src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSw_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_ElecRFCut_VdwLJSw_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_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     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  80     real             rswitch_scalar,d_scalar;
  81     __m128           dummy_mask,cutoff_mask;
  82     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  83     __m128           one     = _mm_set1_ps(1.0);
  84     __m128           two     = _mm_set1_ps(2.0);
  85     x                = xx[0];
  86     f                = ff[0];
  87
  88     nri              = nlist->nri;
  89     iinr             = nlist->iinr;
  90     jindex           = nlist->jindex;
  91     jjnr             = nlist->jjnr;
  92     shiftidx         = nlist->shift;
  93     gid              = nlist->gid;
  94     shiftvec         = fr->shift_vec[0];
  95     fshift           = fr->fshift[0];
  96     facel            = _mm_set1_ps(fr->epsfac);
  97     charge           = mdatoms->chargeA;
  98     krf              = _mm_set1_ps(fr->ic->k_rf);
  99     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 100     crf              = _mm_set1_ps(fr->ic->c_rf);
 101     nvdwtype         = fr->ntype;
 102     vdwparam         = fr->nbfp;
 103     vdwtype          = mdatoms->typeA;
 104
 105     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 106     rcutoff_scalar   = fr->rcoulomb;
 107     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 108     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 109
 110     rswitch_scalar   = fr->rvdw_switch;
 111     rswitch          = _mm_set1_ps(rswitch_scalar);
 112     /* Setup switch parameters */
 113     d_scalar         = rcutoff_scalar-rswitch_scalar;
 114     d                = _mm_set1_ps(d_scalar);
 115     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 116     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 117     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 118     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 119     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 120     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137         shX              = shiftvec[i_shift_offset+XX];
 138         shY              = shiftvec[i_shift_offset+YY];
 139         shZ              = shiftvec[i_shift_offset+ZZ];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 151         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 152         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 153
 154         fix0             = _mm_setzero_ps();
 155         fiy0             = _mm_setzero_ps();
 156         fiz0             = _mm_setzero_ps();
 157
 158         /* Load parameters for i particles */
 159         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 160         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_ps();
 164         vvdwsum          = _mm_setzero_ps();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             jnrC             = jjnr[jidx+2];
 174             jnrD             = jjnr[jidx+3];
 175
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178             j_coord_offsetC  = DIM*jnrC;
 179             j_coord_offsetD  = DIM*jnrD;
 180
 181             /* load j atom coordinates */
 182             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                               x+j_coord_offsetC,x+j_coord_offsetD,
 184                                               &jx0,&jy0,&jz0);
 185
 186             /* Calculate displacement vector */
 187             dx00             = _mm_sub_ps(ix0,jx0);
 188             dy00             = _mm_sub_ps(iy0,jy0);
 189             dz00             = _mm_sub_ps(iz0,jz0);
 190
 191             /* Calculate squared distance and things based on it */
 192             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 193
 194             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 195
 196             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                               charge+jnrC+0,charge+jnrD+0);
 201             vdwjidx0A        = 2*vdwtype[jnrA+0];
 202             vdwjidx0B        = 2*vdwtype[jnrB+0];
 203             vdwjidx0C        = 2*vdwtype[jnrC+0];
 204             vdwjidx0D        = 2*vdwtype[jnrD+0];
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             if (gmx_mm_any_lt(rsq00,rcutoff2))
 211             {
 212
 213             r00              = _mm_mul_ps(rsq00,rinv00);
 214
 215             /* Compute parameters for interactions between i and j atoms */
 216             qq00             = _mm_mul_ps(iq0,jq0);
 217             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 218                                          vdwparam+vdwioffset0+vdwjidx0B,
 219                                          vdwparam+vdwioffset0+vdwjidx0C,
 220                                          vdwparam+vdwioffset0+vdwjidx0D,
 221                                          &c6_00,&c12_00);
 222
 223             /* REACTION-FIELD ELECTROSTATICS */
 224             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 225             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 226
 227             /* LENNARD-JONES DISPERSION/REPULSION */
 228
 229             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 230             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 231             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 232             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 233             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 234
 235             d                = _mm_sub_ps(r00,rswitch);
 236             d                = _mm_max_ps(d,_mm_setzero_ps());
 237             d2               = _mm_mul_ps(d,d);
 238             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 239
 240             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 241
 242             /* Evaluate switch function */
 243             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 244             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 245             vvdw             = _mm_mul_ps(vvdw,sw);
 246             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             velec            = _mm_and_ps(velec,cutoff_mask);
 250             velecsum         = _mm_add_ps(velecsum,velec);
 251             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 252             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 253
 254             fscal            = _mm_add_ps(felec,fvdw);
 255
 256             fscal            = _mm_and_ps(fscal,cutoff_mask);
 257
 258             /* Calculate temporary vectorial force */
 259             tx               = _mm_mul_ps(fscal,dx00);
 260             ty               = _mm_mul_ps(fscal,dy00);
 261             tz               = _mm_mul_ps(fscal,dz00);
 262
 263             /* Update vectorial force */
 264             fix0             = _mm_add_ps(fix0,tx);
 265             fiy0             = _mm_add_ps(fiy0,ty);
 266             fiz0             = _mm_add_ps(fiz0,tz);
 267
 268             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 269                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 270                                                    tx,ty,tz);
 271
 272             }
 273
 274             /* Inner loop uses 70 flops */
 275         }
 276
 277         if(jidx<j_index_end)
 278         {
 279
 280             /* Get j neighbor index, and coordinate index */
 281             jnrA             = jjnr[jidx];
 282             jnrB             = jjnr[jidx+1];
 283             jnrC             = jjnr[jidx+2];
 284             jnrD             = jjnr[jidx+3];
 285
 286             /* Sign of each element will be negative for non-real atoms.
 287              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 288              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 289              */
 290             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 291             jnrA       = (jnrA>=0) ? jnrA : 0;
 292             jnrB       = (jnrB>=0) ? jnrB : 0;
 293             jnrC       = (jnrC>=0) ? jnrC : 0;
 294             jnrD       = (jnrD>=0) ? jnrD : 0;
 295
 296             j_coord_offsetA  = DIM*jnrA;
 297             j_coord_offsetB  = DIM*jnrB;
 298             j_coord_offsetC  = DIM*jnrC;
 299             j_coord_offsetD  = DIM*jnrD;
 300
 301             /* load j atom coordinates */
 302             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 303                                               x+j_coord_offsetC,x+j_coord_offsetD,
 304                                               &jx0,&jy0,&jz0);
 305
 306             /* Calculate displacement vector */
 307             dx00             = _mm_sub_ps(ix0,jx0);
 308             dy00             = _mm_sub_ps(iy0,jy0);
 309             dz00             = _mm_sub_ps(iz0,jz0);
 310
 311             /* Calculate squared distance and things based on it */
 312             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 313
 314             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 315
 316             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 317
 318             /* Load parameters for j particles */
 319             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 320                                                               charge+jnrC+0,charge+jnrD+0);
 321             vdwjidx0A        = 2*vdwtype[jnrA+0];
 322             vdwjidx0B        = 2*vdwtype[jnrB+0];
 323             vdwjidx0C        = 2*vdwtype[jnrC+0];
 324             vdwjidx0D        = 2*vdwtype[jnrD+0];
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             if (gmx_mm_any_lt(rsq00,rcutoff2))
 331             {
 332
 333             r00              = _mm_mul_ps(rsq00,rinv00);
 334             r00              = _mm_andnot_ps(dummy_mask,r00);
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq00             = _mm_mul_ps(iq0,jq0);
 338             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 339                                          vdwparam+vdwioffset0+vdwjidx0B,
 340                                          vdwparam+vdwioffset0+vdwjidx0C,
 341                                          vdwparam+vdwioffset0+vdwjidx0D,
 342                                          &c6_00,&c12_00);
 343
 344             /* REACTION-FIELD ELECTROSTATICS */
 345             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 346             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 347
 348             /* LENNARD-JONES DISPERSION/REPULSION */
 349
 350             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 351             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 352             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 353             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 354             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 355
 356             d                = _mm_sub_ps(r00,rswitch);
 357             d                = _mm_max_ps(d,_mm_setzero_ps());
 358             d2               = _mm_mul_ps(d,d);
 359             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 360
 361             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 362
 363             /* Evaluate switch function */
 364             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 365             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 366             vvdw             = _mm_mul_ps(vvdw,sw);
 367             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velec            = _mm_and_ps(velec,cutoff_mask);
 371             velec            = _mm_andnot_ps(dummy_mask,velec);
 372             velecsum         = _mm_add_ps(velecsum,velec);
 373             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 374             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 375             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 376
 377             fscal            = _mm_add_ps(felec,fvdw);
 378
 379             fscal            = _mm_and_ps(fscal,cutoff_mask);
 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             }
 398
 399             /* Inner loop uses 71 flops */
 400         }
 401
 402         /* End of innermost loop */
 403
 404         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 405                                               f+i_coord_offset,fshift+i_shift_offset);
 406
 407         ggid                        = gid[iidx];
 408         /* Update potential energies */
 409         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 410         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 411
 412         /* Increment number of inner iterations */
 413         inneriter                  += j_index_end - j_index_start;
 414
 415         /* Outer loop uses 12 flops */
 416     }
 417
 418     /* Increment number of outer iterations */
 419     outeriter        += nri;
 420
 421     /* Update outer/inner flops */
 422
 423     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*71);
 424 }
 425 /*
 426  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse2_single
 427  * Electrostatics interaction: ReactionField
 428  * VdW interaction:            LennardJones
 429  * Geometry:                   Particle-Particle
 430  * Calculate force/pot:        Force
 431  */
 432 void
 433 nb_kernel_ElecRFCut_VdwLJSw_GeomP1P1_F_sse2_single
 434                     (t_nblist * gmx_restrict                nlist,
 435                      rvec * gmx_restrict                    xx,
 436                      rvec * gmx_restrict                    ff,
 437                      t_forcerec * gmx_restrict              fr,
 438                      t_mdatoms * gmx_restrict               mdatoms,
 439                      nb_kernel_data_t * gmx_restrict        kernel_data,
 440                      t_nrnb * gmx_restrict                  nrnb)
 441 {
 442     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 443      * just 0 for non-waters.
 444      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 445      * jnr indices corresponding to data put in the four positions in the SIMD register.
 446      */
 447     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 448     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 449     int              jnrA,jnrB,jnrC,jnrD;
 450     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 451     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 452     real             shX,shY,shZ,rcutoff_scalar;
 453     real             *shiftvec,*fshift,*x,*f;
 454     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 455     int              vdwioffset0;
 456     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 457     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 458     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 459     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 460     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 461     real             *charge;
 462     int              nvdwtype;
 463     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 464     int              *vdwtype;
 465     real             *vdwparam;
 466     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 467     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 468     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 469     real             rswitch_scalar,d_scalar;
 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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 495     rcutoff_scalar   = fr->rcoulomb;
 496     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 497     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 498
 499     rswitch_scalar   = fr->rvdw_switch;
 500     rswitch          = _mm_set1_ps(rswitch_scalar);
 501     /* Setup switch parameters */
 502     d_scalar         = rcutoff_scalar-rswitch_scalar;
 503     d                = _mm_set1_ps(d_scalar);
 504     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 505     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 506     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 507     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 508     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 509     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 510
 511     /* Avoid stupid compiler warnings */
 512     jnrA = jnrB = jnrC = jnrD = 0;
 513     j_coord_offsetA = 0;
 514     j_coord_offsetB = 0;
 515     j_coord_offsetC = 0;
 516     j_coord_offsetD = 0;
 517
 518     outeriter        = 0;
 519     inneriter        = 0;
 520
 521     /* Start outer loop over neighborlists */
 522     for(iidx=0; iidx<nri; iidx++)
 523     {
 524         /* Load shift vector for this list */
 525         i_shift_offset   = DIM*shiftidx[iidx];
 526         shX              = shiftvec[i_shift_offset+XX];
 527         shY              = shiftvec[i_shift_offset+YY];
 528         shZ              = shiftvec[i_shift_offset+ZZ];
 529
 530         /* Load limits for loop over neighbors */
 531         j_index_start    = jindex[iidx];
 532         j_index_end      = jindex[iidx+1];
 533
 534         /* Get outer coordinate index */
 535         inr              = iinr[iidx];
 536         i_coord_offset   = DIM*inr;
 537
 538         /* Load i particle coords and add shift vector */
 539         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 540         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 541         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 542
 543         fix0             = _mm_setzero_ps();
 544         fiy0             = _mm_setzero_ps();
 545         fiz0             = _mm_setzero_ps();
 546
 547         /* Load parameters for i particles */
 548         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 549         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 550
 551         /* Start inner kernel loop */
 552         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 553         {
 554
 555             /* Get j neighbor index, and coordinate index */
 556             jnrA             = jjnr[jidx];
 557             jnrB             = jjnr[jidx+1];
 558             jnrC             = jjnr[jidx+2];
 559             jnrD             = jjnr[jidx+3];
 560
 561             j_coord_offsetA  = DIM*jnrA;
 562             j_coord_offsetB  = DIM*jnrB;
 563             j_coord_offsetC  = DIM*jnrC;
 564             j_coord_offsetD  = DIM*jnrD;
 565
 566             /* load j atom coordinates */
 567             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 568                                               x+j_coord_offsetC,x+j_coord_offsetD,
 569                                               &jx0,&jy0,&jz0);
 570
 571             /* Calculate displacement vector */
 572             dx00             = _mm_sub_ps(ix0,jx0);
 573             dy00             = _mm_sub_ps(iy0,jy0);
 574             dz00             = _mm_sub_ps(iz0,jz0);
 575
 576             /* Calculate squared distance and things based on it */
 577             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 578
 579             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 580
 581             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 582
 583             /* Load parameters for j particles */
 584             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 585                                                               charge+jnrC+0,charge+jnrD+0);
 586             vdwjidx0A        = 2*vdwtype[jnrA+0];
 587             vdwjidx0B        = 2*vdwtype[jnrB+0];
 588             vdwjidx0C        = 2*vdwtype[jnrC+0];
 589             vdwjidx0D        = 2*vdwtype[jnrD+0];
 590
 591             /**************************
 592              * CALCULATE INTERACTIONS *
 593              **************************/
 594
 595             if (gmx_mm_any_lt(rsq00,rcutoff2))
 596             {
 597
 598             r00              = _mm_mul_ps(rsq00,rinv00);
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq00             = _mm_mul_ps(iq0,jq0);
 602             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 603                                          vdwparam+vdwioffset0+vdwjidx0B,
 604                                          vdwparam+vdwioffset0+vdwjidx0C,
 605                                          vdwparam+vdwioffset0+vdwjidx0D,
 606                                          &c6_00,&c12_00);
 607
 608             /* REACTION-FIELD ELECTROSTATICS */
 609             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 610
 611             /* LENNARD-JONES DISPERSION/REPULSION */
 612
 613             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 614             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 615             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 616             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 617             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 618
 619             d                = _mm_sub_ps(r00,rswitch);
 620             d                = _mm_max_ps(d,_mm_setzero_ps());
 621             d2               = _mm_mul_ps(d,d);
 622             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 623
 624             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 625
 626             /* Evaluate switch function */
 627             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 628             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 629             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 630
 631             fscal            = _mm_add_ps(felec,fvdw);
 632
 633             fscal            = _mm_and_ps(fscal,cutoff_mask);
 634
 635             /* Calculate temporary vectorial force */
 636             tx               = _mm_mul_ps(fscal,dx00);
 637             ty               = _mm_mul_ps(fscal,dy00);
 638             tz               = _mm_mul_ps(fscal,dz00);
 639
 640             /* Update vectorial force */
 641             fix0             = _mm_add_ps(fix0,tx);
 642             fiy0             = _mm_add_ps(fiy0,ty);
 643             fiz0             = _mm_add_ps(fiz0,tz);
 644
 645             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 646                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 647                                                    tx,ty,tz);
 648
 649             }
 650
 651             /* Inner loop uses 61 flops */
 652         }
 653
 654         if(jidx<j_index_end)
 655         {
 656
 657             /* Get j neighbor index, and coordinate index */
 658             jnrA             = jjnr[jidx];
 659             jnrB             = jjnr[jidx+1];
 660             jnrC             = jjnr[jidx+2];
 661             jnrD             = jjnr[jidx+3];
 662
 663             /* Sign of each element will be negative for non-real atoms.
 664              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 665              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 666              */
 667             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 668             jnrA       = (jnrA>=0) ? jnrA : 0;
 669             jnrB       = (jnrB>=0) ? jnrB : 0;
 670             jnrC       = (jnrC>=0) ? jnrC : 0;
 671             jnrD       = (jnrD>=0) ? jnrD : 0;
 672
 673             j_coord_offsetA  = DIM*jnrA;
 674             j_coord_offsetB  = DIM*jnrB;
 675             j_coord_offsetC  = DIM*jnrC;
 676             j_coord_offsetD  = DIM*jnrD;
 677
 678             /* load j atom coordinates */
 679             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 680                                               x+j_coord_offsetC,x+j_coord_offsetD,
 681                                               &jx0,&jy0,&jz0);
 682
 683             /* Calculate displacement vector */
 684             dx00             = _mm_sub_ps(ix0,jx0);
 685             dy00             = _mm_sub_ps(iy0,jy0);
 686             dz00             = _mm_sub_ps(iz0,jz0);
 687
 688             /* Calculate squared distance and things based on it */
 689             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 690
 691             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 692
 693             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 694
 695             /* Load parameters for j particles */
 696             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 697                                                               charge+jnrC+0,charge+jnrD+0);
 698             vdwjidx0A        = 2*vdwtype[jnrA+0];
 699             vdwjidx0B        = 2*vdwtype[jnrB+0];
 700             vdwjidx0C        = 2*vdwtype[jnrC+0];
 701             vdwjidx0D        = 2*vdwtype[jnrD+0];
 702
 703             /**************************
 704              * CALCULATE INTERACTIONS *
 705              **************************/
 706
 707             if (gmx_mm_any_lt(rsq00,rcutoff2))
 708             {
 709
 710             r00              = _mm_mul_ps(rsq00,rinv00);
 711             r00              = _mm_andnot_ps(dummy_mask,r00);
 712
 713             /* Compute parameters for interactions between i and j atoms */
 714             qq00             = _mm_mul_ps(iq0,jq0);
 715             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 716                                          vdwparam+vdwioffset0+vdwjidx0B,
 717                                          vdwparam+vdwioffset0+vdwjidx0C,
 718                                          vdwparam+vdwioffset0+vdwjidx0D,
 719                                          &c6_00,&c12_00);
 720
 721             /* REACTION-FIELD ELECTROSTATICS */
 722             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 723
 724             /* LENNARD-JONES DISPERSION/REPULSION */
 725
 726             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 727             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 728             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 729             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 730             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 731
 732             d                = _mm_sub_ps(r00,rswitch);
 733             d                = _mm_max_ps(d,_mm_setzero_ps());
 734             d2               = _mm_mul_ps(d,d);
 735             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 736
 737             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 738
 739             /* Evaluate switch function */
 740             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 741             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 742             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 743
 744             fscal            = _mm_add_ps(felec,fvdw);
 745
 746             fscal            = _mm_and_ps(fscal,cutoff_mask);
 747
 748             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm_mul_ps(fscal,dx00);
 752             ty               = _mm_mul_ps(fscal,dy00);
 753             tz               = _mm_mul_ps(fscal,dz00);
 754
 755             /* Update vectorial force */
 756             fix0             = _mm_add_ps(fix0,tx);
 757             fiy0             = _mm_add_ps(fiy0,ty);
 758             fiz0             = _mm_add_ps(fiz0,tz);
 759
 760             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 761                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 762                                                    tx,ty,tz);
 763
 764             }
 765
 766             /* Inner loop uses 62 flops */
 767         }
 768
 769         /* End of innermost loop */
 770
 771         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 772                                               f+i_coord_offset,fshift+i_shift_offset);
 773
 774         /* Increment number of inner iterations */
 775         inneriter                  += j_index_end - j_index_start;
 776
 777         /* Outer loop uses 10 flops */
 778     }
 779
 780     /* Increment number of outer iterations */
 781     outeriter        += nri;
 782
 783     /* Update outer/inner flops */
 784
 785     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*62);
 786 }