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) 2012,2013,2014, 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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 130     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 163
 164         fix0             = _mm_setzero_ps();
 165         fiy0             = _mm_setzero_ps();
 166         fiz0             = _mm_setzero_ps();
 167
 168         /* Load parameters for i particles */
 169         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 170         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_ps();
 174         vvdwsum          = _mm_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               x+j_coord_offsetC,x+j_coord_offsetD,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_ps(ix0,jx0);
 197             dy00             = _mm_sub_ps(iy0,jy0);
 198             dz00             = _mm_sub_ps(iz0,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 202
 203             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 204
 205             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 206
 207             /* Load parameters for j particles */
 208             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 209                                                               charge+jnrC+0,charge+jnrD+0);
 210             vdwjidx0A        = 2*vdwtype[jnrA+0];
 211             vdwjidx0B        = 2*vdwtype[jnrB+0];
 212             vdwjidx0C        = 2*vdwtype[jnrC+0];
 213             vdwjidx0D        = 2*vdwtype[jnrD+0];
 214
 215             /**************************
 216              * CALCULATE INTERACTIONS *
 217              **************************/
 218
 219             if (gmx_mm_any_lt(rsq00,rcutoff2))
 220             {
 221
 222             r00              = _mm_mul_ps(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm_mul_ps(iq0,jq0);
 226             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 227                                          vdwparam+vdwioffset0+vdwjidx0B,
 228                                          vdwparam+vdwioffset0+vdwjidx0C,
 229                                          vdwparam+vdwioffset0+vdwjidx0D,
 230                                          &c6_00,&c12_00);
 231
 232             /* Calculate table index by multiplying r with table scale and truncate to integer */
 233             rt               = _mm_mul_ps(r00,vftabscale);
 234             vfitab           = _mm_cvttps_epi32(rt);
 235             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 236             vfitab           = _mm_slli_epi32(vfitab,3);
 237
 238             /* REACTION-FIELD ELECTROSTATICS */
 239             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 240             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 241
 242             /* CUBIC SPLINE TABLE DISPERSION */
 243             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 244             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 245             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 246             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 247             _MM_TRANSPOSE4_PS(Y,F,G,H);
 248             Heps             = _mm_mul_ps(vfeps,H);
 249             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 250             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 251             vvdw6            = _mm_mul_ps(c6_00,VV);
 252             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 253             fvdw6            = _mm_mul_ps(c6_00,FF);
 254
 255             /* CUBIC SPLINE TABLE REPULSION */
 256             vfitab           = _mm_add_epi32(vfitab,ifour);
 257             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 258             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 259             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 260             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 261             _MM_TRANSPOSE4_PS(Y,F,G,H);
 262             Heps             = _mm_mul_ps(vfeps,H);
 263             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 264             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 265             vvdw12           = _mm_mul_ps(c12_00,VV);
 266             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 267             fvdw12           = _mm_mul_ps(c12_00,FF);
 268             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 269             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 270
 271             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             velec            = _mm_and_ps(velec,cutoff_mask);
 275             velecsum         = _mm_add_ps(velecsum,velec);
 276             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 277             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 278
 279             fscal            = _mm_add_ps(felec,fvdw);
 280
 281             fscal            = _mm_and_ps(fscal,cutoff_mask);
 282
 283             /* Calculate temporary vectorial force */
 284             tx               = _mm_mul_ps(fscal,dx00);
 285             ty               = _mm_mul_ps(fscal,dy00);
 286             tz               = _mm_mul_ps(fscal,dz00);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm_add_ps(fix0,tx);
 290             fiy0             = _mm_add_ps(fiy0,ty);
 291             fiz0             = _mm_add_ps(fiz0,tz);
 292
 293             fjptrA             = f+j_coord_offsetA;
 294             fjptrB             = f+j_coord_offsetB;
 295             fjptrC             = f+j_coord_offsetC;
 296             fjptrD             = f+j_coord_offsetD;
 297             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 298
 299             }
 300
 301             /* Inner loop uses 72 flops */
 302         }
 303
 304         if(jidx<j_index_end)
 305         {
 306
 307             /* Get j neighbor index, and coordinate index */
 308             jnrlistA         = jjnr[jidx];
 309             jnrlistB         = jjnr[jidx+1];
 310             jnrlistC         = jjnr[jidx+2];
 311             jnrlistD         = jjnr[jidx+3];
 312             /* Sign of each element will be negative for non-real atoms.
 313              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 314              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 315              */
 316             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 317             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 318             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 319             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 320             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 321             j_coord_offsetA  = DIM*jnrA;
 322             j_coord_offsetB  = DIM*jnrB;
 323             j_coord_offsetC  = DIM*jnrC;
 324             j_coord_offsetD  = DIM*jnrD;
 325
 326             /* load j atom coordinates */
 327             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 328                                               x+j_coord_offsetC,x+j_coord_offsetD,
 329                                               &jx0,&jy0,&jz0);
 330
 331             /* Calculate displacement vector */
 332             dx00             = _mm_sub_ps(ix0,jx0);
 333             dy00             = _mm_sub_ps(iy0,jy0);
 334             dz00             = _mm_sub_ps(iz0,jz0);
 335
 336             /* Calculate squared distance and things based on it */
 337             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 338
 339             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 340
 341             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 342
 343             /* Load parameters for j particles */
 344             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 345                                                               charge+jnrC+0,charge+jnrD+0);
 346             vdwjidx0A        = 2*vdwtype[jnrA+0];
 347             vdwjidx0B        = 2*vdwtype[jnrB+0];
 348             vdwjidx0C        = 2*vdwtype[jnrC+0];
 349             vdwjidx0D        = 2*vdwtype[jnrD+0];
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             if (gmx_mm_any_lt(rsq00,rcutoff2))
 356             {
 357
 358             r00              = _mm_mul_ps(rsq00,rinv00);
 359             r00              = _mm_andnot_ps(dummy_mask,r00);
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq00             = _mm_mul_ps(iq0,jq0);
 363             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 364                                          vdwparam+vdwioffset0+vdwjidx0B,
 365                                          vdwparam+vdwioffset0+vdwjidx0C,
 366                                          vdwparam+vdwioffset0+vdwjidx0D,
 367                                          &c6_00,&c12_00);
 368
 369             /* Calculate table index by multiplying r with table scale and truncate to integer */
 370             rt               = _mm_mul_ps(r00,vftabscale);
 371             vfitab           = _mm_cvttps_epi32(rt);
 372             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 373             vfitab           = _mm_slli_epi32(vfitab,3);
 374
 375             /* REACTION-FIELD ELECTROSTATICS */
 376             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 377             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 378
 379             /* CUBIC SPLINE TABLE DISPERSION */
 380             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 381             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 382             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 383             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 384             _MM_TRANSPOSE4_PS(Y,F,G,H);
 385             Heps             = _mm_mul_ps(vfeps,H);
 386             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 387             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 388             vvdw6            = _mm_mul_ps(c6_00,VV);
 389             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 390             fvdw6            = _mm_mul_ps(c6_00,FF);
 391
 392             /* CUBIC SPLINE TABLE REPULSION */
 393             vfitab           = _mm_add_epi32(vfitab,ifour);
 394             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 395             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 396             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 397             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 398             _MM_TRANSPOSE4_PS(Y,F,G,H);
 399             Heps             = _mm_mul_ps(vfeps,H);
 400             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 401             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 402             vvdw12           = _mm_mul_ps(c12_00,VV);
 403             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 404             fvdw12           = _mm_mul_ps(c12_00,FF);
 405             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 406             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 407
 408             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velec            = _mm_and_ps(velec,cutoff_mask);
 412             velec            = _mm_andnot_ps(dummy_mask,velec);
 413             velecsum         = _mm_add_ps(velecsum,velec);
 414             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 415             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 416             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 417
 418             fscal            = _mm_add_ps(felec,fvdw);
 419
 420             fscal            = _mm_and_ps(fscal,cutoff_mask);
 421
 422             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 423
 424             /* Calculate temporary vectorial force */
 425             tx               = _mm_mul_ps(fscal,dx00);
 426             ty               = _mm_mul_ps(fscal,dy00);
 427             tz               = _mm_mul_ps(fscal,dz00);
 428
 429             /* Update vectorial force */
 430             fix0             = _mm_add_ps(fix0,tx);
 431             fiy0             = _mm_add_ps(fiy0,ty);
 432             fiz0             = _mm_add_ps(fiz0,tz);
 433
 434             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 435             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 436             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 437             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 438             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 439
 440             }
 441
 442             /* Inner loop uses 73 flops */
 443         }
 444
 445         /* End of innermost loop */
 446
 447         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 448                                               f+i_coord_offset,fshift+i_shift_offset);
 449
 450         ggid                        = gid[iidx];
 451         /* Update potential energies */
 452         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 453         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 454
 455         /* Increment number of inner iterations */
 456         inneriter                  += j_index_end - j_index_start;
 457
 458         /* Outer loop uses 9 flops */
 459     }
 460
 461     /* Increment number of outer iterations */
 462     outeriter        += nri;
 463
 464     /* Update outer/inner flops */
 465
 466     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 467 }
 468 /*
 469  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse2_single
 470  * Electrostatics interaction: ReactionField
 471  * VdW interaction:            CubicSplineTable
 472  * Geometry:                   Particle-Particle
 473  * Calculate force/pot:        Force
 474  */
 475 void
 476 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_sse2_single
 477                     (t_nblist                    * gmx_restrict       nlist,
 478                      rvec                        * gmx_restrict          xx,
 479                      rvec                        * gmx_restrict          ff,
 480                      t_forcerec                  * gmx_restrict          fr,
 481                      t_mdatoms                   * gmx_restrict     mdatoms,
 482                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 483                      t_nrnb                      * gmx_restrict        nrnb)
 484 {
 485     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 486      * just 0 for non-waters.
 487      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 488      * jnr indices corresponding to data put in the four positions in the SIMD register.
 489      */
 490     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 491     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 492     int              jnrA,jnrB,jnrC,jnrD;
 493     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 494     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 495     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 496     real             rcutoff_scalar;
 497     real             *shiftvec,*fshift,*x,*f;
 498     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 499     real             scratch[4*DIM];
 500     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 501     int              vdwioffset0;
 502     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 503     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 504     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 505     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 506     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 507     real             *charge;
 508     int              nvdwtype;
 509     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 510     int              *vdwtype;
 511     real             *vdwparam;
 512     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 513     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 514     __m128i          vfitab;
 515     __m128i          ifour       = _mm_set1_epi32(4);
 516     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 517     real             *vftab;
 518     __m128           dummy_mask,cutoff_mask;
 519     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 520     __m128           one     = _mm_set1_ps(1.0);
 521     __m128           two     = _mm_set1_ps(2.0);
 522     x                = xx[0];
 523     f                = ff[0];
 524
 525     nri              = nlist->nri;
 526     iinr             = nlist->iinr;
 527     jindex           = nlist->jindex;
 528     jjnr             = nlist->jjnr;
 529     shiftidx         = nlist->shift;
 530     gid              = nlist->gid;
 531     shiftvec         = fr->shift_vec[0];
 532     fshift           = fr->fshift[0];
 533     facel            = _mm_set1_ps(fr->epsfac);
 534     charge           = mdatoms->chargeA;
 535     krf              = _mm_set1_ps(fr->ic->k_rf);
 536     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 537     crf              = _mm_set1_ps(fr->ic->c_rf);
 538     nvdwtype         = fr->ntype;
 539     vdwparam         = fr->nbfp;
 540     vdwtype          = mdatoms->typeA;
 541
 542     vftab            = kernel_data->table_vdw->data;
 543     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 544
 545     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 546     rcutoff_scalar   = fr->rcoulomb;
 547     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 548     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 549
 550     /* Avoid stupid compiler warnings */
 551     jnrA = jnrB = jnrC = jnrD = 0;
 552     j_coord_offsetA = 0;
 553     j_coord_offsetB = 0;
 554     j_coord_offsetC = 0;
 555     j_coord_offsetD = 0;
 556
 557     outeriter        = 0;
 558     inneriter        = 0;
 559
 560     for(iidx=0;iidx<4*DIM;iidx++)
 561     {
 562         scratch[iidx] = 0.0;
 563     }
 564
 565     /* Start outer loop over neighborlists */
 566     for(iidx=0; iidx<nri; iidx++)
 567     {
 568         /* Load shift vector for this list */
 569         i_shift_offset   = DIM*shiftidx[iidx];
 570
 571         /* Load limits for loop over neighbors */
 572         j_index_start    = jindex[iidx];
 573         j_index_end      = jindex[iidx+1];
 574
 575         /* Get outer coordinate index */
 576         inr              = iinr[iidx];
 577         i_coord_offset   = DIM*inr;
 578
 579         /* Load i particle coords and add shift vector */
 580         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 581
 582         fix0             = _mm_setzero_ps();
 583         fiy0             = _mm_setzero_ps();
 584         fiz0             = _mm_setzero_ps();
 585
 586         /* Load parameters for i particles */
 587         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 588         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 589
 590         /* Start inner kernel loop */
 591         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 592         {
 593
 594             /* Get j neighbor index, and coordinate index */
 595             jnrA             = jjnr[jidx];
 596             jnrB             = jjnr[jidx+1];
 597             jnrC             = jjnr[jidx+2];
 598             jnrD             = jjnr[jidx+3];
 599             j_coord_offsetA  = DIM*jnrA;
 600             j_coord_offsetB  = DIM*jnrB;
 601             j_coord_offsetC  = DIM*jnrC;
 602             j_coord_offsetD  = DIM*jnrD;
 603
 604             /* load j atom coordinates */
 605             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 606                                               x+j_coord_offsetC,x+j_coord_offsetD,
 607                                               &jx0,&jy0,&jz0);
 608
 609             /* Calculate displacement vector */
 610             dx00             = _mm_sub_ps(ix0,jx0);
 611             dy00             = _mm_sub_ps(iy0,jy0);
 612             dz00             = _mm_sub_ps(iz0,jz0);
 613
 614             /* Calculate squared distance and things based on it */
 615             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 616
 617             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 618
 619             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 620
 621             /* Load parameters for j particles */
 622             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 623                                                               charge+jnrC+0,charge+jnrD+0);
 624             vdwjidx0A        = 2*vdwtype[jnrA+0];
 625             vdwjidx0B        = 2*vdwtype[jnrB+0];
 626             vdwjidx0C        = 2*vdwtype[jnrC+0];
 627             vdwjidx0D        = 2*vdwtype[jnrD+0];
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             if (gmx_mm_any_lt(rsq00,rcutoff2))
 634             {
 635
 636             r00              = _mm_mul_ps(rsq00,rinv00);
 637
 638             /* Compute parameters for interactions between i and j atoms */
 639             qq00             = _mm_mul_ps(iq0,jq0);
 640             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 641                                          vdwparam+vdwioffset0+vdwjidx0B,
 642                                          vdwparam+vdwioffset0+vdwjidx0C,
 643                                          vdwparam+vdwioffset0+vdwjidx0D,
 644                                          &c6_00,&c12_00);
 645
 646             /* Calculate table index by multiplying r with table scale and truncate to integer */
 647             rt               = _mm_mul_ps(r00,vftabscale);
 648             vfitab           = _mm_cvttps_epi32(rt);
 649             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 650             vfitab           = _mm_slli_epi32(vfitab,3);
 651
 652             /* REACTION-FIELD ELECTROSTATICS */
 653             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 654
 655             /* CUBIC SPLINE TABLE DISPERSION */
 656             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 657             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 658             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 659             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 660             _MM_TRANSPOSE4_PS(Y,F,G,H);
 661             Heps             = _mm_mul_ps(vfeps,H);
 662             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 663             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 664             fvdw6            = _mm_mul_ps(c6_00,FF);
 665
 666             /* CUBIC SPLINE TABLE REPULSION */
 667             vfitab           = _mm_add_epi32(vfitab,ifour);
 668             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 669             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 670             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 671             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 672             _MM_TRANSPOSE4_PS(Y,F,G,H);
 673             Heps             = _mm_mul_ps(vfeps,H);
 674             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 675             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 676             fvdw12           = _mm_mul_ps(c12_00,FF);
 677             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 678
 679             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 680
 681             fscal            = _mm_add_ps(felec,fvdw);
 682
 683             fscal            = _mm_and_ps(fscal,cutoff_mask);
 684
 685             /* Calculate temporary vectorial force */
 686             tx               = _mm_mul_ps(fscal,dx00);
 687             ty               = _mm_mul_ps(fscal,dy00);
 688             tz               = _mm_mul_ps(fscal,dz00);
 689
 690             /* Update vectorial force */
 691             fix0             = _mm_add_ps(fix0,tx);
 692             fiy0             = _mm_add_ps(fiy0,ty);
 693             fiz0             = _mm_add_ps(fiz0,tz);
 694
 695             fjptrA             = f+j_coord_offsetA;
 696             fjptrB             = f+j_coord_offsetB;
 697             fjptrC             = f+j_coord_offsetC;
 698             fjptrD             = f+j_coord_offsetD;
 699             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 700
 701             }
 702
 703             /* Inner loop uses 57 flops */
 704         }
 705
 706         if(jidx<j_index_end)
 707         {
 708
 709             /* Get j neighbor index, and coordinate index */
 710             jnrlistA         = jjnr[jidx];
 711             jnrlistB         = jjnr[jidx+1];
 712             jnrlistC         = jjnr[jidx+2];
 713             jnrlistD         = jjnr[jidx+3];
 714             /* Sign of each element will be negative for non-real atoms.
 715              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 716              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 717              */
 718             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 719             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 720             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 721             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 722             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 723             j_coord_offsetA  = DIM*jnrA;
 724             j_coord_offsetB  = DIM*jnrB;
 725             j_coord_offsetC  = DIM*jnrC;
 726             j_coord_offsetD  = DIM*jnrD;
 727
 728             /* load j atom coordinates */
 729             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 730                                               x+j_coord_offsetC,x+j_coord_offsetD,
 731                                               &jx0,&jy0,&jz0);
 732
 733             /* Calculate displacement vector */
 734             dx00             = _mm_sub_ps(ix0,jx0);
 735             dy00             = _mm_sub_ps(iy0,jy0);
 736             dz00             = _mm_sub_ps(iz0,jz0);
 737
 738             /* Calculate squared distance and things based on it */
 739             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 740
 741             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 742
 743             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 744
 745             /* Load parameters for j particles */
 746             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 747                                                               charge+jnrC+0,charge+jnrD+0);
 748             vdwjidx0A        = 2*vdwtype[jnrA+0];
 749             vdwjidx0B        = 2*vdwtype[jnrB+0];
 750             vdwjidx0C        = 2*vdwtype[jnrC+0];
 751             vdwjidx0D        = 2*vdwtype[jnrD+0];
 752
 753             /**************************
 754              * CALCULATE INTERACTIONS *
 755              **************************/
 756
 757             if (gmx_mm_any_lt(rsq00,rcutoff2))
 758             {
 759
 760             r00              = _mm_mul_ps(rsq00,rinv00);
 761             r00              = _mm_andnot_ps(dummy_mask,r00);
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq00             = _mm_mul_ps(iq0,jq0);
 765             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 766                                          vdwparam+vdwioffset0+vdwjidx0B,
 767                                          vdwparam+vdwioffset0+vdwjidx0C,
 768                                          vdwparam+vdwioffset0+vdwjidx0D,
 769                                          &c6_00,&c12_00);
 770
 771             /* Calculate table index by multiplying r with table scale and truncate to integer */
 772             rt               = _mm_mul_ps(r00,vftabscale);
 773             vfitab           = _mm_cvttps_epi32(rt);
 774             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 775             vfitab           = _mm_slli_epi32(vfitab,3);
 776
 777             /* REACTION-FIELD ELECTROSTATICS */
 778             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 779
 780             /* CUBIC SPLINE TABLE DISPERSION */
 781             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 782             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 783             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 784             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 785             _MM_TRANSPOSE4_PS(Y,F,G,H);
 786             Heps             = _mm_mul_ps(vfeps,H);
 787             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 788             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 789             fvdw6            = _mm_mul_ps(c6_00,FF);
 790
 791             /* CUBIC SPLINE TABLE REPULSION */
 792             vfitab           = _mm_add_epi32(vfitab,ifour);
 793             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 794             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 795             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 796             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 797             _MM_TRANSPOSE4_PS(Y,F,G,H);
 798             Heps             = _mm_mul_ps(vfeps,H);
 799             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 800             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 801             fvdw12           = _mm_mul_ps(c12_00,FF);
 802             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 803
 804             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 805
 806             fscal            = _mm_add_ps(felec,fvdw);
 807
 808             fscal            = _mm_and_ps(fscal,cutoff_mask);
 809
 810             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 811
 812             /* Calculate temporary vectorial force */
 813             tx               = _mm_mul_ps(fscal,dx00);
 814             ty               = _mm_mul_ps(fscal,dy00);
 815             tz               = _mm_mul_ps(fscal,dz00);
 816
 817             /* Update vectorial force */
 818             fix0             = _mm_add_ps(fix0,tx);
 819             fiy0             = _mm_add_ps(fiy0,ty);
 820             fiz0             = _mm_add_ps(fiz0,tz);
 821
 822             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 823             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 824             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 825             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 826             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 827
 828             }
 829
 830             /* Inner loop uses 58 flops */
 831         }
 832
 833         /* End of innermost loop */
 834
 835         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 836                                               f+i_coord_offset,fshift+i_shift_offset);
 837
 838         /* Increment number of inner iterations */
 839         inneriter                  += j_index_end - j_index_start;
 840
 841         /* Outer loop uses 7 flops */
 842     }
 843
 844     /* Increment number of outer iterations */
 845     outeriter        += nri;
 846
 847     /* Update outer/inner flops */
 848
 849     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*58);
 850 }