 753             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 754             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 755             _MM_TRANSPOSE4_PS(Y,F,G,H);
 756             Heps             = _mm_mul_ps(vfeps,H);
 757             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 758             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 759             fvdw12           = _mm_mul_ps(c12_00,FF);
 760             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 761
 762             fscal            = _mm_add_ps(felec,fvdw);
 763
 764             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 765
 766             /* Calculate temporary vectorial force */
 767             tx               = _mm_mul_ps(fscal,dx00);
 768             ty               = _mm_mul_ps(fscal,dy00);
 769             tz               = _mm_mul_ps(fscal,dz00);
 770
 771             /* Update vectorial force */
 772             fix0             = _mm_add_ps(fix0,tx);
 773             fiy0             = _mm_add_ps(fiy0,ty);
 774             fiz0             = _mm_add_ps(fiz0,tz);
 775
 776             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 777             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 778             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 779             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 780             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 781
 782             /* Inner loop uses 55 flops */
 783         }
 784
 785         /* End of innermost loop */
 786
 787         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 788                                               f+i_coord_offset,fshift+i_shift_offset);
 789
 790         /* Increment number of inner iterations */
 791         inneriter                  += j_index_end - j_index_start;
 792
 793         /* Outer loop uses 7 flops */
 794     }
 795
 796     /* Increment number of outer iterations */
 797     outeriter        += nri;
 798
 799     /* Update outer/inner flops */
 800
 801     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);
 802 }