src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse2_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  95     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  96     __m128i          vfitab;
  97     __m128i          ifour       = _mm_set1_epi32(4);
  98     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  99     real             *vftab;
 100     __m128           dummy_mask,cutoff_mask;
 101     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 102     __m128           one     = _mm_set1_ps(1.0);
 103     __m128           two     = _mm_set1_ps(2.0);
 104     x                = xx[0];
 105     f                = ff[0];
 106
 107     nri              = nlist->nri;
 108     iinr             = nlist->iinr;
 109     jindex           = nlist->jindex;
 110     jjnr             = nlist->jjnr;
 111     shiftidx         = nlist->shift;
 112     gid              = nlist->gid;
 113     shiftvec         = fr->shift_vec[0];
 114     fshift           = fr->fshift[0];
 115     facel            = _mm_set1_ps(fr->epsfac);
 116     charge           = mdatoms->chargeA;
 117     krf              = _mm_set1_ps(fr->ic->k_rf);
 118     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 119     crf              = _mm_set1_ps(fr->ic->c_rf);
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     vftab            = kernel_data->table_vdw->data;
 125     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162
 163         /* Load parameters for i particles */
 164         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 165         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 166
 167         /* Reset potential sums */
 168         velecsum         = _mm_setzero_ps();
 169         vvdwsum          = _mm_setzero_ps();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               x+j_coord_offsetC,x+j_coord_offsetD,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_ps(ix0,jx0);
 192             dy00             = _mm_sub_ps(iy0,jy0);
 193             dz00             = _mm_sub_ps(iz0,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 197
 198             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 199
 200             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 201
 202             /* Load parameters for j particles */
 203             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 204                                                               charge+jnrC+0,charge+jnrD+0);
 205             vdwjidx0A        = 2*vdwtype[jnrA+0];
 206             vdwjidx0B        = 2*vdwtype[jnrB+0];
 207             vdwjidx0C        = 2*vdwtype[jnrC+0];
 208             vdwjidx0D        = 2*vdwtype[jnrD+0];
 209
 210             /**************************
 211              * CALCULATE INTERACTIONS *
 212              **************************/
 213
 214             r00              = _mm_mul_ps(rsq00,rinv00);
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq00             = _mm_mul_ps(iq0,jq0);
 218             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 219                                          vdwparam+vdwioffset0+vdwjidx0B,
 220                                          vdwparam+vdwioffset0+vdwjidx0C,
 221                                          vdwparam+vdwioffset0+vdwjidx0D,
 222                                          &c6_00,&c12_00);
 223
 224             /* Calculate table index by multiplying r with table scale and truncate to integer */
 225             rt               = _mm_mul_ps(r00,vftabscale);
 226             vfitab           = _mm_cvttps_epi32(rt);
 227             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 228             vfitab           = _mm_slli_epi32(vfitab,3);
 229
 230             /* REACTION-FIELD ELECTROSTATICS */
 231             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 232             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 233
 234             /* CUBIC SPLINE TABLE DISPERSION */
 235             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 236             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 237             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 238             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 239             _MM_TRANSPOSE4_PS(Y,F,G,H);
 240             Heps             = _mm_mul_ps(vfeps,H);
 241             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 242             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 243             vvdw6            = _mm_mul_ps(c6_00,VV);
 244             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 245             fvdw6            = _mm_mul_ps(c6_00,FF);
 246
 247             /* CUBIC SPLINE TABLE REPULSION */
 248             vfitab           = _mm_add_epi32(vfitab,ifour);
 249             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 250             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 251             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 252             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 253             _MM_TRANSPOSE4_PS(Y,F,G,H);
 254             Heps             = _mm_mul_ps(vfeps,H);
 255             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 256             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 257             vvdw12           = _mm_mul_ps(c12_00,VV);
 258             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 259             fvdw12           = _mm_mul_ps(c12_00,FF);
 260             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 261             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velecsum         = _mm_add_ps(velecsum,velec);
 265             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 266
 267             fscal            = _mm_add_ps(felec,fvdw);
 268
 269             /* Calculate temporary vectorial force */
 270             tx               = _mm_mul_ps(fscal,dx00);
 271             ty               = _mm_mul_ps(fscal,dy00);
 272             tz               = _mm_mul_ps(fscal,dz00);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm_add_ps(fix0,tx);
 276             fiy0             = _mm_add_ps(fiy0,ty);
 277             fiz0             = _mm_add_ps(fiz0,tz);
 278
 279             fjptrA             = f+j_coord_offsetA;
 280             fjptrB             = f+j_coord_offsetB;
 281             fjptrC             = f+j_coord_offsetC;
 282             fjptrD             = f+j_coord_offsetD;
 283             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 284
 285             /* Inner loop uses 67 flops */
 286         }
 287
 288         if(jidx<j_index_end)
 289         {
 290
 291             /* Get j neighbor index, and coordinate index */
 292             jnrlistA         = jjnr[jidx];
 293             jnrlistB         = jjnr[jidx+1];
 294             jnrlistC         = jjnr[jidx+2];
 295             jnrlistD         = jjnr[jidx+3];
 296             /* Sign of each element will be negative for non-real atoms.
 297              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 298              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 299              */
 300             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 301             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 302             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 303             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 304             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 305             j_coord_offsetA  = DIM*jnrA;
 306             j_coord_offsetB  = DIM*jnrB;
 307             j_coord_offsetC  = DIM*jnrC;
 308             j_coord_offsetD  = DIM*jnrD;
 309
 310             /* load j atom coordinates */
 311             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 312                                               x+j_coord_offsetC,x+j_coord_offsetD,
 313                                               &jx0,&jy0,&jz0);
 314
 315             /* Calculate displacement vector */
 316             dx00             = _mm_sub_ps(ix0,jx0);
 317             dy00             = _mm_sub_ps(iy0,jy0);
 318             dz00             = _mm_sub_ps(iz0,jz0);
 319
 320             /* Calculate squared distance and things based on it */
 321             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 322
 323             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 324
 325             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 326
 327             /* Load parameters for j particles */
 328             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 329                                                               charge+jnrC+0,charge+jnrD+0);
 330             vdwjidx0A        = 2*vdwtype[jnrA+0];
 331             vdwjidx0B        = 2*vdwtype[jnrB+0];
 332             vdwjidx0C        = 2*vdwtype[jnrC+0];
 333             vdwjidx0D        = 2*vdwtype[jnrD+0];
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             r00              = _mm_mul_ps(rsq00,rinv00);
 340             r00              = _mm_andnot_ps(dummy_mask,r00);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq00             = _mm_mul_ps(iq0,jq0);
 344             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 345                                          vdwparam+vdwioffset0+vdwjidx0B,
 346                                          vdwparam+vdwioffset0+vdwjidx0C,
 347                                          vdwparam+vdwioffset0+vdwjidx0D,
 348                                          &c6_00,&c12_00);
 349
 350             /* Calculate table index by multiplying r with table scale and truncate to integer */
 351             rt               = _mm_mul_ps(r00,vftabscale);
 352             vfitab           = _mm_cvttps_epi32(rt);
 353             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 354             vfitab           = _mm_slli_epi32(vfitab,3);
 355
 356             /* REACTION-FIELD ELECTROSTATICS */
 357             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 358             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 359
 360             /* CUBIC SPLINE TABLE DISPERSION */
 361             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 362             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 363             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 364             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 365             _MM_TRANSPOSE4_PS(Y,F,G,H);
 366             Heps             = _mm_mul_ps(vfeps,H);
 367             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 368             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 369             vvdw6            = _mm_mul_ps(c6_00,VV);
 370             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 371             fvdw6            = _mm_mul_ps(c6_00,FF);
 372
 373             /* CUBIC SPLINE TABLE REPULSION */
 374             vfitab           = _mm_add_epi32(vfitab,ifour);
 375             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 376             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 377             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 378             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 379             _MM_TRANSPOSE4_PS(Y,F,G,H);
 380             Heps             = _mm_mul_ps(vfeps,H);
 381             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 382             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 383             vvdw12           = _mm_mul_ps(c12_00,VV);
 384             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 385             fvdw12           = _mm_mul_ps(c12_00,FF);
 386             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 387             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velec            = _mm_andnot_ps(dummy_mask,velec);
 391             velecsum         = _mm_add_ps(velecsum,velec);
 392             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 393             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 394
 395             fscal            = _mm_add_ps(felec,fvdw);
 396
 397             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm_mul_ps(fscal,dx00);
 401             ty               = _mm_mul_ps(fscal,dy00);
 402             tz               = _mm_mul_ps(fscal,dz00);
 403
 404             /* Update vectorial force */
 405             fix0             = _mm_add_ps(fix0,tx);
 406             fiy0             = _mm_add_ps(fiy0,ty);
 407             fiz0             = _mm_add_ps(fiz0,tz);
 408
 409             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 410             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 411             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 412             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 413             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 414
 415             /* Inner loop uses 68 flops */
 416         }
 417
 418         /* End of innermost loop */
 419
 420         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 421                                               f+i_coord_offset,fshift+i_shift_offset);
 422
 423         ggid                        = gid[iidx];
 424         /* Update potential energies */
 425         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 426         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 427
 428         /* Increment number of inner iterations */
 429         inneriter                  += j_index_end - j_index_start;
 430
 431         /* Outer loop uses 9 flops */
 432     }
 433
 434     /* Increment number of outer iterations */
 435     outeriter        += nri;
 436
 437     /* Update outer/inner flops */
 438
 439     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*68);
 440 }
 441 /*
 442  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_single
 443  * Electrostatics interaction: ReactionField
 444  * VdW interaction:            CubicSplineTable
 445  * Geometry:                   Particle-Particle
 446  * Calculate force/pot:        Force
 447  */
 448 void
 449 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_single
 450                     (t_nblist                    * gmx_restrict       nlist,
 451                      rvec                        * gmx_restrict          xx,
 452                      rvec                        * gmx_restrict          ff,
 453                      t_forcerec                  * gmx_restrict          fr,
 454                      t_mdatoms                   * gmx_restrict     mdatoms,
 455                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 456                      t_nrnb                      * gmx_restrict        nrnb)
 457 {
 458     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 459      * just 0 for non-waters.
 460      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 461      * jnr indices corresponding to data put in the four positions in the SIMD register.
 462      */
 463     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 464     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 465     int              jnrA,jnrB,jnrC,jnrD;
 466     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 467     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 468     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 469     real             rcutoff_scalar;
 470     real             *shiftvec,*fshift,*x,*f;
 471     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 472     real             scratch[4*DIM];
 473     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 474     int              vdwioffset0;
 475     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 476     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 477     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 478     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 479     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 480     real             *charge;
 481     int              nvdwtype;
 482     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 483     int              *vdwtype;
 484     real             *vdwparam;
 485     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 486     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 487     __m128i          vfitab;
 488     __m128i          ifour       = _mm_set1_epi32(4);
 489     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 490     real             *vftab;
 491     __m128           dummy_mask,cutoff_mask;
 492     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 493     __m128           one     = _mm_set1_ps(1.0);
 494     __m128           two     = _mm_set1_ps(2.0);
 495     x                = xx[0];
 496     f                = ff[0];
 497
 498     nri              = nlist->nri;
 499     iinr             = nlist->iinr;
 500     jindex           = nlist->jindex;
 501     jjnr             = nlist->jjnr;
 502     shiftidx         = nlist->shift;
 503     gid              = nlist->gid;
 504     shiftvec         = fr->shift_vec[0];
 505     fshift           = fr->fshift[0];
 506     facel            = _mm_set1_ps(fr->epsfac);
 507     charge           = mdatoms->chargeA;
 508     krf              = _mm_set1_ps(fr->ic->k_rf);
 509     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 510     crf              = _mm_set1_ps(fr->ic->c_rf);
 511     nvdwtype         = fr->ntype;
 512     vdwparam         = fr->nbfp;
 513     vdwtype          = mdatoms->typeA;
 514
 515     vftab            = kernel_data->table_vdw->data;
 516     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 517
 518     /* Avoid stupid compiler warnings */
 519     jnrA = jnrB = jnrC = jnrD = 0;
 520     j_coord_offsetA = 0;
 521     j_coord_offsetB = 0;
 522     j_coord_offsetC = 0;
 523     j_coord_offsetD = 0;
 524
 525     outeriter        = 0;
 526     inneriter        = 0;
 527
 528     for(iidx=0;iidx<4*DIM;iidx++)
 529     {
 530         scratch[iidx] = 0.0;
 531     }
 532
 533     /* Start outer loop over neighborlists */
 534     for(iidx=0; iidx<nri; iidx++)
 535     {
 536         /* Load shift vector for this list */
 537         i_shift_offset   = DIM*shiftidx[iidx];
 538
 539         /* Load limits for loop over neighbors */
 540         j_index_start    = jindex[iidx];
 541         j_index_end      = jindex[iidx+1];
 542
 543         /* Get outer coordinate index */
 544         inr              = iinr[iidx];
 545         i_coord_offset   = DIM*inr;
 546
 547         /* Load i particle coords and add shift vector */
 548         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 549
 550         fix0             = _mm_setzero_ps();
 551         fiy0             = _mm_setzero_ps();
 552         fiz0             = _mm_setzero_ps();
 553
 554         /* Load parameters for i particles */
 555         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 556         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 557
 558         /* Start inner kernel loop */
 559         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 560         {
 561
 562             /* Get j neighbor index, and coordinate index */
 563             jnrA             = jjnr[jidx];
 564             jnrB             = jjnr[jidx+1];
 565             jnrC             = jjnr[jidx+2];
 566             jnrD             = jjnr[jidx+3];
 567             j_coord_offsetA  = DIM*jnrA;
 568             j_coord_offsetB  = DIM*jnrB;
 569             j_coord_offsetC  = DIM*jnrC;
 570             j_coord_offsetD  = DIM*jnrD;
 571
 572             /* load j atom coordinates */
 573             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 574                                               x+j_coord_offsetC,x+j_coord_offsetD,
 575                                               &jx0,&jy0,&jz0);
 576
 577             /* Calculate displacement vector */
 578             dx00             = _mm_sub_ps(ix0,jx0);
 579             dy00             = _mm_sub_ps(iy0,jy0);
 580             dz00             = _mm_sub_ps(iz0,jz0);
 581
 582             /* Calculate squared distance and things based on it */
 583             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 584
 585             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 586
 587             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 588
 589             /* Load parameters for j particles */
 590             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 591                                                               charge+jnrC+0,charge+jnrD+0);
 592             vdwjidx0A        = 2*vdwtype[jnrA+0];
 593             vdwjidx0B        = 2*vdwtype[jnrB+0];
 594             vdwjidx0C        = 2*vdwtype[jnrC+0];
 595             vdwjidx0D        = 2*vdwtype[jnrD+0];
 596
 597             /**************************
 598              * CALCULATE INTERACTIONS *
 599              **************************/
 600
 601             r00              = _mm_mul_ps(rsq00,rinv00);
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq00             = _mm_mul_ps(iq0,jq0);
 605             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 606                                          vdwparam+vdwioffset0+vdwjidx0B,
 607                                          vdwparam+vdwioffset0+vdwjidx0C,
 608                                          vdwparam+vdwioffset0+vdwjidx0D,
 609                                          &c6_00,&c12_00);
 610
 611             /* Calculate table index by multiplying r with table scale and truncate to integer */
 612             rt               = _mm_mul_ps(r00,vftabscale);
 613             vfitab           = _mm_cvttps_epi32(rt);
 614             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 615             vfitab           = _mm_slli_epi32(vfitab,3);
 616
 617             /* REACTION-FIELD ELECTROSTATICS */
 618             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 619
 620             /* CUBIC SPLINE TABLE DISPERSION */
 621             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 622             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 623             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 624             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 625             _MM_TRANSPOSE4_PS(Y,F,G,H);
 626             Heps             = _mm_mul_ps(vfeps,H);
 627             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 628             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 629             fvdw6            = _mm_mul_ps(c6_00,FF);
 630
 631             /* CUBIC SPLINE TABLE REPULSION */
 632             vfitab           = _mm_add_epi32(vfitab,ifour);
 633             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 634             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 635             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 636             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 637             _MM_TRANSPOSE4_PS(Y,F,G,H);
 638             Heps             = _mm_mul_ps(vfeps,H);
 639             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 640             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 641             fvdw12           = _mm_mul_ps(c12_00,FF);
 642             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 643
 644             fscal            = _mm_add_ps(felec,fvdw);
 645
 646             /* Calculate temporary vectorial force */
 647             tx               = _mm_mul_ps(fscal,dx00);
 648             ty               = _mm_mul_ps(fscal,dy00);
 649             tz               = _mm_mul_ps(fscal,dz00);
 650
 651             /* Update vectorial force */
 652             fix0             = _mm_add_ps(fix0,tx);
 653             fiy0             = _mm_add_ps(fiy0,ty);
 654             fiz0             = _mm_add_ps(fiz0,tz);
 655
 656             fjptrA             = f+j_coord_offsetA;
 657             fjptrB             = f+j_coord_offsetB;
 658             fjptrC             = f+j_coord_offsetC;
 659             fjptrD             = f+j_coord_offsetD;
 660             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 661
 662             /* Inner loop uses 54 flops */
 663         }
 664
 665         if(jidx<j_index_end)
 666         {
 667
 668             /* Get j neighbor index, and coordinate index */
 669             jnrlistA         = jjnr[jidx];
 670             jnrlistB         = jjnr[jidx+1];
 671             jnrlistC         = jjnr[jidx+2];
 672             jnrlistD         = jjnr[jidx+3];
 673             /* Sign of each element will be negative for non-real atoms.
 674              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 675              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 676              */
 677             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 678             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 679             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 680             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 681             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 682             j_coord_offsetA  = DIM*jnrA;
 683             j_coord_offsetB  = DIM*jnrB;
 684             j_coord_offsetC  = DIM*jnrC;
 685             j_coord_offsetD  = DIM*jnrD;
 686
 687             /* load j atom coordinates */
 688             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 689                                               x+j_coord_offsetC,x+j_coord_offsetD,
 690                                               &jx0,&jy0,&jz0);
 691
 692             /* Calculate displacement vector */
 693             dx00             = _mm_sub_ps(ix0,jx0);
 694             dy00             = _mm_sub_ps(iy0,jy0);
 695             dz00             = _mm_sub_ps(iz0,jz0);
 696
 697             /* Calculate squared distance and things based on it */
 698             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 699
 700             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 701
 702             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 703
 704             /* Load parameters for j particles */
 705             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 706                                                               charge+jnrC+0,charge+jnrD+0);
 707             vdwjidx0A        = 2*vdwtype[jnrA+0];
 708             vdwjidx0B        = 2*vdwtype[jnrB+0];
 709             vdwjidx0C        = 2*vdwtype[jnrC+0];
 710             vdwjidx0D        = 2*vdwtype[jnrD+0];
 711
 712             /**************************
 713              * CALCULATE INTERACTIONS *
 714              **************************/
 715
 716             r00              = _mm_mul_ps(rsq00,rinv00);
 717             r00              = _mm_andnot_ps(dummy_mask,r00);
 718
 719             /* Compute parameters for interactions between i and j atoms */
 720             qq00             = _mm_mul_ps(iq0,jq0);
 721             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 722                                          vdwparam+vdwioffset0+vdwjidx0B,
 723                                          vdwparam+vdwioffset0+vdwjidx0C,
 724                                          vdwparam+vdwioffset0+vdwjidx0D,
 725                                          &c6_00,&c12_00);
 726
 727             /* Calculate table index by multiplying r with table scale and truncate to integer */
 728             rt               = _mm_mul_ps(r00,vftabscale);
 729             vfitab           = _mm_cvttps_epi32(rt);
 730             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 731             vfitab           = _mm_slli_epi32(vfitab,3);
 732
 733             /* REACTION-FIELD ELECTROSTATICS */
 734             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 735
 736             /* CUBIC SPLINE TABLE DISPERSION */
 737             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 738             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 739             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 740             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 741             _MM_TRANSPOSE4_PS(Y,F,G,H);
 742             Heps             = _mm_mul_ps(vfeps,H);
 743             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 744             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 745             fvdw6            = _mm_mul_ps(c6_00,FF);
 746
 747             /* CUBIC SPLINE TABLE REPULSION */
 748             vfitab           = _mm_add_epi32(vfitab,ifour);
 749             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 750             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 751             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 752             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 753             _MM_TRANSPOSE4_PS(Y,F,G,H);
 754             Heps             = _mm_mul_ps(vfeps,H);
 755             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 756             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 757             fvdw12           = _mm_mul_ps(c12_00,FF);
 758             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 759
 760             fscal            = _mm_add_ps(felec,fvdw);
 761
 762             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 763
 764             /* Calculate temporary vectorial force */
 765             tx               = _mm_mul_ps(fscal,dx00);
 766             ty               = _mm_mul_ps(fscal,dy00);
 767             tz               = _mm_mul_ps(fscal,dz00);
 768
 769             /* Update vectorial force */
 770             fix0             = _mm_add_ps(fix0,tx);
 771             fiy0             = _mm_add_ps(fiy0,ty);
 772             fiz0             = _mm_add_ps(fiz0,tz);
 773
 774             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 775             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 776             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 777             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 778             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 779
 780             /* Inner loop uses 55 flops */
 781         }
 782
 783         /* End of innermost loop */
 784
 785         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 786                                               f+i_coord_offset,fshift+i_shift_offset);
 787
 788         /* Increment number of inner iterations */
 789         inneriter                  += j_index_end - j_index_start;
 790
 791         /* Outer loop uses 7 flops */
 792     }
 793
 794     /* Increment number of outer iterations */
 795     outeriter        += nri;
 796
 797     /* Update outer/inner flops */
 798
 799     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*55);
 800 }