src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  69     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  78     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  79     __m128i          vfitab;
  80     __m128i          ifour       = _mm_set1_epi32(4);
  81     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  82     real             *vftab;
  83     __m128           dummy_mask,cutoff_mask;
  84     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  85     __m128           one     = _mm_set1_ps(1.0);
  86     __m128           two     = _mm_set1_ps(2.0);
  87     x                = xx[0];
  88     f                = ff[0];
  89
  90     nri              = nlist->nri;
  91     iinr             = nlist->iinr;
  92     jindex           = nlist->jindex;
  93     jjnr             = nlist->jjnr;
  94     shiftidx         = nlist->shift;
  95     gid              = nlist->gid;
  96     shiftvec         = fr->shift_vec[0];
  97     fshift           = fr->fshift[0];
  98     facel            = _mm_set1_ps(fr->epsfac);
  99     charge           = mdatoms->chargeA;
 100     krf              = _mm_set1_ps(fr->ic->k_rf);
 101     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 102     crf              = _mm_set1_ps(fr->ic->c_rf);
 103     nvdwtype         = fr->ntype;
 104     vdwparam         = fr->nbfp;
 105     vdwtype          = mdatoms->typeA;
 106
 107     vftab            = kernel_data->table_vdw->data;
 108     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 109
 110     /* Avoid stupid compiler warnings */
 111     jnrA = jnrB = jnrC = jnrD = 0;
 112     j_coord_offsetA = 0;
 113     j_coord_offsetB = 0;
 114     j_coord_offsetC = 0;
 115     j_coord_offsetD = 0;
 116
 117     outeriter        = 0;
 118     inneriter        = 0;
 119
 120     /* Start outer loop over neighborlists */
 121     for(iidx=0; iidx<nri; iidx++)
 122     {
 123         /* Load shift vector for this list */
 124         i_shift_offset   = DIM*shiftidx[iidx];
 125         shX              = shiftvec[i_shift_offset+XX];
 126         shY              = shiftvec[i_shift_offset+YY];
 127         shZ              = shiftvec[i_shift_offset+ZZ];
 128
 129         /* Load limits for loop over neighbors */
 130         j_index_start    = jindex[iidx];
 131         j_index_end      = jindex[iidx+1];
 132
 133         /* Get outer coordinate index */
 134         inr              = iinr[iidx];
 135         i_coord_offset   = DIM*inr;
 136
 137         /* Load i particle coords and add shift vector */
 138         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 139         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 140         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 141
 142         fix0             = _mm_setzero_ps();
 143         fiy0             = _mm_setzero_ps();
 144         fiz0             = _mm_setzero_ps();
 145
 146         /* Load parameters for i particles */
 147         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 148         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm_setzero_ps();
 152         vvdwsum          = _mm_setzero_ps();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             jnrC             = jjnr[jidx+2];
 162             jnrD             = jjnr[jidx+3];
 163
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               x+j_coord_offsetC,x+j_coord_offsetD,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_ps(ix0,jx0);
 176             dy00             = _mm_sub_ps(iy0,jy0);
 177             dz00             = _mm_sub_ps(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 183
 184             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 188                                                               charge+jnrC+0,charge+jnrD+0);
 189             vdwjidx0A        = 2*vdwtype[jnrA+0];
 190             vdwjidx0B        = 2*vdwtype[jnrB+0];
 191             vdwjidx0C        = 2*vdwtype[jnrC+0];
 192             vdwjidx0D        = 2*vdwtype[jnrD+0];
 193
 194             /**************************
 195              * CALCULATE INTERACTIONS *
 196              **************************/
 197
 198             r00              = _mm_mul_ps(rsq00,rinv00);
 199
 200             /* Compute parameters for interactions between i and j atoms */
 201             qq00             = _mm_mul_ps(iq0,jq0);
 202             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 203                                          vdwparam+vdwioffset0+vdwjidx0B,
 204                                          vdwparam+vdwioffset0+vdwjidx0C,
 205                                          vdwparam+vdwioffset0+vdwjidx0D,
 206                                          &c6_00,&c12_00);
 207
 208             /* Calculate table index by multiplying r with table scale and truncate to integer */
 209             rt               = _mm_mul_ps(r00,vftabscale);
 210             vfitab           = _mm_cvttps_epi32(rt);
 211             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 212             vfitab           = _mm_slli_epi32(vfitab,3);
 213
 214             /* REACTION-FIELD ELECTROSTATICS */
 215             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 216             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 217
 218             /* CUBIC SPLINE TABLE DISPERSION */
 219             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 220             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 221             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 222             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 223             _MM_TRANSPOSE4_PS(Y,F,G,H);
 224             Heps             = _mm_mul_ps(vfeps,H);
 225             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 226             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 227             vvdw6            = _mm_mul_ps(c6_00,VV);
 228             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 229             fvdw6            = _mm_mul_ps(c6_00,FF);
 230
 231             /* CUBIC SPLINE TABLE REPULSION */
 232             vfitab           = _mm_add_epi32(vfitab,ifour);
 233             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 234             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 235             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 236             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 237             _MM_TRANSPOSE4_PS(Y,F,G,H);
 238             Heps             = _mm_mul_ps(vfeps,H);
 239             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 240             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 241             vvdw12           = _mm_mul_ps(c12_00,VV);
 242             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 243             fvdw12           = _mm_mul_ps(c12_00,FF);
 244             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 245             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velecsum         = _mm_add_ps(velecsum,velec);
 249             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 250
 251             fscal            = _mm_add_ps(felec,fvdw);
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_ps(fscal,dx00);
 255             ty               = _mm_mul_ps(fscal,dy00);
 256             tz               = _mm_mul_ps(fscal,dz00);
 257
 258             /* Update vectorial force */
 259             fix0             = _mm_add_ps(fix0,tx);
 260             fiy0             = _mm_add_ps(fiy0,ty);
 261             fiz0             = _mm_add_ps(fiz0,tz);
 262
 263             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 264                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 265                                                    tx,ty,tz);
 266
 267             /* Inner loop uses 67 flops */
 268         }
 269
 270         if(jidx<j_index_end)
 271         {
 272
 273             /* Get j neighbor index, and coordinate index */
 274             jnrA             = jjnr[jidx];
 275             jnrB             = jjnr[jidx+1];
 276             jnrC             = jjnr[jidx+2];
 277             jnrD             = jjnr[jidx+3];
 278
 279             /* Sign of each element will be negative for non-real atoms.
 280              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 281              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 282              */
 283             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 284             jnrA       = (jnrA>=0) ? jnrA : 0;
 285             jnrB       = (jnrB>=0) ? jnrB : 0;
 286             jnrC       = (jnrC>=0) ? jnrC : 0;
 287             jnrD       = (jnrD>=0) ? jnrD : 0;
 288
 289             j_coord_offsetA  = DIM*jnrA;
 290             j_coord_offsetB  = DIM*jnrB;
 291             j_coord_offsetC  = DIM*jnrC;
 292             j_coord_offsetD  = DIM*jnrD;
 293
 294             /* load j atom coordinates */
 295             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 296                                               x+j_coord_offsetC,x+j_coord_offsetD,
 297                                               &jx0,&jy0,&jz0);
 298
 299             /* Calculate displacement vector */
 300             dx00             = _mm_sub_ps(ix0,jx0);
 301             dy00             = _mm_sub_ps(iy0,jy0);
 302             dz00             = _mm_sub_ps(iz0,jz0);
 303
 304             /* Calculate squared distance and things based on it */
 305             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 306
 307             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 308
 309             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 310
 311             /* Load parameters for j particles */
 312             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 313                                                               charge+jnrC+0,charge+jnrD+0);
 314             vdwjidx0A        = 2*vdwtype[jnrA+0];
 315             vdwjidx0B        = 2*vdwtype[jnrB+0];
 316             vdwjidx0C        = 2*vdwtype[jnrC+0];
 317             vdwjidx0D        = 2*vdwtype[jnrD+0];
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             r00              = _mm_mul_ps(rsq00,rinv00);
 324             r00              = _mm_andnot_ps(dummy_mask,r00);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq00             = _mm_mul_ps(iq0,jq0);
 328             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 329                                          vdwparam+vdwioffset0+vdwjidx0B,
 330                                          vdwparam+vdwioffset0+vdwjidx0C,
 331                                          vdwparam+vdwioffset0+vdwjidx0D,
 332                                          &c6_00,&c12_00);
 333
 334             /* Calculate table index by multiplying r with table scale and truncate to integer */
 335             rt               = _mm_mul_ps(r00,vftabscale);
 336             vfitab           = _mm_cvttps_epi32(rt);
 337             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 338             vfitab           = _mm_slli_epi32(vfitab,3);
 339
 340             /* REACTION-FIELD ELECTROSTATICS */
 341             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 342             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 343
 344             /* CUBIC SPLINE TABLE DISPERSION */
 345             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 346             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 347             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 348             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 349             _MM_TRANSPOSE4_PS(Y,F,G,H);
 350             Heps             = _mm_mul_ps(vfeps,H);
 351             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 352             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 353             vvdw6            = _mm_mul_ps(c6_00,VV);
 354             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 355             fvdw6            = _mm_mul_ps(c6_00,FF);
 356
 357             /* CUBIC SPLINE TABLE REPULSION */
 358             vfitab           = _mm_add_epi32(vfitab,ifour);
 359             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 360             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 361             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 362             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 363             _MM_TRANSPOSE4_PS(Y,F,G,H);
 364             Heps             = _mm_mul_ps(vfeps,H);
 365             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 366             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 367             vvdw12           = _mm_mul_ps(c12_00,VV);
 368             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 369             fvdw12           = _mm_mul_ps(c12_00,FF);
 370             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 371             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _mm_andnot_ps(dummy_mask,velec);
 375             velecsum         = _mm_add_ps(velecsum,velec);
 376             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 377             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 378
 379             fscal            = _mm_add_ps(felec,fvdw);
 380
 381             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 382
 383             /* Calculate temporary vectorial force */
 384             tx               = _mm_mul_ps(fscal,dx00);
 385             ty               = _mm_mul_ps(fscal,dy00);
 386             tz               = _mm_mul_ps(fscal,dz00);
 387
 388             /* Update vectorial force */
 389             fix0             = _mm_add_ps(fix0,tx);
 390             fiy0             = _mm_add_ps(fiy0,ty);
 391             fiz0             = _mm_add_ps(fiz0,tz);
 392
 393             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 394                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 395                                                    tx,ty,tz);
 396
 397             /* Inner loop uses 68 flops */
 398         }
 399
 400         /* End of innermost loop */
 401
 402         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 403                                               f+i_coord_offset,fshift+i_shift_offset);
 404
 405         ggid                        = gid[iidx];
 406         /* Update potential energies */
 407         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 408         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 409
 410         /* Increment number of inner iterations */
 411         inneriter                  += j_index_end - j_index_start;
 412
 413         /* Outer loop uses 12 flops */
 414     }
 415
 416     /* Increment number of outer iterations */
 417     outeriter        += nri;
 418
 419     /* Update outer/inner flops */
 420
 421     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*68);
 422 }
 423 /*
 424  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_single
 425  * Electrostatics interaction: ReactionField
 426  * VdW interaction:            CubicSplineTable
 427  * Geometry:                   Particle-Particle
 428  * Calculate force/pot:        Force
 429  */
 430 void
 431 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse2_single
 432                     (t_nblist * gmx_restrict                nlist,
 433                      rvec * gmx_restrict                    xx,
 434                      rvec * gmx_restrict                    ff,
 435                      t_forcerec * gmx_restrict              fr,
 436                      t_mdatoms * gmx_restrict               mdatoms,
 437                      nb_kernel_data_t * gmx_restrict        kernel_data,
 438                      t_nrnb * gmx_restrict                  nrnb)
 439 {
 440     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 441      * just 0 for non-waters.
 442      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 443      * jnr indices corresponding to data put in the four positions in the SIMD register.
 444      */
 445     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 446     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 447     int              jnrA,jnrB,jnrC,jnrD;
 448     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 449     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 450     real             shX,shY,shZ,rcutoff_scalar;
 451     real             *shiftvec,*fshift,*x,*f;
 452     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 453     int              vdwioffset0;
 454     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 455     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 456     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 457     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 458     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 459     real             *charge;
 460     int              nvdwtype;
 461     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 462     int              *vdwtype;
 463     real             *vdwparam;
 464     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 465     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 466     __m128i          vfitab;
 467     __m128i          ifour       = _mm_set1_epi32(4);
 468     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 469     real             *vftab;
 470     __m128           dummy_mask,cutoff_mask;
 471     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 472     __m128           one     = _mm_set1_ps(1.0);
 473     __m128           two     = _mm_set1_ps(2.0);
 474     x                = xx[0];
 475     f                = ff[0];
 476
 477     nri              = nlist->nri;
 478     iinr             = nlist->iinr;
 479     jindex           = nlist->jindex;
 480     jjnr             = nlist->jjnr;
 481     shiftidx         = nlist->shift;
 482     gid              = nlist->gid;
 483     shiftvec         = fr->shift_vec[0];
 484     fshift           = fr->fshift[0];
 485     facel            = _mm_set1_ps(fr->epsfac);
 486     charge           = mdatoms->chargeA;
 487     krf              = _mm_set1_ps(fr->ic->k_rf);
 488     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 489     crf              = _mm_set1_ps(fr->ic->c_rf);
 490     nvdwtype         = fr->ntype;
 491     vdwparam         = fr->nbfp;
 492     vdwtype          = mdatoms->typeA;
 493
 494     vftab            = kernel_data->table_vdw->data;
 495     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 496
 497     /* Avoid stupid compiler warnings */
 498     jnrA = jnrB = jnrC = jnrD = 0;
 499     j_coord_offsetA = 0;
 500     j_coord_offsetB = 0;
 501     j_coord_offsetC = 0;
 502     j_coord_offsetD = 0;
 503
 504     outeriter        = 0;
 505     inneriter        = 0;
 506
 507     /* Start outer loop over neighborlists */
 508     for(iidx=0; iidx<nri; iidx++)
 509     {
 510         /* Load shift vector for this list */
 511         i_shift_offset   = DIM*shiftidx[iidx];
 512         shX              = shiftvec[i_shift_offset+XX];
 513         shY              = shiftvec[i_shift_offset+YY];
 514         shZ              = shiftvec[i_shift_offset+ZZ];
 515
 516         /* Load limits for loop over neighbors */
 517         j_index_start    = jindex[iidx];
 518         j_index_end      = jindex[iidx+1];
 519
 520         /* Get outer coordinate index */
 521         inr              = iinr[iidx];
 522         i_coord_offset   = DIM*inr;
 523
 524         /* Load i particle coords and add shift vector */
 525         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 526         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 527         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 528
 529         fix0             = _mm_setzero_ps();
 530         fiy0             = _mm_setzero_ps();
 531         fiz0             = _mm_setzero_ps();
 532
 533         /* Load parameters for i particles */
 534         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 535         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 536
 537         /* Start inner kernel loop */
 538         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 539         {
 540
 541             /* Get j neighbor index, and coordinate index */
 542             jnrA             = jjnr[jidx];
 543             jnrB             = jjnr[jidx+1];
 544             jnrC             = jjnr[jidx+2];
 545             jnrD             = jjnr[jidx+3];
 546
 547             j_coord_offsetA  = DIM*jnrA;
 548             j_coord_offsetB  = DIM*jnrB;
 549             j_coord_offsetC  = DIM*jnrC;
 550             j_coord_offsetD  = DIM*jnrD;
 551
 552             /* load j atom coordinates */
 553             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 554                                               x+j_coord_offsetC,x+j_coord_offsetD,
 555                                               &jx0,&jy0,&jz0);
 556
 557             /* Calculate displacement vector */
 558             dx00             = _mm_sub_ps(ix0,jx0);
 559             dy00             = _mm_sub_ps(iy0,jy0);
 560             dz00             = _mm_sub_ps(iz0,jz0);
 561
 562             /* Calculate squared distance and things based on it */
 563             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 564
 565             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 566
 567             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 568
 569             /* Load parameters for j particles */
 570             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 571                                                               charge+jnrC+0,charge+jnrD+0);
 572             vdwjidx0A        = 2*vdwtype[jnrA+0];
 573             vdwjidx0B        = 2*vdwtype[jnrB+0];
 574             vdwjidx0C        = 2*vdwtype[jnrC+0];
 575             vdwjidx0D        = 2*vdwtype[jnrD+0];
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             r00              = _mm_mul_ps(rsq00,rinv00);
 582
 583             /* Compute parameters for interactions between i and j atoms */
 584             qq00             = _mm_mul_ps(iq0,jq0);
 585             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 586                                          vdwparam+vdwioffset0+vdwjidx0B,
 587                                          vdwparam+vdwioffset0+vdwjidx0C,
 588                                          vdwparam+vdwioffset0+vdwjidx0D,
 589                                          &c6_00,&c12_00);
 590
 591             /* Calculate table index by multiplying r with table scale and truncate to integer */
 592             rt               = _mm_mul_ps(r00,vftabscale);
 593             vfitab           = _mm_cvttps_epi32(rt);
 594             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 595             vfitab           = _mm_slli_epi32(vfitab,3);
 596
 597             /* REACTION-FIELD ELECTROSTATICS */
 598             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 599
 600             /* CUBIC SPLINE TABLE DISPERSION */
 601             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 602             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 603             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 604             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 605             _MM_TRANSPOSE4_PS(Y,F,G,H);
 606             Heps             = _mm_mul_ps(vfeps,H);
 607             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 608             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 609             fvdw6            = _mm_mul_ps(c6_00,FF);
 610
 611             /* CUBIC SPLINE TABLE REPULSION */
 612             vfitab           = _mm_add_epi32(vfitab,ifour);
 613             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 614             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 615             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 616             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 617             _MM_TRANSPOSE4_PS(Y,F,G,H);
 618             Heps             = _mm_mul_ps(vfeps,H);
 619             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 620             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 621             fvdw12           = _mm_mul_ps(c12_00,FF);
 622             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 623
 624             fscal            = _mm_add_ps(felec,fvdw);
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm_mul_ps(fscal,dx00);
 628             ty               = _mm_mul_ps(fscal,dy00);
 629             tz               = _mm_mul_ps(fscal,dz00);
 630
 631             /* Update vectorial force */
 632             fix0             = _mm_add_ps(fix0,tx);
 633             fiy0             = _mm_add_ps(fiy0,ty);
 634             fiz0             = _mm_add_ps(fiz0,tz);
 635
 636             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 637                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 638                                                    tx,ty,tz);
 639
 640             /* Inner loop uses 54 flops */
 641         }
 642
 643         if(jidx<j_index_end)
 644         {
 645
 646             /* Get j neighbor index, and coordinate index */
 647             jnrA             = jjnr[jidx];
 648             jnrB             = jjnr[jidx+1];
 649             jnrC             = jjnr[jidx+2];
 650             jnrD             = jjnr[jidx+3];
 651
 652             /* Sign of each element will be negative for non-real atoms.
 653              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 654              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 655              */
 656             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 657             jnrA       = (jnrA>=0) ? jnrA : 0;
 658             jnrB       = (jnrB>=0) ? jnrB : 0;
 659             jnrC       = (jnrC>=0) ? jnrC : 0;
 660             jnrD       = (jnrD>=0) ? jnrD : 0;
 661
 662             j_coord_offsetA  = DIM*jnrA;
 663             j_coord_offsetB  = DIM*jnrB;
 664             j_coord_offsetC  = DIM*jnrC;
 665             j_coord_offsetD  = DIM*jnrD;
 666
 667             /* load j atom coordinates */
 668             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 669                                               x+j_coord_offsetC,x+j_coord_offsetD,
 670                                               &jx0,&jy0,&jz0);
 671
 672             /* Calculate displacement vector */
 673             dx00             = _mm_sub_ps(ix0,jx0);
 674             dy00             = _mm_sub_ps(iy0,jy0);
 675             dz00             = _mm_sub_ps(iz0,jz0);
 676
 677             /* Calculate squared distance and things based on it */
 678             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 679
 680             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 681
 682             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 683
 684             /* Load parameters for j particles */
 685             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 686                                                               charge+jnrC+0,charge+jnrD+0);
 687             vdwjidx0A        = 2*vdwtype[jnrA+0];
 688             vdwjidx0B        = 2*vdwtype[jnrB+0];
 689             vdwjidx0C        = 2*vdwtype[jnrC+0];
 690             vdwjidx0D        = 2*vdwtype[jnrD+0];
 691
 692             /**************************
 693              * CALCULATE INTERACTIONS *
 694              **************************/
 695
 696             r00              = _mm_mul_ps(rsq00,rinv00);
 697             r00              = _mm_andnot_ps(dummy_mask,r00);
 698
 699             /* Compute parameters for interactions between i and j atoms */
 700             qq00             = _mm_mul_ps(iq0,jq0);
 701             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 702                                          vdwparam+vdwioffset0+vdwjidx0B,
 703                                          vdwparam+vdwioffset0+vdwjidx0C,
 704                                          vdwparam+vdwioffset0+vdwjidx0D,
 705                                          &c6_00,&c12_00);
 706
 707             /* Calculate table index by multiplying r with table scale and truncate to integer */
 708             rt               = _mm_mul_ps(r00,vftabscale);
 709             vfitab           = _mm_cvttps_epi32(rt);
 710             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 711             vfitab           = _mm_slli_epi32(vfitab,3);
 712
 713             /* REACTION-FIELD ELECTROSTATICS */
 714             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 715
 716             /* CUBIC SPLINE TABLE DISPERSION */
 717             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 718             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 719             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 720             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 721             _MM_TRANSPOSE4_PS(Y,F,G,H);
 722             Heps             = _mm_mul_ps(vfeps,H);
 723             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 724             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 725             fvdw6            = _mm_mul_ps(c6_00,FF);
 726
 727             /* CUBIC SPLINE TABLE REPULSION */
 728             vfitab           = _mm_add_epi32(vfitab,ifour);
 729             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 730             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 731             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 732             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 733             _MM_TRANSPOSE4_PS(Y,F,G,H);
 734             Heps             = _mm_mul_ps(vfeps,H);
 735             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 736             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 737             fvdw12           = _mm_mul_ps(c12_00,FF);
 738             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 739
 740             fscal            = _mm_add_ps(felec,fvdw);
 741
 742             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 743
 744             /* Calculate temporary vectorial force */
 745             tx               = _mm_mul_ps(fscal,dx00);
 746             ty               = _mm_mul_ps(fscal,dy00);
 747             tz               = _mm_mul_ps(fscal,dz00);
 748
 749             /* Update vectorial force */
 750             fix0             = _mm_add_ps(fix0,tx);
 751             fiy0             = _mm_add_ps(fiy0,ty);
 752             fiz0             = _mm_add_ps(fiz0,tz);
 753
 754             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 755                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 756                                                    tx,ty,tz);
 757
 758             /* Inner loop uses 55 flops */
 759         }
 760
 761         /* End of innermost loop */
 762
 763         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 764                                               f+i_coord_offset,fshift+i_shift_offset);
 765
 766         /* Increment number of inner iterations */
 767         inneriter                  += j_index_end - j_index_start;
 768
 769         /* Outer loop uses 10 flops */
 770     }
 771
 772     /* Increment number of outer iterations */
 773     outeriter        += nri;
 774
 775     /* Update outer/inner flops */
 776
 777     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*55);
 778 }