src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecNone_VdwLJSh_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_ElecNone_VdwLJSh_GeomP1P1_VF_sse2_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_VdwLJSh_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     int              nvdwtype;
  72     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  73     int              *vdwtype;
  74     real             *vdwparam;
  75     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  76     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  77     __m128           dummy_mask,cutoff_mask;
  78     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  79     __m128           one     = _mm_set1_ps(1.0);
  80     __m128           two     = _mm_set1_ps(2.0);
  81     x                = xx[0];
  82     f                = ff[0];
  83
  84     nri              = nlist->nri;
  85     iinr             = nlist->iinr;
  86     jindex           = nlist->jindex;
  87     jjnr             = nlist->jjnr;
  88     shiftidx         = nlist->shift;
  89     gid              = nlist->gid;
  90     shiftvec         = fr->shift_vec[0];
  91     fshift           = fr->fshift[0];
  92     nvdwtype         = fr->ntype;
  93     vdwparam         = fr->nbfp;
  94     vdwtype          = mdatoms->typeA;
  95
  96     rcutoff_scalar   = fr->rvdw;
  97     rcutoff          = _mm_set1_ps(rcutoff_scalar);
  98     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
  99
 100     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 101     rvdw             = _mm_set1_ps(fr->rvdw);
 102
 103     /* Avoid stupid compiler warnings */
 104     jnrA = jnrB = jnrC = jnrD = 0;
 105     j_coord_offsetA = 0;
 106     j_coord_offsetB = 0;
 107     j_coord_offsetC = 0;
 108     j_coord_offsetD = 0;
 109
 110     outeriter        = 0;
 111     inneriter        = 0;
 112
 113     /* Start outer loop over neighborlists */
 114     for(iidx=0; iidx<nri; iidx++)
 115     {
 116         /* Load shift vector for this list */
 117         i_shift_offset   = DIM*shiftidx[iidx];
 118         shX              = shiftvec[i_shift_offset+XX];
 119         shY              = shiftvec[i_shift_offset+YY];
 120         shZ              = shiftvec[i_shift_offset+ZZ];
 121
 122         /* Load limits for loop over neighbors */
 123         j_index_start    = jindex[iidx];
 124         j_index_end      = jindex[iidx+1];
 125
 126         /* Get outer coordinate index */
 127         inr              = iinr[iidx];
 128         i_coord_offset   = DIM*inr;
 129
 130         /* Load i particle coords and add shift vector */
 131         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 132         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 133         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 134
 135         fix0             = _mm_setzero_ps();
 136         fiy0             = _mm_setzero_ps();
 137         fiz0             = _mm_setzero_ps();
 138
 139         /* Load parameters for i particles */
 140         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142         /* Reset potential sums */
 143         vvdwsum          = _mm_setzero_ps();
 144
 145         /* Start inner kernel loop */
 146         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 147         {
 148
 149             /* Get j neighbor index, and coordinate index */
 150             jnrA             = jjnr[jidx];
 151             jnrB             = jjnr[jidx+1];
 152             jnrC             = jjnr[jidx+2];
 153             jnrD             = jjnr[jidx+3];
 154
 155             j_coord_offsetA  = DIM*jnrA;
 156             j_coord_offsetB  = DIM*jnrB;
 157             j_coord_offsetC  = DIM*jnrC;
 158             j_coord_offsetD  = DIM*jnrD;
 159
 160             /* load j atom coordinates */
 161             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 162                                               x+j_coord_offsetC,x+j_coord_offsetD,
 163                                               &jx0,&jy0,&jz0);
 164
 165             /* Calculate displacement vector */
 166             dx00             = _mm_sub_ps(ix0,jx0);
 167             dy00             = _mm_sub_ps(iy0,jy0);
 168             dz00             = _mm_sub_ps(iz0,jz0);
 169
 170             /* Calculate squared distance and things based on it */
 171             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 172
 173             rinvsq00         = gmx_mm_inv_ps(rsq00);
 174
 175             /* Load parameters for j particles */
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178             vdwjidx0C        = 2*vdwtype[jnrC+0];
 179             vdwjidx0D        = 2*vdwtype[jnrD+0];
 180
 181             /**************************
 182              * CALCULATE INTERACTIONS *
 183              **************************/
 184
 185             if (gmx_mm_any_lt(rsq00,rcutoff2))
 186             {
 187
 188             /* Compute parameters for interactions between i and j atoms */
 189             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 190                                          vdwparam+vdwioffset0+vdwjidx0B,
 191                                          vdwparam+vdwioffset0+vdwjidx0C,
 192                                          vdwparam+vdwioffset0+vdwjidx0D,
 193                                          &c6_00,&c12_00);
 194
 195             /* LENNARD-JONES DISPERSION/REPULSION */
 196
 197             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 198             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 199             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 200             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 201                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 202             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 203
 204             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 205
 206             /* Update potential sum for this i atom from the interaction with this j atom. */
 207             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 208             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 209
 210             fscal            = fvdw;
 211
 212             fscal            = _mm_and_ps(fscal,cutoff_mask);
 213
 214             /* Calculate temporary vectorial force */
 215             tx               = _mm_mul_ps(fscal,dx00);
 216             ty               = _mm_mul_ps(fscal,dy00);
 217             tz               = _mm_mul_ps(fscal,dz00);
 218
 219             /* Update vectorial force */
 220             fix0             = _mm_add_ps(fix0,tx);
 221             fiy0             = _mm_add_ps(fiy0,ty);
 222             fiz0             = _mm_add_ps(fiz0,tz);
 223
 224             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 225                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 226                                                    tx,ty,tz);
 227
 228             }
 229
 230             /* Inner loop uses 41 flops */
 231         }
 232
 233         if(jidx<j_index_end)
 234         {
 235
 236             /* Get j neighbor index, and coordinate index */
 237             jnrA             = jjnr[jidx];
 238             jnrB             = jjnr[jidx+1];
 239             jnrC             = jjnr[jidx+2];
 240             jnrD             = jjnr[jidx+3];
 241
 242             /* Sign of each element will be negative for non-real atoms.
 243              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 244              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 245              */
 246             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 247             jnrA       = (jnrA>=0) ? jnrA : 0;
 248             jnrB       = (jnrB>=0) ? jnrB : 0;
 249             jnrC       = (jnrC>=0) ? jnrC : 0;
 250             jnrD       = (jnrD>=0) ? jnrD : 0;
 251
 252             j_coord_offsetA  = DIM*jnrA;
 253             j_coord_offsetB  = DIM*jnrB;
 254             j_coord_offsetC  = DIM*jnrC;
 255             j_coord_offsetD  = DIM*jnrD;
 256
 257             /* load j atom coordinates */
 258             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 259                                               x+j_coord_offsetC,x+j_coord_offsetD,
 260                                               &jx0,&jy0,&jz0);
 261
 262             /* Calculate displacement vector */
 263             dx00             = _mm_sub_ps(ix0,jx0);
 264             dy00             = _mm_sub_ps(iy0,jy0);
 265             dz00             = _mm_sub_ps(iz0,jz0);
 266
 267             /* Calculate squared distance and things based on it */
 268             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 269
 270             rinvsq00         = gmx_mm_inv_ps(rsq00);
 271
 272             /* Load parameters for j particles */
 273             vdwjidx0A        = 2*vdwtype[jnrA+0];
 274             vdwjidx0B        = 2*vdwtype[jnrB+0];
 275             vdwjidx0C        = 2*vdwtype[jnrC+0];
 276             vdwjidx0D        = 2*vdwtype[jnrD+0];
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             if (gmx_mm_any_lt(rsq00,rcutoff2))
 283             {
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 287                                          vdwparam+vdwioffset0+vdwjidx0B,
 288                                          vdwparam+vdwioffset0+vdwjidx0C,
 289                                          vdwparam+vdwioffset0+vdwjidx0D,
 290                                          &c6_00,&c12_00);
 291
 292             /* LENNARD-JONES DISPERSION/REPULSION */
 293
 294             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 295             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 296             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 297             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 298                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 299             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 300
 301             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 305             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 306             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 307
 308             fscal            = fvdw;
 309
 310             fscal            = _mm_and_ps(fscal,cutoff_mask);
 311
 312             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm_mul_ps(fscal,dx00);
 316             ty               = _mm_mul_ps(fscal,dy00);
 317             tz               = _mm_mul_ps(fscal,dz00);
 318
 319             /* Update vectorial force */
 320             fix0             = _mm_add_ps(fix0,tx);
 321             fiy0             = _mm_add_ps(fiy0,ty);
 322             fiz0             = _mm_add_ps(fiz0,tz);
 323
 324             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 325                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 326                                                    tx,ty,tz);
 327
 328             }
 329
 330             /* Inner loop uses 41 flops */
 331         }
 332
 333         /* End of innermost loop */
 334
 335         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 336                                               f+i_coord_offset,fshift+i_shift_offset);
 337
 338         ggid                        = gid[iidx];
 339         /* Update potential energies */
 340         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 341
 342         /* Increment number of inner iterations */
 343         inneriter                  += j_index_end - j_index_start;
 344
 345         /* Outer loop uses 10 flops */
 346     }
 347
 348     /* Increment number of outer iterations */
 349     outeriter        += nri;
 350
 351     /* Update outer/inner flops */
 352
 353     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*10 + inneriter*41);
 354 }
 355 /*
 356  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse2_single
 357  * Electrostatics interaction: None
 358  * VdW interaction:            LennardJones
 359  * Geometry:                   Particle-Particle
 360  * Calculate force/pot:        Force
 361  */
 362 void
 363 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse2_single
 364                     (t_nblist * gmx_restrict                nlist,
 365                      rvec * gmx_restrict                    xx,
 366                      rvec * gmx_restrict                    ff,
 367                      t_forcerec * gmx_restrict              fr,
 368                      t_mdatoms * gmx_restrict               mdatoms,
 369                      nb_kernel_data_t * gmx_restrict        kernel_data,
 370                      t_nrnb * gmx_restrict                  nrnb)
 371 {
 372     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 373      * just 0 for non-waters.
 374      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 375      * jnr indices corresponding to data put in the four positions in the SIMD register.
 376      */
 377     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 378     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 379     int              jnrA,jnrB,jnrC,jnrD;
 380     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 381     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 382     real             shX,shY,shZ,rcutoff_scalar;
 383     real             *shiftvec,*fshift,*x,*f;
 384     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 385     int              vdwioffset0;
 386     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 387     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 388     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 389     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 390     int              nvdwtype;
 391     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 392     int              *vdwtype;
 393     real             *vdwparam;
 394     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 395     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 396     __m128           dummy_mask,cutoff_mask;
 397     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 398     __m128           one     = _mm_set1_ps(1.0);
 399     __m128           two     = _mm_set1_ps(2.0);
 400     x                = xx[0];
 401     f                = ff[0];
 402
 403     nri              = nlist->nri;
 404     iinr             = nlist->iinr;
 405     jindex           = nlist->jindex;
 406     jjnr             = nlist->jjnr;
 407     shiftidx         = nlist->shift;
 408     gid              = nlist->gid;
 409     shiftvec         = fr->shift_vec[0];
 410     fshift           = fr->fshift[0];
 411     nvdwtype         = fr->ntype;
 412     vdwparam         = fr->nbfp;
 413     vdwtype          = mdatoms->typeA;
 414
 415     rcutoff_scalar   = fr->rvdw;
 416     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 417     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 418
 419     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 420     rvdw             = _mm_set1_ps(fr->rvdw);
 421
 422     /* Avoid stupid compiler warnings */
 423     jnrA = jnrB = jnrC = jnrD = 0;
 424     j_coord_offsetA = 0;
 425     j_coord_offsetB = 0;
 426     j_coord_offsetC = 0;
 427     j_coord_offsetD = 0;
 428
 429     outeriter        = 0;
 430     inneriter        = 0;
 431
 432     /* Start outer loop over neighborlists */
 433     for(iidx=0; iidx<nri; iidx++)
 434     {
 435         /* Load shift vector for this list */
 436         i_shift_offset   = DIM*shiftidx[iidx];
 437         shX              = shiftvec[i_shift_offset+XX];
 438         shY              = shiftvec[i_shift_offset+YY];
 439         shZ              = shiftvec[i_shift_offset+ZZ];
 440
 441         /* Load limits for loop over neighbors */
 442         j_index_start    = jindex[iidx];
 443         j_index_end      = jindex[iidx+1];
 444
 445         /* Get outer coordinate index */
 446         inr              = iinr[iidx];
 447         i_coord_offset   = DIM*inr;
 448
 449         /* Load i particle coords and add shift vector */
 450         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 451         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 452         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 453
 454         fix0             = _mm_setzero_ps();
 455         fiy0             = _mm_setzero_ps();
 456         fiz0             = _mm_setzero_ps();
 457
 458         /* Load parameters for i particles */
 459         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 460
 461         /* Start inner kernel loop */
 462         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 463         {
 464
 465             /* Get j neighbor index, and coordinate index */
 466             jnrA             = jjnr[jidx];
 467             jnrB             = jjnr[jidx+1];
 468             jnrC             = jjnr[jidx+2];
 469             jnrD             = jjnr[jidx+3];
 470
 471             j_coord_offsetA  = DIM*jnrA;
 472             j_coord_offsetB  = DIM*jnrB;
 473             j_coord_offsetC  = DIM*jnrC;
 474             j_coord_offsetD  = DIM*jnrD;
 475
 476             /* load j atom coordinates */
 477             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 478                                               x+j_coord_offsetC,x+j_coord_offsetD,
 479                                               &jx0,&jy0,&jz0);
 480
 481             /* Calculate displacement vector */
 482             dx00             = _mm_sub_ps(ix0,jx0);
 483             dy00             = _mm_sub_ps(iy0,jy0);
 484             dz00             = _mm_sub_ps(iz0,jz0);
 485
 486             /* Calculate squared distance and things based on it */
 487             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 488
 489             rinvsq00         = gmx_mm_inv_ps(rsq00);
 490
 491             /* Load parameters for j particles */
 492             vdwjidx0A        = 2*vdwtype[jnrA+0];
 493             vdwjidx0B        = 2*vdwtype[jnrB+0];
 494             vdwjidx0C        = 2*vdwtype[jnrC+0];
 495             vdwjidx0D        = 2*vdwtype[jnrD+0];
 496
 497             /**************************
 498              * CALCULATE INTERACTIONS *
 499              **************************/
 500
 501             if (gmx_mm_any_lt(rsq00,rcutoff2))
 502             {
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 506                                          vdwparam+vdwioffset0+vdwjidx0B,
 507                                          vdwparam+vdwioffset0+vdwjidx0C,
 508                                          vdwparam+vdwioffset0+vdwjidx0D,
 509                                          &c6_00,&c12_00);
 510
 511             /* LENNARD-JONES DISPERSION/REPULSION */
 512
 513             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 514             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 515
 516             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 517
 518             fscal            = fvdw;
 519
 520             fscal            = _mm_and_ps(fscal,cutoff_mask);
 521
 522             /* Calculate temporary vectorial force */
 523             tx               = _mm_mul_ps(fscal,dx00);
 524             ty               = _mm_mul_ps(fscal,dy00);
 525             tz               = _mm_mul_ps(fscal,dz00);
 526
 527             /* Update vectorial force */
 528             fix0             = _mm_add_ps(fix0,tx);
 529             fiy0             = _mm_add_ps(fiy0,ty);
 530             fiz0             = _mm_add_ps(fiz0,tz);
 531
 532             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 533                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 534                                                    tx,ty,tz);
 535
 536             }
 537
 538             /* Inner loop uses 30 flops */
 539         }
 540
 541         if(jidx<j_index_end)
 542         {
 543
 544             /* Get j neighbor index, and coordinate index */
 545             jnrA             = jjnr[jidx];
 546             jnrB             = jjnr[jidx+1];
 547             jnrC             = jjnr[jidx+2];
 548             jnrD             = jjnr[jidx+3];
 549
 550             /* Sign of each element will be negative for non-real atoms.
 551              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 552              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 553              */
 554             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 555             jnrA       = (jnrA>=0) ? jnrA : 0;
 556             jnrB       = (jnrB>=0) ? jnrB : 0;
 557             jnrC       = (jnrC>=0) ? jnrC : 0;
 558             jnrD       = (jnrD>=0) ? jnrD : 0;
 559
 560             j_coord_offsetA  = DIM*jnrA;
 561             j_coord_offsetB  = DIM*jnrB;
 562             j_coord_offsetC  = DIM*jnrC;
 563             j_coord_offsetD  = DIM*jnrD;
 564
 565             /* load j atom coordinates */
 566             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 567                                               x+j_coord_offsetC,x+j_coord_offsetD,
 568                                               &jx0,&jy0,&jz0);
 569
 570             /* Calculate displacement vector */
 571             dx00             = _mm_sub_ps(ix0,jx0);
 572             dy00             = _mm_sub_ps(iy0,jy0);
 573             dz00             = _mm_sub_ps(iz0,jz0);
 574
 575             /* Calculate squared distance and things based on it */
 576             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 577
 578             rinvsq00         = gmx_mm_inv_ps(rsq00);
 579
 580             /* Load parameters for j particles */
 581             vdwjidx0A        = 2*vdwtype[jnrA+0];
 582             vdwjidx0B        = 2*vdwtype[jnrB+0];
 583             vdwjidx0C        = 2*vdwtype[jnrC+0];
 584             vdwjidx0D        = 2*vdwtype[jnrD+0];
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             if (gmx_mm_any_lt(rsq00,rcutoff2))
 591             {
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 595                                          vdwparam+vdwioffset0+vdwjidx0B,
 596                                          vdwparam+vdwioffset0+vdwjidx0C,
 597                                          vdwparam+vdwioffset0+vdwjidx0D,
 598                                          &c6_00,&c12_00);
 599
 600             /* LENNARD-JONES DISPERSION/REPULSION */
 601
 602             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 603             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 604
 605             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 606
 607             fscal            = fvdw;
 608
 609             fscal            = _mm_and_ps(fscal,cutoff_mask);
 610
 611             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_ps(fscal,dx00);
 615             ty               = _mm_mul_ps(fscal,dy00);
 616             tz               = _mm_mul_ps(fscal,dz00);
 617
 618             /* Update vectorial force */
 619             fix0             = _mm_add_ps(fix0,tx);
 620             fiy0             = _mm_add_ps(fiy0,ty);
 621             fiz0             = _mm_add_ps(fiz0,tz);
 622
 623             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 624                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 625                                                    tx,ty,tz);
 626
 627             }
 628
 629             /* Inner loop uses 30 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 635                                               f+i_coord_offset,fshift+i_shift_offset);
 636
 637         /* Increment number of inner iterations */
 638         inneriter                  += j_index_end - j_index_start;
 639
 640         /* Outer loop uses 9 flops */
 641     }
 642
 643     /* Increment number of outer iterations */
 644     outeriter        += nri;
 645
 646     /* Update outer/inner flops */
 647
 648     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*9 + inneriter*30);
 649 }