src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecNone_VdwLJSw_GeomP1P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse4_1_single
  38  * Electrostatics interaction: None
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sse4_1_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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     int              nvdwtype;
  75     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  76     int              *vdwtype;
  77     real             *vdwparam;
  78     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  79     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  80     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  81     real             rswitch_scalar,d_scalar;
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     nvdwtype         = fr->ntype;
  98     vdwparam         = fr->nbfp;
  99     vdwtype          = mdatoms->typeA;
 100
 101     rcutoff_scalar   = fr->rvdw;
 102     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 103     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 104
 105     rswitch_scalar   = fr->rvdw_switch;
 106     rswitch          = _mm_set1_ps(rswitch_scalar);
 107     /* Setup switch parameters */
 108     d_scalar         = rcutoff_scalar-rswitch_scalar;
 109     d                = _mm_set1_ps(d_scalar);
 110     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 111     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 112     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 113     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 114     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 115     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = jnrC = jnrD = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121     j_coord_offsetC = 0;
 122     j_coord_offsetD = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     for(iidx=0;iidx<4*DIM;iidx++)
 128     {
 129         scratch[iidx] = 0.0;
 130     }
 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
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 148
 149         fix0             = _mm_setzero_ps();
 150         fiy0             = _mm_setzero_ps();
 151         fiz0             = _mm_setzero_ps();
 152
 153         /* Load parameters for i particles */
 154         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 155
 156         /* Reset potential sums */
 157         vvdwsum          = _mm_setzero_ps();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170             j_coord_offsetC  = DIM*jnrC;
 171             j_coord_offsetD  = DIM*jnrD;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               x+j_coord_offsetC,x+j_coord_offsetD,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _mm_sub_ps(ix0,jx0);
 180             dy00             = _mm_sub_ps(iy0,jy0);
 181             dz00             = _mm_sub_ps(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 187
 188             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 189
 190             /* Load parameters for j particles */
 191             vdwjidx0A        = 2*vdwtype[jnrA+0];
 192             vdwjidx0B        = 2*vdwtype[jnrB+0];
 193             vdwjidx0C        = 2*vdwtype[jnrC+0];
 194             vdwjidx0D        = 2*vdwtype[jnrD+0];
 195
 196             /**************************
 197              * CALCULATE INTERACTIONS *
 198              **************************/
 199
 200             if (gmx_mm_any_lt(rsq00,rcutoff2))
 201             {
 202
 203             r00              = _mm_mul_ps(rsq00,rinv00);
 204
 205             /* Compute parameters for interactions between i and j atoms */
 206             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,
 208                                          vdwparam+vdwioffset0+vdwjidx0C,
 209                                          vdwparam+vdwioffset0+vdwjidx0D,
 210                                          &c6_00,&c12_00);
 211
 212             /* LENNARD-JONES DISPERSION/REPULSION */
 213
 214             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 215             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 216             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 217             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 218             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 219
 220             d                = _mm_sub_ps(r00,rswitch);
 221             d                = _mm_max_ps(d,_mm_setzero_ps());
 222             d2               = _mm_mul_ps(d,d);
 223             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)))))));
 224
 225             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 226
 227             /* Evaluate switch function */
 228             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 229             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 230             vvdw             = _mm_mul_ps(vvdw,sw);
 231             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 235             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 236
 237             fscal            = fvdw;
 238
 239             fscal            = _mm_and_ps(fscal,cutoff_mask);
 240
 241             /* Calculate temporary vectorial force */
 242             tx               = _mm_mul_ps(fscal,dx00);
 243             ty               = _mm_mul_ps(fscal,dy00);
 244             tz               = _mm_mul_ps(fscal,dz00);
 245
 246             /* Update vectorial force */
 247             fix0             = _mm_add_ps(fix0,tx);
 248             fiy0             = _mm_add_ps(fiy0,ty);
 249             fiz0             = _mm_add_ps(fiz0,tz);
 250
 251             fjptrA             = f+j_coord_offsetA;
 252             fjptrB             = f+j_coord_offsetB;
 253             fjptrC             = f+j_coord_offsetC;
 254             fjptrD             = f+j_coord_offsetD;
 255             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 256
 257             }
 258
 259             /* Inner loop uses 59 flops */
 260         }
 261
 262         if(jidx<j_index_end)
 263         {
 264
 265             /* Get j neighbor index, and coordinate index */
 266             jnrlistA         = jjnr[jidx];
 267             jnrlistB         = jjnr[jidx+1];
 268             jnrlistC         = jjnr[jidx+2];
 269             jnrlistD         = jjnr[jidx+3];
 270             /* Sign of each element will be negative for non-real atoms.
 271              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 272              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 273              */
 274             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 275             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 276             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 277             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 278             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 279             j_coord_offsetA  = DIM*jnrA;
 280             j_coord_offsetB  = DIM*jnrB;
 281             j_coord_offsetC  = DIM*jnrC;
 282             j_coord_offsetD  = DIM*jnrD;
 283
 284             /* load j atom coordinates */
 285             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 286                                               x+j_coord_offsetC,x+j_coord_offsetD,
 287                                               &jx0,&jy0,&jz0);
 288
 289             /* Calculate displacement vector */
 290             dx00             = _mm_sub_ps(ix0,jx0);
 291             dy00             = _mm_sub_ps(iy0,jy0);
 292             dz00             = _mm_sub_ps(iz0,jz0);
 293
 294             /* Calculate squared distance and things based on it */
 295             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 296
 297             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 298
 299             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 300
 301             /* Load parameters for j particles */
 302             vdwjidx0A        = 2*vdwtype[jnrA+0];
 303             vdwjidx0B        = 2*vdwtype[jnrB+0];
 304             vdwjidx0C        = 2*vdwtype[jnrC+0];
 305             vdwjidx0D        = 2*vdwtype[jnrD+0];
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             if (gmx_mm_any_lt(rsq00,rcutoff2))
 312             {
 313
 314             r00              = _mm_mul_ps(rsq00,rinv00);
 315             r00              = _mm_andnot_ps(dummy_mask,r00);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 319                                          vdwparam+vdwioffset0+vdwjidx0B,
 320                                          vdwparam+vdwioffset0+vdwjidx0C,
 321                                          vdwparam+vdwioffset0+vdwjidx0D,
 322                                          &c6_00,&c12_00);
 323
 324             /* LENNARD-JONES DISPERSION/REPULSION */
 325
 326             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 327             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 328             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 329             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 330             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 331
 332             d                = _mm_sub_ps(r00,rswitch);
 333             d                = _mm_max_ps(d,_mm_setzero_ps());
 334             d2               = _mm_mul_ps(d,d);
 335             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)))))));
 336
 337             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 338
 339             /* Evaluate switch function */
 340             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 341             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 342             vvdw             = _mm_mul_ps(vvdw,sw);
 343             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 347             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 348             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 349
 350             fscal            = fvdw;
 351
 352             fscal            = _mm_and_ps(fscal,cutoff_mask);
 353
 354             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm_mul_ps(fscal,dx00);
 358             ty               = _mm_mul_ps(fscal,dy00);
 359             tz               = _mm_mul_ps(fscal,dz00);
 360
 361             /* Update vectorial force */
 362             fix0             = _mm_add_ps(fix0,tx);
 363             fiy0             = _mm_add_ps(fiy0,ty);
 364             fiz0             = _mm_add_ps(fiz0,tz);
 365
 366             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 367             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 368             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 369             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 370             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 371
 372             }
 373
 374             /* Inner loop uses 60 flops */
 375         }
 376
 377         /* End of innermost loop */
 378
 379         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 380                                               f+i_coord_offset,fshift+i_shift_offset);
 381
 382         ggid                        = gid[iidx];
 383         /* Update potential energies */
 384         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 385
 386         /* Increment number of inner iterations */
 387         inneriter                  += j_index_end - j_index_start;
 388
 389         /* Outer loop uses 7 flops */
 390     }
 391
 392     /* Increment number of outer iterations */
 393     outeriter        += nri;
 394
 395     /* Update outer/inner flops */
 396
 397     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*60);
 398 }
 399 /*
 400  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse4_1_single
 401  * Electrostatics interaction: None
 402  * VdW interaction:            LennardJones
 403  * Geometry:                   Particle-Particle
 404  * Calculate force/pot:        Force
 405  */
 406 void
 407 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sse4_1_single
 408                     (t_nblist * gmx_restrict                nlist,
 409                      rvec * gmx_restrict                    xx,
 410                      rvec * gmx_restrict                    ff,
 411                      t_forcerec * gmx_restrict              fr,
 412                      t_mdatoms * gmx_restrict               mdatoms,
 413                      nb_kernel_data_t * gmx_restrict        kernel_data,
 414                      t_nrnb * gmx_restrict                  nrnb)
 415 {
 416     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 417      * just 0 for non-waters.
 418      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 419      * jnr indices corresponding to data put in the four positions in the SIMD register.
 420      */
 421     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 422     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 423     int              jnrA,jnrB,jnrC,jnrD;
 424     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 425     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 426     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 427     real             rcutoff_scalar;
 428     real             *shiftvec,*fshift,*x,*f;
 429     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 430     real             scratch[4*DIM];
 431     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 432     int              vdwioffset0;
 433     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 434     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 435     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 436     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 437     int              nvdwtype;
 438     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 439     int              *vdwtype;
 440     real             *vdwparam;
 441     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 442     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 443     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 444     real             rswitch_scalar,d_scalar;
 445     __m128           dummy_mask,cutoff_mask;
 446     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 447     __m128           one     = _mm_set1_ps(1.0);
 448     __m128           two     = _mm_set1_ps(2.0);
 449     x                = xx[0];
 450     f                = ff[0];
 451
 452     nri              = nlist->nri;
 453     iinr             = nlist->iinr;
 454     jindex           = nlist->jindex;
 455     jjnr             = nlist->jjnr;
 456     shiftidx         = nlist->shift;
 457     gid              = nlist->gid;
 458     shiftvec         = fr->shift_vec[0];
 459     fshift           = fr->fshift[0];
 460     nvdwtype         = fr->ntype;
 461     vdwparam         = fr->nbfp;
 462     vdwtype          = mdatoms->typeA;
 463
 464     rcutoff_scalar   = fr->rvdw;
 465     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 466     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 467
 468     rswitch_scalar   = fr->rvdw_switch;
 469     rswitch          = _mm_set1_ps(rswitch_scalar);
 470     /* Setup switch parameters */
 471     d_scalar         = rcutoff_scalar-rswitch_scalar;
 472     d                = _mm_set1_ps(d_scalar);
 473     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 474     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 475     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 476     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 477     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 478     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 479
 480     /* Avoid stupid compiler warnings */
 481     jnrA = jnrB = jnrC = jnrD = 0;
 482     j_coord_offsetA = 0;
 483     j_coord_offsetB = 0;
 484     j_coord_offsetC = 0;
 485     j_coord_offsetD = 0;
 486
 487     outeriter        = 0;
 488     inneriter        = 0;
 489
 490     for(iidx=0;iidx<4*DIM;iidx++)
 491     {
 492         scratch[iidx] = 0.0;
 493     }
 494
 495     /* Start outer loop over neighborlists */
 496     for(iidx=0; iidx<nri; iidx++)
 497     {
 498         /* Load shift vector for this list */
 499         i_shift_offset   = DIM*shiftidx[iidx];
 500
 501         /* Load limits for loop over neighbors */
 502         j_index_start    = jindex[iidx];
 503         j_index_end      = jindex[iidx+1];
 504
 505         /* Get outer coordinate index */
 506         inr              = iinr[iidx];
 507         i_coord_offset   = DIM*inr;
 508
 509         /* Load i particle coords and add shift vector */
 510         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 511
 512         fix0             = _mm_setzero_ps();
 513         fiy0             = _mm_setzero_ps();
 514         fiz0             = _mm_setzero_ps();
 515
 516         /* Load parameters for i particles */
 517         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 518
 519         /* Start inner kernel loop */
 520         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 521         {
 522
 523             /* Get j neighbor index, and coordinate index */
 524             jnrA             = jjnr[jidx];
 525             jnrB             = jjnr[jidx+1];
 526             jnrC             = jjnr[jidx+2];
 527             jnrD             = jjnr[jidx+3];
 528             j_coord_offsetA  = DIM*jnrA;
 529             j_coord_offsetB  = DIM*jnrB;
 530             j_coord_offsetC  = DIM*jnrC;
 531             j_coord_offsetD  = DIM*jnrD;
 532
 533             /* load j atom coordinates */
 534             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 535                                               x+j_coord_offsetC,x+j_coord_offsetD,
 536                                               &jx0,&jy0,&jz0);
 537
 538             /* Calculate displacement vector */
 539             dx00             = _mm_sub_ps(ix0,jx0);
 540             dy00             = _mm_sub_ps(iy0,jy0);
 541             dz00             = _mm_sub_ps(iz0,jz0);
 542
 543             /* Calculate squared distance and things based on it */
 544             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 545
 546             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 547
 548             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 549
 550             /* Load parameters for j particles */
 551             vdwjidx0A        = 2*vdwtype[jnrA+0];
 552             vdwjidx0B        = 2*vdwtype[jnrB+0];
 553             vdwjidx0C        = 2*vdwtype[jnrC+0];
 554             vdwjidx0D        = 2*vdwtype[jnrD+0];
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             if (gmx_mm_any_lt(rsq00,rcutoff2))
 561             {
 562
 563             r00              = _mm_mul_ps(rsq00,rinv00);
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 567                                          vdwparam+vdwioffset0+vdwjidx0B,
 568                                          vdwparam+vdwioffset0+vdwjidx0C,
 569                                          vdwparam+vdwioffset0+vdwjidx0D,
 570                                          &c6_00,&c12_00);
 571
 572             /* LENNARD-JONES DISPERSION/REPULSION */
 573
 574             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 575             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 576             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 577             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 578             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 579
 580             d                = _mm_sub_ps(r00,rswitch);
 581             d                = _mm_max_ps(d,_mm_setzero_ps());
 582             d2               = _mm_mul_ps(d,d);
 583             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)))))));
 584
 585             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 586
 587             /* Evaluate switch function */
 588             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 589             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 590             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 591
 592             fscal            = fvdw;
 593
 594             fscal            = _mm_and_ps(fscal,cutoff_mask);
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm_mul_ps(fscal,dx00);
 598             ty               = _mm_mul_ps(fscal,dy00);
 599             tz               = _mm_mul_ps(fscal,dz00);
 600
 601             /* Update vectorial force */
 602             fix0             = _mm_add_ps(fix0,tx);
 603             fiy0             = _mm_add_ps(fiy0,ty);
 604             fiz0             = _mm_add_ps(fiz0,tz);
 605
 606             fjptrA             = f+j_coord_offsetA;
 607             fjptrB             = f+j_coord_offsetB;
 608             fjptrC             = f+j_coord_offsetC;
 609             fjptrD             = f+j_coord_offsetD;
 610             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 611
 612             }
 613
 614             /* Inner loop uses 56 flops */
 615         }
 616
 617         if(jidx<j_index_end)
 618         {
 619
 620             /* Get j neighbor index, and coordinate index */
 621             jnrlistA         = jjnr[jidx];
 622             jnrlistB         = jjnr[jidx+1];
 623             jnrlistC         = jjnr[jidx+2];
 624             jnrlistD         = jjnr[jidx+3];
 625             /* Sign of each element will be negative for non-real atoms.
 626              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 627              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 628              */
 629             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 630             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 631             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 632             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 633             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 634             j_coord_offsetA  = DIM*jnrA;
 635             j_coord_offsetB  = DIM*jnrB;
 636             j_coord_offsetC  = DIM*jnrC;
 637             j_coord_offsetD  = DIM*jnrD;
 638
 639             /* load j atom coordinates */
 640             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 641                                               x+j_coord_offsetC,x+j_coord_offsetD,
 642                                               &jx0,&jy0,&jz0);
 643
 644             /* Calculate displacement vector */
 645             dx00             = _mm_sub_ps(ix0,jx0);
 646             dy00             = _mm_sub_ps(iy0,jy0);
 647             dz00             = _mm_sub_ps(iz0,jz0);
 648
 649             /* Calculate squared distance and things based on it */
 650             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 651
 652             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 653
 654             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 655
 656             /* Load parameters for j particles */
 657             vdwjidx0A        = 2*vdwtype[jnrA+0];
 658             vdwjidx0B        = 2*vdwtype[jnrB+0];
 659             vdwjidx0C        = 2*vdwtype[jnrC+0];
 660             vdwjidx0D        = 2*vdwtype[jnrD+0];
 661
 662             /**************************
 663              * CALCULATE INTERACTIONS *
 664              **************************/
 665
 666             if (gmx_mm_any_lt(rsq00,rcutoff2))
 667             {
 668
 669             r00              = _mm_mul_ps(rsq00,rinv00);
 670             r00              = _mm_andnot_ps(dummy_mask,r00);
 671
 672             /* Compute parameters for interactions between i and j atoms */
 673             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 674                                          vdwparam+vdwioffset0+vdwjidx0B,
 675                                          vdwparam+vdwioffset0+vdwjidx0C,
 676                                          vdwparam+vdwioffset0+vdwjidx0D,
 677                                          &c6_00,&c12_00);
 678
 679             /* LENNARD-JONES DISPERSION/REPULSION */
 680
 681             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 682             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 683             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 684             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 685             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 686
 687             d                = _mm_sub_ps(r00,rswitch);
 688             d                = _mm_max_ps(d,_mm_setzero_ps());
 689             d2               = _mm_mul_ps(d,d);
 690             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)))))));
 691
 692             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 693
 694             /* Evaluate switch function */
 695             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 696             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 697             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 698
 699             fscal            = fvdw;
 700
 701             fscal            = _mm_and_ps(fscal,cutoff_mask);
 702
 703             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm_mul_ps(fscal,dx00);
 707             ty               = _mm_mul_ps(fscal,dy00);
 708             tz               = _mm_mul_ps(fscal,dz00);
 709
 710             /* Update vectorial force */
 711             fix0             = _mm_add_ps(fix0,tx);
 712             fiy0             = _mm_add_ps(fiy0,ty);
 713             fiz0             = _mm_add_ps(fiz0,tz);
 714
 715             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 716             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 717             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 718             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 719             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 720
 721             }
 722
 723             /* Inner loop uses 57 flops */
 724         }
 725
 726         /* End of innermost loop */
 727
 728         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 729                                               f+i_coord_offset,fshift+i_shift_offset);
 730
 731         /* Increment number of inner iterations */
 732         inneriter                  += j_index_end - j_index_start;
 733
 734         /* Outer loop uses 6 flops */
 735     }
 736
 737     /* Increment number of outer iterations */
 738     outeriter        += nri;
 739
 740     /* Update outer/inner flops */
 741
 742     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*57);
 743 }