src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecNone_VdwLJSh_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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 AVX_128, 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           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_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 203                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),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              /* Update vectorial force */
 217             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 218             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 219             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 220
 221             fjptrA             = f+j_coord_offsetA;
 222             fjptrB             = f+j_coord_offsetB;
 223             fjptrC             = f+j_coord_offsetC;
 224             fjptrD             = f+j_coord_offsetD;
 225             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 226                                                    _mm_mul_ps(dx00,fscal),
 227                                                    _mm_mul_ps(dy00,fscal),
 228                                                    _mm_mul_ps(dz00,fscal));
 229
 230             }
 231
 232             /* Inner loop uses 44 flops */
 233         }
 234
 235         if(jidx<j_index_end)
 236         {
 237
 238             /* Get j neighbor index, and coordinate index */
 239             jnrlistA         = jjnr[jidx];
 240             jnrlistB         = jjnr[jidx+1];
 241             jnrlistC         = jjnr[jidx+2];
 242             jnrlistD         = jjnr[jidx+3];
 243             /* Sign of each element will be negative for non-real atoms.
 244              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 245              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 246              */
 247             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 248             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 249             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 250             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 251             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 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_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 298                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),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              /* Update vectorial force */
 315             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 316             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 317             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 318
 319             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 320             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 321             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 322             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 323             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 324                                                    _mm_mul_ps(dx00,fscal),
 325                                                    _mm_mul_ps(dy00,fscal),
 326                                                    _mm_mul_ps(dz00,fscal));
 327
 328             }
 329
 330             /* Inner loop uses 44 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 7 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*7 + inneriter*44);
 354 }
 355 /*
 356  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSh_GeomP1P1_F_avx_128_fma_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_avx_128_fma_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 AVX_128, 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 381     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 382     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 383     real             rcutoff_scalar;
 384     real             *shiftvec,*fshift,*x,*f;
 385     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 386     real             scratch[4*DIM];
 387     __m128           fscal,rcutoff,rcutoff2,jidxall;
 388     int              vdwioffset0;
 389     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 390     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 391     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 392     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 393     int              nvdwtype;
 394     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 395     int              *vdwtype;
 396     real             *vdwparam;
 397     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 398     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 399     __m128           dummy_mask,cutoff_mask;
 400     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 401     __m128           one     = _mm_set1_ps(1.0);
 402     __m128           two     = _mm_set1_ps(2.0);
 403     x                = xx[0];
 404     f                = ff[0];
 405
 406     nri              = nlist->nri;
 407     iinr             = nlist->iinr;
 408     jindex           = nlist->jindex;
 409     jjnr             = nlist->jjnr;
 410     shiftidx         = nlist->shift;
 411     gid              = nlist->gid;
 412     shiftvec         = fr->shift_vec[0];
 413     fshift           = fr->fshift[0];
 414     nvdwtype         = fr->ntype;
 415     vdwparam         = fr->nbfp;
 416     vdwtype          = mdatoms->typeA;
 417
 418     rcutoff_scalar   = fr->rvdw;
 419     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 420     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 421
 422     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 423     rvdw             = _mm_set1_ps(fr->rvdw);
 424
 425     /* Avoid stupid compiler warnings */
 426     jnrA = jnrB = jnrC = jnrD = 0;
 427     j_coord_offsetA = 0;
 428     j_coord_offsetB = 0;
 429     j_coord_offsetC = 0;
 430     j_coord_offsetD = 0;
 431
 432     outeriter        = 0;
 433     inneriter        = 0;
 434
 435     for(iidx=0;iidx<4*DIM;iidx++)
 436     {
 437         scratch[iidx] = 0.0;
 438     }
 439
 440     /* Start outer loop over neighborlists */
 441     for(iidx=0; iidx<nri; iidx++)
 442     {
 443         /* Load shift vector for this list */
 444         i_shift_offset   = DIM*shiftidx[iidx];
 445
 446         /* Load limits for loop over neighbors */
 447         j_index_start    = jindex[iidx];
 448         j_index_end      = jindex[iidx+1];
 449
 450         /* Get outer coordinate index */
 451         inr              = iinr[iidx];
 452         i_coord_offset   = DIM*inr;
 453
 454         /* Load i particle coords and add shift vector */
 455         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 456
 457         fix0             = _mm_setzero_ps();
 458         fiy0             = _mm_setzero_ps();
 459         fiz0             = _mm_setzero_ps();
 460
 461         /* Load parameters for i particles */
 462         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 463
 464         /* Start inner kernel loop */
 465         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 466         {
 467
 468             /* Get j neighbor index, and coordinate index */
 469             jnrA             = jjnr[jidx];
 470             jnrB             = jjnr[jidx+1];
 471             jnrC             = jjnr[jidx+2];
 472             jnrD             = jjnr[jidx+3];
 473             j_coord_offsetA  = DIM*jnrA;
 474             j_coord_offsetB  = DIM*jnrB;
 475             j_coord_offsetC  = DIM*jnrC;
 476             j_coord_offsetD  = DIM*jnrD;
 477
 478             /* load j atom coordinates */
 479             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 480                                               x+j_coord_offsetC,x+j_coord_offsetD,
 481                                               &jx0,&jy0,&jz0);
 482
 483             /* Calculate displacement vector */
 484             dx00             = _mm_sub_ps(ix0,jx0);
 485             dy00             = _mm_sub_ps(iy0,jy0);
 486             dz00             = _mm_sub_ps(iz0,jz0);
 487
 488             /* Calculate squared distance and things based on it */
 489             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 490
 491             rinvsq00         = gmx_mm_inv_ps(rsq00);
 492
 493             /* Load parameters for j particles */
 494             vdwjidx0A        = 2*vdwtype[jnrA+0];
 495             vdwjidx0B        = 2*vdwtype[jnrB+0];
 496             vdwjidx0C        = 2*vdwtype[jnrC+0];
 497             vdwjidx0D        = 2*vdwtype[jnrD+0];
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             if (gmx_mm_any_lt(rsq00,rcutoff2))
 504             {
 505
 506             /* Compute parameters for interactions between i and j atoms */
 507             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 508                                          vdwparam+vdwioffset0+vdwjidx0B,
 509                                          vdwparam+vdwioffset0+vdwjidx0C,
 510                                          vdwparam+vdwioffset0+vdwjidx0D,
 511                                          &c6_00,&c12_00);
 512
 513             /* LENNARD-JONES DISPERSION/REPULSION */
 514
 515             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 516             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 517
 518             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 519
 520             fscal            = fvdw;
 521
 522             fscal            = _mm_and_ps(fscal,cutoff_mask);
 523
 524              /* Update vectorial force */
 525             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 526             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 527             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 528
 529             fjptrA             = f+j_coord_offsetA;
 530             fjptrB             = f+j_coord_offsetB;
 531             fjptrC             = f+j_coord_offsetC;
 532             fjptrD             = f+j_coord_offsetD;
 533             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 534                                                    _mm_mul_ps(dx00,fscal),
 535                                                    _mm_mul_ps(dy00,fscal),
 536                                                    _mm_mul_ps(dz00,fscal));
 537
 538             }
 539
 540             /* Inner loop uses 33 flops */
 541         }
 542
 543         if(jidx<j_index_end)
 544         {
 545
 546             /* Get j neighbor index, and coordinate index */
 547             jnrlistA         = jjnr[jidx];
 548             jnrlistB         = jjnr[jidx+1];
 549             jnrlistC         = jjnr[jidx+2];
 550             jnrlistD         = jjnr[jidx+3];
 551             /* Sign of each element will be negative for non-real atoms.
 552              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 553              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 554              */
 555             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 556             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 557             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 558             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 559             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 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_msub_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              /* Update vectorial force */
 614             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 615             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 616             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 617
 618             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 619             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 620             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 621             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 622             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 623                                                    _mm_mul_ps(dx00,fscal),
 624                                                    _mm_mul_ps(dy00,fscal),
 625                                                    _mm_mul_ps(dz00,fscal));
 626
 627             }
 628
 629             /* Inner loop uses 33 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 6 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*6 + inneriter*33);
 649 }