src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_sse2_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 sse2_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 "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_sse2_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 130     rcutoff_scalar   = fr->rcoulomb;
 131     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 132     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 165
 166         fix0             = _mm_setzero_ps();
 167         fiy0             = _mm_setzero_ps();
 168         fiz0             = _mm_setzero_ps();
 169
 170         /* Load parameters for i particles */
 171         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 172         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vvdwsum          = _mm_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_ps(ix0,jx0);
 199             dy00             = _mm_sub_ps(iy0,jy0);
 200             dz00             = _mm_sub_ps(iz0,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 204
 205             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 206
 207             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 211                                                               charge+jnrC+0,charge+jnrD+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214             vdwjidx0C        = 2*vdwtype[jnrC+0];
 215             vdwjidx0D        = 2*vdwtype[jnrD+0];
 216
 217             /**************************
 218              * CALCULATE INTERACTIONS *
 219              **************************/
 220
 221             if (gmx_mm_any_lt(rsq00,rcutoff2))
 222             {
 223
 224             r00              = _mm_mul_ps(rsq00,rinv00);
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq00             = _mm_mul_ps(iq0,jq0);
 228             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 229                                          vdwparam+vdwioffset0+vdwjidx0B,
 230                                          vdwparam+vdwioffset0+vdwjidx0C,
 231                                          vdwparam+vdwioffset0+vdwjidx0D,
 232                                          &c6_00,&c12_00);
 233
 234             /* Calculate table index by multiplying r with table scale and truncate to integer */
 235             rt               = _mm_mul_ps(r00,vftabscale);
 236             vfitab           = _mm_cvttps_epi32(rt);
 237             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 238             vfitab           = _mm_slli_epi32(vfitab,3);
 239
 240             /* REACTION-FIELD ELECTROSTATICS */
 241             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 242             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 243
 244             /* CUBIC SPLINE TABLE DISPERSION */
 245             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 246             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 247             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 248             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 249             _MM_TRANSPOSE4_PS(Y,F,G,H);
 250             Heps             = _mm_mul_ps(vfeps,H);
 251             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 252             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 253             vvdw6            = _mm_mul_ps(c6_00,VV);
 254             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 255             fvdw6            = _mm_mul_ps(c6_00,FF);
 256
 257             /* CUBIC SPLINE TABLE REPULSION */
 258             vfitab           = _mm_add_epi32(vfitab,ifour);
 259             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 260             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 261             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 262             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 263             _MM_TRANSPOSE4_PS(Y,F,G,H);
 264             Heps             = _mm_mul_ps(vfeps,H);
 265             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 266             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 267             vvdw12           = _mm_mul_ps(c12_00,VV);
 268             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 269             fvdw12           = _mm_mul_ps(c12_00,FF);
 270             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 271             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 272
 273             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm_and_ps(velec,cutoff_mask);
 277             velecsum         = _mm_add_ps(velecsum,velec);
 278             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 279             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 280
 281             fscal            = _mm_add_ps(felec,fvdw);
 282
 283             fscal            = _mm_and_ps(fscal,cutoff_mask);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_ps(fscal,dx00);
 287             ty               = _mm_mul_ps(fscal,dy00);
 288             tz               = _mm_mul_ps(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_add_ps(fix0,tx);
 292             fiy0             = _mm_add_ps(fiy0,ty);
 293             fiz0             = _mm_add_ps(fiz0,tz);
 294
 295             fjptrA             = f+j_coord_offsetA;
 296             fjptrB             = f+j_coord_offsetB;
 297             fjptrC             = f+j_coord_offsetC;
 298             fjptrD             = f+j_coord_offsetD;
 299             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 300
 301             }
 302
 303             /* Inner loop uses 72 flops */
 304         }
 305
 306         if(jidx<j_index_end)
 307         {
 308
 309             /* Get j neighbor index, and coordinate index */
 310             jnrlistA         = jjnr[jidx];
 311             jnrlistB         = jjnr[jidx+1];
 312             jnrlistC         = jjnr[jidx+2];
 313             jnrlistD         = jjnr[jidx+3];
 314             /* Sign of each element will be negative for non-real atoms.
 315              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 316              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 317              */
 318             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 319             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 320             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 321             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 322             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 323             j_coord_offsetA  = DIM*jnrA;
 324             j_coord_offsetB  = DIM*jnrB;
 325             j_coord_offsetC  = DIM*jnrC;
 326             j_coord_offsetD  = DIM*jnrD;
 327
 328             /* load j atom coordinates */
 329             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 330                                               x+j_coord_offsetC,x+j_coord_offsetD,
 331                                               &jx0,&jy0,&jz0);
 332
 333             /* Calculate displacement vector */
 334             dx00             = _mm_sub_ps(ix0,jx0);
 335             dy00             = _mm_sub_ps(iy0,jy0);
 336             dz00             = _mm_sub_ps(iz0,jz0);
 337
 338             /* Calculate squared distance and things based on it */
 339             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 340
 341             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 342
 343             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 344
 345             /* Load parameters for j particles */
 346             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 347                                                               charge+jnrC+0,charge+jnrD+0);
 348             vdwjidx0A        = 2*vdwtype[jnrA+0];
 349             vdwjidx0B        = 2*vdwtype[jnrB+0];
 350             vdwjidx0C        = 2*vdwtype[jnrC+0];
 351             vdwjidx0D        = 2*vdwtype[jnrD+0];
 352
 353             /**************************
 354              * CALCULATE INTERACTIONS *
 355              **************************/
 356
 357             if (gmx_mm_any_lt(rsq00,rcutoff2))
 358             {
 359
 360             r00              = _mm_mul_ps(rsq00,rinv00);
 361             r00              = _mm_andnot_ps(dummy_mask,r00);
 362
 363             /* Compute parameters for interactions between i and j atoms */
 364             qq00             = _mm_mul_ps(iq0,jq0);
 365             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 366                                          vdwparam+vdwioffset0+vdwjidx0B,
 367                                          vdwparam+vdwioffset0+vdwjidx0C,
 368                                          vdwparam+vdwioffset0+vdwjidx0D,
 369                                          &c6_00,&c12_00);
 370
 371             /* Calculate table index by multiplying r with table scale and truncate to integer */
 372             rt               = _mm_mul_ps(r00,vftabscale);
 373             vfitab           = _mm_cvttps_epi32(rt);
 374             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 375             vfitab           = _mm_slli_epi32(vfitab,3);
 376
 377             /* REACTION-FIELD ELECTROSTATICS */
 378             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 379             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 380
 381             /* CUBIC SPLINE TABLE DISPERSION */
 382             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 383             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 384             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 385             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 386             _MM_TRANSPOSE4_PS(Y,F,G,H);
 387             Heps             = _mm_mul_ps(vfeps,H);
 388             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 389             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 390             vvdw6            = _mm_mul_ps(c6_00,VV);
 391             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 392             fvdw6            = _mm_mul_ps(c6_00,FF);
 393
 394             /* CUBIC SPLINE TABLE REPULSION */
 395             vfitab           = _mm_add_epi32(vfitab,ifour);
 396             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 397             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 398             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 399             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 400             _MM_TRANSPOSE4_PS(Y,F,G,H);
 401             Heps             = _mm_mul_ps(vfeps,H);
 402             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 403             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 404             vvdw12           = _mm_mul_ps(c12_00,VV);
 405             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 406             fvdw12           = _mm_mul_ps(c12_00,FF);
 407             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 408             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 409
 410             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 411
 412             /* Update potential sum for this i atom from the interaction with this j atom. */
 413             velec            = _mm_and_ps(velec,cutoff_mask);
 414             velec            = _mm_andnot_ps(dummy_mask,velec);
 415             velecsum         = _mm_add_ps(velecsum,velec);
 416             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 417             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 418             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 419
 420             fscal            = _mm_add_ps(felec,fvdw);
 421
 422             fscal            = _mm_and_ps(fscal,cutoff_mask);
 423
 424             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 425
 426             /* Calculate temporary vectorial force */
 427             tx               = _mm_mul_ps(fscal,dx00);
 428             ty               = _mm_mul_ps(fscal,dy00);
 429             tz               = _mm_mul_ps(fscal,dz00);
 430
 431             /* Update vectorial force */
 432             fix0             = _mm_add_ps(fix0,tx);
 433             fiy0             = _mm_add_ps(fiy0,ty);
 434             fiz0             = _mm_add_ps(fiz0,tz);
 435
 436             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 437             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 438             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 439             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 440             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 441
 442             }
 443
 444             /* Inner loop uses 73 flops */
 445         }
 446
 447         /* End of innermost loop */
 448
 449         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 450                                               f+i_coord_offset,fshift+i_shift_offset);
 451
 452         ggid                        = gid[iidx];
 453         /* Update potential energies */
 454         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 455         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 456
 457         /* Increment number of inner iterations */
 458         inneriter                  += j_index_end - j_index_start;
 459
 460         /* Outer loop uses 9 flops */
 461     }
 462
 463     /* Increment number of outer iterations */
 464     outeriter        += nri;
 465
 466     /* Update outer/inner flops */
 467
 468     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 469 }
 470 /*
 471  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse2_single
 472  * Electrostatics interaction: ReactionField
 473  * VdW interaction:            CubicSplineTable
 474  * Geometry:                   Particle-Particle
 475  * Calculate force/pot:        Force
 476  */
 477 void
 478 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse2_single
 479                     (t_nblist                    * gmx_restrict       nlist,
 480                      rvec                        * gmx_restrict          xx,
 481                      rvec                        * gmx_restrict          ff,
 482                      t_forcerec                  * gmx_restrict          fr,
 483                      t_mdatoms                   * gmx_restrict     mdatoms,
 484                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 485                      t_nrnb                      * gmx_restrict        nrnb)
 486 {
 487     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 488      * just 0 for non-waters.
 489      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 490      * jnr indices corresponding to data put in the four positions in the SIMD register.
 491      */
 492     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 493     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 494     int              jnrA,jnrB,jnrC,jnrD;
 495     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 496     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 497     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 498     real             rcutoff_scalar;
 499     real             *shiftvec,*fshift,*x,*f;
 500     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 501     real             scratch[4*DIM];
 502     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 503     int              vdwioffset0;
 504     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 505     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 506     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 507     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 508     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 509     real             *charge;
 510     int              nvdwtype;
 511     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 512     int              *vdwtype;
 513     real             *vdwparam;
 514     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 515     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 516     __m128i          vfitab;
 517     __m128i          ifour       = _mm_set1_epi32(4);
 518     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 519     real             *vftab;
 520     __m128           dummy_mask,cutoff_mask;
 521     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 522     __m128           one     = _mm_set1_ps(1.0);
 523     __m128           two     = _mm_set1_ps(2.0);
 524     x                = xx[0];
 525     f                = ff[0];
 526
 527     nri              = nlist->nri;
 528     iinr             = nlist->iinr;
 529     jindex           = nlist->jindex;
 530     jjnr             = nlist->jjnr;
 531     shiftidx         = nlist->shift;
 532     gid              = nlist->gid;
 533     shiftvec         = fr->shift_vec[0];
 534     fshift           = fr->fshift[0];
 535     facel            = _mm_set1_ps(fr->epsfac);
 536     charge           = mdatoms->chargeA;
 537     krf              = _mm_set1_ps(fr->ic->k_rf);
 538     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 539     crf              = _mm_set1_ps(fr->ic->c_rf);
 540     nvdwtype         = fr->ntype;
 541     vdwparam         = fr->nbfp;
 542     vdwtype          = mdatoms->typeA;
 543
 544     vftab            = kernel_data->table_vdw->data;
 545     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 546
 547     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 548     rcutoff_scalar   = fr->rcoulomb;
 549     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 550     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 551
 552     /* Avoid stupid compiler warnings */
 553     jnrA = jnrB = jnrC = jnrD = 0;
 554     j_coord_offsetA = 0;
 555     j_coord_offsetB = 0;
 556     j_coord_offsetC = 0;
 557     j_coord_offsetD = 0;
 558
 559     outeriter        = 0;
 560     inneriter        = 0;
 561
 562     for(iidx=0;iidx<4*DIM;iidx++)
 563     {
 564         scratch[iidx] = 0.0;
 565     }
 566
 567     /* Start outer loop over neighborlists */
 568     for(iidx=0; iidx<nri; iidx++)
 569     {
 570         /* Load shift vector for this list */
 571         i_shift_offset   = DIM*shiftidx[iidx];
 572
 573         /* Load limits for loop over neighbors */
 574         j_index_start    = jindex[iidx];
 575         j_index_end      = jindex[iidx+1];
 576
 577         /* Get outer coordinate index */
 578         inr              = iinr[iidx];
 579         i_coord_offset   = DIM*inr;
 580
 581         /* Load i particle coords and add shift vector */
 582         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 583
 584         fix0             = _mm_setzero_ps();
 585         fiy0             = _mm_setzero_ps();
 586         fiz0             = _mm_setzero_ps();
 587
 588         /* Load parameters for i particles */
 589         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 590         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 591
 592         /* Start inner kernel loop */
 593         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 594         {
 595
 596             /* Get j neighbor index, and coordinate index */
 597             jnrA             = jjnr[jidx];
 598             jnrB             = jjnr[jidx+1];
 599             jnrC             = jjnr[jidx+2];
 600             jnrD             = jjnr[jidx+3];
 601             j_coord_offsetA  = DIM*jnrA;
 602             j_coord_offsetB  = DIM*jnrB;
 603             j_coord_offsetC  = DIM*jnrC;
 604             j_coord_offsetD  = DIM*jnrD;
 605
 606             /* load j atom coordinates */
 607             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 608                                               x+j_coord_offsetC,x+j_coord_offsetD,
 609                                               &jx0,&jy0,&jz0);
 610
 611             /* Calculate displacement vector */
 612             dx00             = _mm_sub_ps(ix0,jx0);
 613             dy00             = _mm_sub_ps(iy0,jy0);
 614             dz00             = _mm_sub_ps(iz0,jz0);
 615
 616             /* Calculate squared distance and things based on it */
 617             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 618
 619             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 620
 621             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 622
 623             /* Load parameters for j particles */
 624             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 625                                                               charge+jnrC+0,charge+jnrD+0);
 626             vdwjidx0A        = 2*vdwtype[jnrA+0];
 627             vdwjidx0B        = 2*vdwtype[jnrB+0];
 628             vdwjidx0C        = 2*vdwtype[jnrC+0];
 629             vdwjidx0D        = 2*vdwtype[jnrD+0];
 630
 631             /**************************
 632              * CALCULATE INTERACTIONS *
 633              **************************/
 634
 635             if (gmx_mm_any_lt(rsq00,rcutoff2))
 636             {
 637
 638             r00              = _mm_mul_ps(rsq00,rinv00);
 639
 640             /* Compute parameters for interactions between i and j atoms */
 641             qq00             = _mm_mul_ps(iq0,jq0);
 642             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 643                                          vdwparam+vdwioffset0+vdwjidx0B,
 644                                          vdwparam+vdwioffset0+vdwjidx0C,
 645                                          vdwparam+vdwioffset0+vdwjidx0D,
 646                                          &c6_00,&c12_00);
 647
 648             /* Calculate table index by multiplying r with table scale and truncate to integer */
 649             rt               = _mm_mul_ps(r00,vftabscale);
 650             vfitab           = _mm_cvttps_epi32(rt);
 651             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 652             vfitab           = _mm_slli_epi32(vfitab,3);
 653
 654             /* REACTION-FIELD ELECTROSTATICS */
 655             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 656
 657             /* CUBIC SPLINE TABLE DISPERSION */
 658             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 659             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 660             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 661             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 662             _MM_TRANSPOSE4_PS(Y,F,G,H);
 663             Heps             = _mm_mul_ps(vfeps,H);
 664             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 665             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 666             fvdw6            = _mm_mul_ps(c6_00,FF);
 667
 668             /* CUBIC SPLINE TABLE REPULSION */
 669             vfitab           = _mm_add_epi32(vfitab,ifour);
 670             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 671             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 672             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 673             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 674             _MM_TRANSPOSE4_PS(Y,F,G,H);
 675             Heps             = _mm_mul_ps(vfeps,H);
 676             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 677             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 678             fvdw12           = _mm_mul_ps(c12_00,FF);
 679             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 680
 681             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 682
 683             fscal            = _mm_add_ps(felec,fvdw);
 684
 685             fscal            = _mm_and_ps(fscal,cutoff_mask);
 686
 687             /* Calculate temporary vectorial force */
 688             tx               = _mm_mul_ps(fscal,dx00);
 689             ty               = _mm_mul_ps(fscal,dy00);
 690             tz               = _mm_mul_ps(fscal,dz00);
 691
 692             /* Update vectorial force */
 693             fix0             = _mm_add_ps(fix0,tx);
 694             fiy0             = _mm_add_ps(fiy0,ty);
 695             fiz0             = _mm_add_ps(fiz0,tz);
 696
 697             fjptrA             = f+j_coord_offsetA;
 698             fjptrB             = f+j_coord_offsetB;
 699             fjptrC             = f+j_coord_offsetC;
 700             fjptrD             = f+j_coord_offsetD;
 701             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 702
 703             }
 704
 705             /* Inner loop uses 57 flops */
 706         }
 707
 708         if(jidx<j_index_end)
 709         {
 710
 711             /* Get j neighbor index, and coordinate index */
 712             jnrlistA         = jjnr[jidx];
 713             jnrlistB         = jjnr[jidx+1];
 714             jnrlistC         = jjnr[jidx+2];
 715             jnrlistD         = jjnr[jidx+3];
 716             /* Sign of each element will be negative for non-real atoms.
 717              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 718              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 719              */
 720             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 721             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 722             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 723             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 724             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 725             j_coord_offsetA  = DIM*jnrA;
 726             j_coord_offsetB  = DIM*jnrB;
 727             j_coord_offsetC  = DIM*jnrC;
 728             j_coord_offsetD  = DIM*jnrD;
 729
 730             /* load j atom coordinates */
 731             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 732                                               x+j_coord_offsetC,x+j_coord_offsetD,
 733                                               &jx0,&jy0,&jz0);
 734
 735             /* Calculate displacement vector */
 736             dx00             = _mm_sub_ps(ix0,jx0);
 737             dy00             = _mm_sub_ps(iy0,jy0);
 738             dz00             = _mm_sub_ps(iz0,jz0);
 739
 740             /* Calculate squared distance and things based on it */
 741             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 742
 743             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 744
 745             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 746
 747             /* Load parameters for j particles */
 748             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 749                                                               charge+jnrC+0,charge+jnrD+0);
 750             vdwjidx0A        = 2*vdwtype[jnrA+0];
 751             vdwjidx0B        = 2*vdwtype[jnrB+0];
 752             vdwjidx0C        = 2*vdwtype[jnrC+0];
 753             vdwjidx0D        = 2*vdwtype[jnrD+0];
 754
 755             /**************************
 756              * CALCULATE INTERACTIONS *
 757              **************************/
 758
 759             if (gmx_mm_any_lt(rsq00,rcutoff2))
 760             {
 761
 762             r00              = _mm_mul_ps(rsq00,rinv00);
 763             r00              = _mm_andnot_ps(dummy_mask,r00);
 764
 765             /* Compute parameters for interactions between i and j atoms */
 766             qq00             = _mm_mul_ps(iq0,jq0);
 767             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 768                                          vdwparam+vdwioffset0+vdwjidx0B,
 769                                          vdwparam+vdwioffset0+vdwjidx0C,
 770                                          vdwparam+vdwioffset0+vdwjidx0D,
 771                                          &c6_00,&c12_00);
 772
 773             /* Calculate table index by multiplying r with table scale and truncate to integer */
 774             rt               = _mm_mul_ps(r00,vftabscale);
 775             vfitab           = _mm_cvttps_epi32(rt);
 776             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 777             vfitab           = _mm_slli_epi32(vfitab,3);
 778
 779             /* REACTION-FIELD ELECTROSTATICS */
 780             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 781
 782             /* CUBIC SPLINE TABLE DISPERSION */
 783             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 784             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 785             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 786             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 787             _MM_TRANSPOSE4_PS(Y,F,G,H);
 788             Heps             = _mm_mul_ps(vfeps,H);
 789             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 790             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 791             fvdw6            = _mm_mul_ps(c6_00,FF);
 792
 793             /* CUBIC SPLINE TABLE REPULSION */
 794             vfitab           = _mm_add_epi32(vfitab,ifour);
 795             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 796             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 797             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 798             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 799             _MM_TRANSPOSE4_PS(Y,F,G,H);
 800             Heps             = _mm_mul_ps(vfeps,H);
 801             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 802             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 803             fvdw12           = _mm_mul_ps(c12_00,FF);
 804             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 805
 806             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 807
 808             fscal            = _mm_add_ps(felec,fvdw);
 809
 810             fscal            = _mm_and_ps(fscal,cutoff_mask);
 811
 812             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 813
 814             /* Calculate temporary vectorial force */
 815             tx               = _mm_mul_ps(fscal,dx00);
 816             ty               = _mm_mul_ps(fscal,dy00);
 817             tz               = _mm_mul_ps(fscal,dz00);
 818
 819             /* Update vectorial force */
 820             fix0             = _mm_add_ps(fix0,tx);
 821             fiy0             = _mm_add_ps(fiy0,ty);
 822             fiz0             = _mm_add_ps(fiz0,tz);
 823
 824             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 825             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 826             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 827             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 828             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 829
 830             }
 831
 832             /* Inner loop uses 58 flops */
 833         }
 834
 835         /* End of innermost loop */
 836
 837         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 838                                               f+i_coord_offset,fshift+i_shift_offset);
 839
 840         /* Increment number of inner iterations */
 841         inneriter                  += j_index_end - j_index_start;
 842
 843         /* Outer loop uses 7 flops */
 844     }
 845
 846     /* Increment number of outer iterations */
 847     outeriter        += nri;
 848
 849     /* Update outer/inner flops */
 850
 851     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*58);
 852 }