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              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           dummy_mask,cutoff_mask;
  81     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  82     __m128           one     = _mm_set1_ps(1.0);
  83     __m128           two     = _mm_set1_ps(2.0);
  84     x                = xx[0];
  85     f                = ff[0];
  86
  87     nri              = nlist->nri;
  88     iinr             = nlist->iinr;
  89     jindex           = nlist->jindex;
  90     jjnr             = nlist->jjnr;
  91     shiftidx         = nlist->shift;
  92     gid              = nlist->gid;
  93     shiftvec         = fr->shift_vec[0];
  94     fshift           = fr->fshift[0];
  95     nvdwtype         = fr->ntype;
  96     vdwparam         = fr->nbfp;
  97     vdwtype          = mdatoms->typeA;
  98
  99     rcutoff_scalar   = fr->rvdw;
 100     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 101     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 102
 103     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 104     rvdw             = _mm_set1_ps(fr->rvdw);
 105
 106     /* Avoid stupid compiler warnings */
 107     jnrA = jnrB = jnrC = jnrD = 0;
 108     j_coord_offsetA = 0;
 109     j_coord_offsetB = 0;
 110     j_coord_offsetC = 0;
 111     j_coord_offsetD = 0;
 112
 113     outeriter        = 0;
 114     inneriter        = 0;
 115
 116     for(iidx=0;iidx<4*DIM;iidx++)
 117     {
 118         scratch[iidx] = 0.0;
 119     }
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 137
 138         fix0             = _mm_setzero_ps();
 139         fiy0             = _mm_setzero_ps();
 140         fiz0             = _mm_setzero_ps();
 141
 142         /* Load parameters for i particles */
 143         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 144
 145         /* Reset potential sums */
 146         vvdwsum          = _mm_setzero_ps();
 147
 148         /* Start inner kernel loop */
 149         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 150         {
 151
 152             /* Get j neighbor index, and coordinate index */
 153             jnrA             = jjnr[jidx];
 154             jnrB             = jjnr[jidx+1];
 155             jnrC             = jjnr[jidx+2];
 156             jnrD             = jjnr[jidx+3];
 157             j_coord_offsetA  = DIM*jnrA;
 158             j_coord_offsetB  = DIM*jnrB;
 159             j_coord_offsetC  = DIM*jnrC;
 160             j_coord_offsetD  = DIM*jnrD;
 161
 162             /* load j atom coordinates */
 163             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 164                                               x+j_coord_offsetC,x+j_coord_offsetD,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_ps(ix0,jx0);
 169             dy00             = _mm_sub_ps(iy0,jy0);
 170             dz00             = _mm_sub_ps(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 174
 175             rinvsq00         = gmx_mm_inv_ps(rsq00);
 176
 177             /* Load parameters for j particles */
 178             vdwjidx0A        = 2*vdwtype[jnrA+0];
 179             vdwjidx0B        = 2*vdwtype[jnrB+0];
 180             vdwjidx0C        = 2*vdwtype[jnrC+0];
 181             vdwjidx0D        = 2*vdwtype[jnrD+0];
 182
 183             /**************************
 184              * CALCULATE INTERACTIONS *
 185              **************************/
 186
 187             if (gmx_mm_any_lt(rsq00,rcutoff2))
 188             {
 189
 190             /* Compute parameters for interactions between i and j atoms */
 191             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 192                                          vdwparam+vdwioffset0+vdwjidx0B,
 193                                          vdwparam+vdwioffset0+vdwjidx0C,
 194                                          vdwparam+vdwioffset0+vdwjidx0D,
 195                                          &c6_00,&c12_00);
 196
 197             /* LENNARD-JONES DISPERSION/REPULSION */
 198
 199             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 200             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 201             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 202             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) ,
 203                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 204             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 205
 206             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 207
 208             /* Update potential sum for this i atom from the interaction with this j atom. */
 209             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 210             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 211
 212             fscal            = fvdw;
 213
 214             fscal            = _mm_and_ps(fscal,cutoff_mask);
 215
 216             /* Calculate temporary vectorial force */
 217             tx               = _mm_mul_ps(fscal,dx00);
 218             ty               = _mm_mul_ps(fscal,dy00);
 219             tz               = _mm_mul_ps(fscal,dz00);
 220
 221             /* Update vectorial force */
 222             fix0             = _mm_add_ps(fix0,tx);
 223             fiy0             = _mm_add_ps(fiy0,ty);
 224             fiz0             = _mm_add_ps(fiz0,tz);
 225
 226             fjptrA             = f+j_coord_offsetA;
 227             fjptrB             = f+j_coord_offsetB;
 228             fjptrC             = f+j_coord_offsetC;
 229             fjptrD             = f+j_coord_offsetD;
 230             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 231
 232             }
 233
 234             /* Inner loop uses 41 flops */
 235         }
 236
 237         if(jidx<j_index_end)
 238         {
 239
 240             /* Get j neighbor index, and coordinate index */
 241             jnrlistA         = jjnr[jidx];
 242             jnrlistB         = jjnr[jidx+1];
 243             jnrlistC         = jjnr[jidx+2];
 244             jnrlistD         = jjnr[jidx+3];
 245             /* Sign of each element will be negative for non-real atoms.
 246              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 247              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 248              */
 249             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 250             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 251             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 252             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 253             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 254             j_coord_offsetA  = DIM*jnrA;
 255             j_coord_offsetB  = DIM*jnrB;
 256             j_coord_offsetC  = DIM*jnrC;
 257             j_coord_offsetD  = DIM*jnrD;
 258
 259             /* load j atom coordinates */
 260             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 261                                               x+j_coord_offsetC,x+j_coord_offsetD,
 262                                               &jx0,&jy0,&jz0);
 263
 264             /* Calculate displacement vector */
 265             dx00             = _mm_sub_ps(ix0,jx0);
 266             dy00             = _mm_sub_ps(iy0,jy0);
 267             dz00             = _mm_sub_ps(iz0,jz0);
 268
 269             /* Calculate squared distance and things based on it */
 270             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 271
 272             rinvsq00         = gmx_mm_inv_ps(rsq00);
 273
 274             /* Load parameters for j particles */
 275             vdwjidx0A        = 2*vdwtype[jnrA+0];
 276             vdwjidx0B        = 2*vdwtype[jnrB+0];
 277             vdwjidx0C        = 2*vdwtype[jnrC+0];
 278             vdwjidx0D        = 2*vdwtype[jnrD+0];
 279
 280             /**************************
 281              * CALCULATE INTERACTIONS *
 282              **************************/
 283
 284             if (gmx_mm_any_lt(rsq00,rcutoff2))
 285             {
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 289                                          vdwparam+vdwioffset0+vdwjidx0B,
 290                                          vdwparam+vdwioffset0+vdwjidx0C,
 291                                          vdwparam+vdwioffset0+vdwjidx0D,
 292                                          &c6_00,&c12_00);
 293
 294             /* LENNARD-JONES DISPERSION/REPULSION */
 295
 296             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 297             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 298             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 299             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) ,
 300                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 301             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 302
 303             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 307             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 308             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 309
 310             fscal            = fvdw;
 311
 312             fscal            = _mm_and_ps(fscal,cutoff_mask);
 313
 314             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_ps(fscal,dx00);
 318             ty               = _mm_mul_ps(fscal,dy00);
 319             tz               = _mm_mul_ps(fscal,dz00);
 320
 321             /* Update vectorial force */
 322             fix0             = _mm_add_ps(fix0,tx);
 323             fiy0             = _mm_add_ps(fiy0,ty);
 324             fiz0             = _mm_add_ps(fiz0,tz);
 325
 326             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 327             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 328             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 329             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 330             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 331
 332             }
 333
 334             /* Inner loop uses 41 flops */
 335         }
 336
 337         /* End of innermost loop */
 338
 339         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 340                                               f+i_coord_offset,fshift+i_shift_offset);
 341
 342         ggid                        = gid[iidx];
 343         /* Update potential energies */
 344         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 345
 346         /* Increment number of inner iterations */
 347         inneriter                  += j_index_end - j_index_start;
 348
 349         /* Outer loop uses 7 flops */
 350     }
 351
 352     /* Increment number of outer iterations */
 353     outeriter        += nri;
 354
 355     /* Update outer/inner flops */
 356
 357     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*41);
 358 }
 359 /*
 360  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse2_single
 361  * Electrostatics interaction: None
 362  * VdW interaction:            LennardJones
 363  * Geometry:                   Particle-Particle
 364  * Calculate force/pot:        Force
 365  */
 366 void
 367 nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_sse2_single
 368                     (t_nblist * gmx_restrict                nlist,
 369                      rvec * gmx_restrict                    xx,
 370                      rvec * gmx_restrict                    ff,
 371                      t_forcerec * gmx_restrict              fr,
 372                      t_mdatoms * gmx_restrict               mdatoms,
 373                      nb_kernel_data_t * gmx_restrict        kernel_data,
 374                      t_nrnb * gmx_restrict                  nrnb)
 375 {
 376     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 377      * just 0 for non-waters.
 378      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 379      * jnr indices corresponding to data put in the four positions in the SIMD register.
 380      */
 381     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 382     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 383     int              jnrA,jnrB,jnrC,jnrD;
 384     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 385     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 386     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 387     real             rcutoff_scalar;
 388     real             *shiftvec,*fshift,*x,*f;
 389     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 390     real             scratch[4*DIM];
 391     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 392     int              vdwioffset0;
 393     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 394     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 395     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 396     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 397     int              nvdwtype;
 398     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 399     int              *vdwtype;
 400     real             *vdwparam;
 401     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 402     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 403     __m128           dummy_mask,cutoff_mask;
 404     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 405     __m128           one     = _mm_set1_ps(1.0);
 406     __m128           two     = _mm_set1_ps(2.0);
 407     x                = xx[0];
 408     f                = ff[0];
 409
 410     nri              = nlist->nri;
 411     iinr             = nlist->iinr;
 412     jindex           = nlist->jindex;
 413     jjnr             = nlist->jjnr;
 414     shiftidx         = nlist->shift;
 415     gid              = nlist->gid;
 416     shiftvec         = fr->shift_vec[0];
 417     fshift           = fr->fshift[0];
 418     nvdwtype         = fr->ntype;
 419     vdwparam         = fr->nbfp;
 420     vdwtype          = mdatoms->typeA;
 421
 422     rcutoff_scalar   = fr->rvdw;
 423     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 424     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 425
 426     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 427     rvdw             = _mm_set1_ps(fr->rvdw);
 428
 429     /* Avoid stupid compiler warnings */
 430     jnrA = jnrB = jnrC = jnrD = 0;
 431     j_coord_offsetA = 0;
 432     j_coord_offsetB = 0;
 433     j_coord_offsetC = 0;
 434     j_coord_offsetD = 0;
 435
 436     outeriter        = 0;
 437     inneriter        = 0;
 438
 439     for(iidx=0;iidx<4*DIM;iidx++)
 440     {
 441         scratch[iidx] = 0.0;
 442     }
 443
 444     /* Start outer loop over neighborlists */
 445     for(iidx=0; iidx<nri; iidx++)
 446     {
 447         /* Load shift vector for this list */
 448         i_shift_offset   = DIM*shiftidx[iidx];
 449
 450         /* Load limits for loop over neighbors */
 451         j_index_start    = jindex[iidx];
 452         j_index_end      = jindex[iidx+1];
 453
 454         /* Get outer coordinate index */
 455         inr              = iinr[iidx];
 456         i_coord_offset   = DIM*inr;
 457
 458         /* Load i particle coords and add shift vector */
 459         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 460
 461         fix0             = _mm_setzero_ps();
 462         fiy0             = _mm_setzero_ps();
 463         fiz0             = _mm_setzero_ps();
 464
 465         /* Load parameters for i particles */
 466         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 467
 468         /* Start inner kernel loop */
 469         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 470         {
 471
 472             /* Get j neighbor index, and coordinate index */
 473             jnrA             = jjnr[jidx];
 474             jnrB             = jjnr[jidx+1];
 475             jnrC             = jjnr[jidx+2];
 476             jnrD             = jjnr[jidx+3];
 477             j_coord_offsetA  = DIM*jnrA;
 478             j_coord_offsetB  = DIM*jnrB;
 479             j_coord_offsetC  = DIM*jnrC;
 480             j_coord_offsetD  = DIM*jnrD;
 481
 482             /* load j atom coordinates */
 483             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 484                                               x+j_coord_offsetC,x+j_coord_offsetD,
 485                                               &jx0,&jy0,&jz0);
 486
 487             /* Calculate displacement vector */
 488             dx00             = _mm_sub_ps(ix0,jx0);
 489             dy00             = _mm_sub_ps(iy0,jy0);
 490             dz00             = _mm_sub_ps(iz0,jz0);
 491
 492             /* Calculate squared distance and things based on it */
 493             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 494
 495             rinvsq00         = gmx_mm_inv_ps(rsq00);
 496
 497             /* Load parameters for j particles */
 498             vdwjidx0A        = 2*vdwtype[jnrA+0];
 499             vdwjidx0B        = 2*vdwtype[jnrB+0];
 500             vdwjidx0C        = 2*vdwtype[jnrC+0];
 501             vdwjidx0D        = 2*vdwtype[jnrD+0];
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             if (gmx_mm_any_lt(rsq00,rcutoff2))
 508             {
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 512                                          vdwparam+vdwioffset0+vdwjidx0B,
 513                                          vdwparam+vdwioffset0+vdwjidx0C,
 514                                          vdwparam+vdwioffset0+vdwjidx0D,
 515                                          &c6_00,&c12_00);
 516
 517             /* LENNARD-JONES DISPERSION/REPULSION */
 518
 519             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 520             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 521
 522             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 523
 524             fscal            = fvdw;
 525
 526             fscal            = _mm_and_ps(fscal,cutoff_mask);
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_ps(fscal,dx00);
 530             ty               = _mm_mul_ps(fscal,dy00);
 531             tz               = _mm_mul_ps(fscal,dz00);
 532
 533             /* Update vectorial force */
 534             fix0             = _mm_add_ps(fix0,tx);
 535             fiy0             = _mm_add_ps(fiy0,ty);
 536             fiz0             = _mm_add_ps(fiz0,tz);
 537
 538             fjptrA             = f+j_coord_offsetA;
 539             fjptrB             = f+j_coord_offsetB;
 540             fjptrC             = f+j_coord_offsetC;
 541             fjptrD             = f+j_coord_offsetD;
 542             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 543
 544             }
 545
 546             /* Inner loop uses 30 flops */
 547         }
 548
 549         if(jidx<j_index_end)
 550         {
 551
 552             /* Get j neighbor index, and coordinate index */
 553             jnrlistA         = jjnr[jidx];
 554             jnrlistB         = jjnr[jidx+1];
 555             jnrlistC         = jjnr[jidx+2];
 556             jnrlistD         = jjnr[jidx+3];
 557             /* Sign of each element will be negative for non-real atoms.
 558              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 559              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 560              */
 561             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 562             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 563             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 564             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 565             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 566             j_coord_offsetA  = DIM*jnrA;
 567             j_coord_offsetB  = DIM*jnrB;
 568             j_coord_offsetC  = DIM*jnrC;
 569             j_coord_offsetD  = DIM*jnrD;
 570
 571             /* load j atom coordinates */
 572             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 573                                               x+j_coord_offsetC,x+j_coord_offsetD,
 574                                               &jx0,&jy0,&jz0);
 575
 576             /* Calculate displacement vector */
 577             dx00             = _mm_sub_ps(ix0,jx0);
 578             dy00             = _mm_sub_ps(iy0,jy0);
 579             dz00             = _mm_sub_ps(iz0,jz0);
 580
 581             /* Calculate squared distance and things based on it */
 582             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 583
 584             rinvsq00         = gmx_mm_inv_ps(rsq00);
 585
 586             /* Load parameters for j particles */
 587             vdwjidx0A        = 2*vdwtype[jnrA+0];
 588             vdwjidx0B        = 2*vdwtype[jnrB+0];
 589             vdwjidx0C        = 2*vdwtype[jnrC+0];
 590             vdwjidx0D        = 2*vdwtype[jnrD+0];
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             if (gmx_mm_any_lt(rsq00,rcutoff2))
 597             {
 598
 599             /* Compute parameters for interactions between i and j atoms */
 600             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 601                                          vdwparam+vdwioffset0+vdwjidx0B,
 602                                          vdwparam+vdwioffset0+vdwjidx0C,
 603                                          vdwparam+vdwioffset0+vdwjidx0D,
 604                                          &c6_00,&c12_00);
 605
 606             /* LENNARD-JONES DISPERSION/REPULSION */
 607
 608             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 609             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 610
 611             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 612
 613             fscal            = fvdw;
 614
 615             fscal            = _mm_and_ps(fscal,cutoff_mask);
 616
 617             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_ps(fscal,dx00);
 621             ty               = _mm_mul_ps(fscal,dy00);
 622             tz               = _mm_mul_ps(fscal,dz00);
 623
 624             /* Update vectorial force */
 625             fix0             = _mm_add_ps(fix0,tx);
 626             fiy0             = _mm_add_ps(fiy0,ty);
 627             fiz0             = _mm_add_ps(fiz0,tz);
 628
 629             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 630             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 631             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 632             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 633             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 634
 635             }
 636
 637             /* Inner loop uses 30 flops */
 638         }
 639
 640         /* End of innermost loop */
 641
 642         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 643                                               f+i_coord_offset,fshift+i_shift_offset);
 644
 645         /* Increment number of inner iterations */
 646         inneriter                  += j_index_end - j_index_start;
 647
 648         /* Outer loop uses 6 flops */
 649     }
 650
 651     /* Increment number of outer iterations */
 652     outeriter        += nri;
 653
 654     /* Update outer/inner flops */
 655
 656     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 657 }