src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwLJ_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_ElecCoul_VdwLJ_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  69     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  78     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  79     __m128           dummy_mask,cutoff_mask;
  80     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  81     __m128           one     = _mm_set1_ps(1.0);
  82     __m128           two     = _mm_set1_ps(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm_set1_ps(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96     nvdwtype         = fr->ntype;
  97     vdwparam         = fr->nbfp;
  98     vdwtype          = mdatoms->typeA;
  99
 100     /* Avoid stupid compiler warnings */
 101     jnrA = jnrB = jnrC = jnrD = 0;
 102     j_coord_offsetA = 0;
 103     j_coord_offsetB = 0;
 104     j_coord_offsetC = 0;
 105     j_coord_offsetD = 0;
 106
 107     outeriter        = 0;
 108     inneriter        = 0;
 109
 110     /* Start outer loop over neighborlists */
 111     for(iidx=0; iidx<nri; iidx++)
 112     {
 113         /* Load shift vector for this list */
 114         i_shift_offset   = DIM*shiftidx[iidx];
 115         shX              = shiftvec[i_shift_offset+XX];
 116         shY              = shiftvec[i_shift_offset+YY];
 117         shZ              = shiftvec[i_shift_offset+ZZ];
 118
 119         /* Load limits for loop over neighbors */
 120         j_index_start    = jindex[iidx];
 121         j_index_end      = jindex[iidx+1];
 122
 123         /* Get outer coordinate index */
 124         inr              = iinr[iidx];
 125         i_coord_offset   = DIM*inr;
 126
 127         /* Load i particle coords and add shift vector */
 128         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 129         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 130         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 131
 132         fix0             = _mm_setzero_ps();
 133         fiy0             = _mm_setzero_ps();
 134         fiz0             = _mm_setzero_ps();
 135
 136         /* Load parameters for i particles */
 137         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 138         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140         /* Reset potential sums */
 141         velecsum         = _mm_setzero_ps();
 142         vvdwsum          = _mm_setzero_ps();
 143
 144         /* Start inner kernel loop */
 145         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 146         {
 147
 148             /* Get j neighbor index, and coordinate index */
 149             jnrA             = jjnr[jidx];
 150             jnrB             = jjnr[jidx+1];
 151             jnrC             = jjnr[jidx+2];
 152             jnrD             = jjnr[jidx+3];
 153
 154             j_coord_offsetA  = DIM*jnrA;
 155             j_coord_offsetB  = DIM*jnrB;
 156             j_coord_offsetC  = DIM*jnrC;
 157             j_coord_offsetD  = DIM*jnrD;
 158
 159             /* load j atom coordinates */
 160             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 161                                               x+j_coord_offsetC,x+j_coord_offsetD,
 162                                               &jx0,&jy0,&jz0);
 163
 164             /* Calculate displacement vector */
 165             dx00             = _mm_sub_ps(ix0,jx0);
 166             dy00             = _mm_sub_ps(iy0,jy0);
 167             dz00             = _mm_sub_ps(iz0,jz0);
 168
 169             /* Calculate squared distance and things based on it */
 170             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 171
 172             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 173
 174             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 175
 176             /* Load parameters for j particles */
 177             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 178                                                               charge+jnrC+0,charge+jnrD+0);
 179             vdwjidx0A        = 2*vdwtype[jnrA+0];
 180             vdwjidx0B        = 2*vdwtype[jnrB+0];
 181             vdwjidx0C        = 2*vdwtype[jnrC+0];
 182             vdwjidx0D        = 2*vdwtype[jnrD+0];
 183
 184             /**************************
 185              * CALCULATE INTERACTIONS *
 186              **************************/
 187
 188             /* Compute parameters for interactions between i and j atoms */
 189             qq00             = _mm_mul_ps(iq0,jq0);
 190             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 191                                          vdwparam+vdwioffset0+vdwjidx0B,
 192                                          vdwparam+vdwioffset0+vdwjidx0C,
 193                                          vdwparam+vdwioffset0+vdwjidx0D,
 194                                          &c6_00,&c12_00);
 195
 196             /* COULOMB ELECTROSTATICS */
 197             velec            = _mm_mul_ps(qq00,rinv00);
 198             felec            = _mm_mul_ps(velec,rinvsq00);
 199
 200             /* LENNARD-JONES DISPERSION/REPULSION */
 201
 202             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 203             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 204             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 205             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 206             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 207
 208             /* Update potential sum for this i atom from the interaction with this j atom. */
 209             velecsum         = _mm_add_ps(velecsum,velec);
 210             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 211
 212             fscal            = _mm_add_ps(felec,fvdw);
 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             /* Inner loop uses 40 flops */
 229         }
 230
 231         if(jidx<j_index_end)
 232         {
 233
 234             /* Get j neighbor index, and coordinate index */
 235             jnrA             = jjnr[jidx];
 236             jnrB             = jjnr[jidx+1];
 237             jnrC             = jjnr[jidx+2];
 238             jnrD             = jjnr[jidx+3];
 239
 240             /* Sign of each element will be negative for non-real atoms.
 241              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 242              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 243              */
 244             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 245             jnrA       = (jnrA>=0) ? jnrA : 0;
 246             jnrB       = (jnrB>=0) ? jnrB : 0;
 247             jnrC       = (jnrC>=0) ? jnrC : 0;
 248             jnrD       = (jnrD>=0) ? jnrD : 0;
 249
 250             j_coord_offsetA  = DIM*jnrA;
 251             j_coord_offsetB  = DIM*jnrB;
 252             j_coord_offsetC  = DIM*jnrC;
 253             j_coord_offsetD  = DIM*jnrD;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 257                                               x+j_coord_offsetC,x+j_coord_offsetD,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_ps(ix0,jx0);
 262             dy00             = _mm_sub_ps(iy0,jy0);
 263             dz00             = _mm_sub_ps(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 267
 268             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 269
 270             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 271
 272             /* Load parameters for j particles */
 273             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 274                                                               charge+jnrC+0,charge+jnrD+0);
 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             /* Compute parameters for interactions between i and j atoms */
 285             qq00             = _mm_mul_ps(iq0,jq0);
 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             /* COULOMB ELECTROSTATICS */
 293             velec            = _mm_mul_ps(qq00,rinv00);
 294             felec            = _mm_mul_ps(velec,rinvsq00);
 295
 296             /* LENNARD-JONES DISPERSION/REPULSION */
 297
 298             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 299             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 300             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 301             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 302             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm_andnot_ps(dummy_mask,velec);
 306             velecsum         = _mm_add_ps(velecsum,velec);
 307             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 308             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 309
 310             fscal            = _mm_add_ps(felec,fvdw);
 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             /* Inner loop uses 40 flops */
 329         }
 330
 331         /* End of innermost loop */
 332
 333         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 334                                               f+i_coord_offset,fshift+i_shift_offset);
 335
 336         ggid                        = gid[iidx];
 337         /* Update potential energies */
 338         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 339         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 340
 341         /* Increment number of inner iterations */
 342         inneriter                  += j_index_end - j_index_start;
 343
 344         /* Outer loop uses 12 flops */
 345     }
 346
 347     /* Increment number of outer iterations */
 348     outeriter        += nri;
 349
 350     /* Update outer/inner flops */
 351
 352     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*40);
 353 }
 354 /*
 355  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_sse2_single
 356  * Electrostatics interaction: Coulomb
 357  * VdW interaction:            LennardJones
 358  * Geometry:                   Particle-Particle
 359  * Calculate force/pot:        Force
 360  */
 361 void
 362 nb_kernel_ElecCoul_VdwLJ_GeomP1P1_F_sse2_single
 363                     (t_nblist * gmx_restrict                nlist,
 364                      rvec * gmx_restrict                    xx,
 365                      rvec * gmx_restrict                    ff,
 366                      t_forcerec * gmx_restrict              fr,
 367                      t_mdatoms * gmx_restrict               mdatoms,
 368                      nb_kernel_data_t * gmx_restrict        kernel_data,
 369                      t_nrnb * gmx_restrict                  nrnb)
 370 {
 371     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 372      * just 0 for non-waters.
 373      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 374      * jnr indices corresponding to data put in the four positions in the SIMD register.
 375      */
 376     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 377     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 378     int              jnrA,jnrB,jnrC,jnrD;
 379     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 380     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 381     real             shX,shY,shZ,rcutoff_scalar;
 382     real             *shiftvec,*fshift,*x,*f;
 383     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 384     int              vdwioffset0;
 385     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 386     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 387     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 388     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 389     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 390     real             *charge;
 391     int              nvdwtype;
 392     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 393     int              *vdwtype;
 394     real             *vdwparam;
 395     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 396     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 397     __m128           dummy_mask,cutoff_mask;
 398     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 399     __m128           one     = _mm_set1_ps(1.0);
 400     __m128           two     = _mm_set1_ps(2.0);
 401     x                = xx[0];
 402     f                = ff[0];
 403
 404     nri              = nlist->nri;
 405     iinr             = nlist->iinr;
 406     jindex           = nlist->jindex;
 407     jjnr             = nlist->jjnr;
 408     shiftidx         = nlist->shift;
 409     gid              = nlist->gid;
 410     shiftvec         = fr->shift_vec[0];
 411     fshift           = fr->fshift[0];
 412     facel            = _mm_set1_ps(fr->epsfac);
 413     charge           = mdatoms->chargeA;
 414     nvdwtype         = fr->ntype;
 415     vdwparam         = fr->nbfp;
 416     vdwtype          = mdatoms->typeA;
 417
 418     /* Avoid stupid compiler warnings */
 419     jnrA = jnrB = jnrC = jnrD = 0;
 420     j_coord_offsetA = 0;
 421     j_coord_offsetB = 0;
 422     j_coord_offsetC = 0;
 423     j_coord_offsetD = 0;
 424
 425     outeriter        = 0;
 426     inneriter        = 0;
 427
 428     /* Start outer loop over neighborlists */
 429     for(iidx=0; iidx<nri; iidx++)
 430     {
 431         /* Load shift vector for this list */
 432         i_shift_offset   = DIM*shiftidx[iidx];
 433         shX              = shiftvec[i_shift_offset+XX];
 434         shY              = shiftvec[i_shift_offset+YY];
 435         shZ              = shiftvec[i_shift_offset+ZZ];
 436
 437         /* Load limits for loop over neighbors */
 438         j_index_start    = jindex[iidx];
 439         j_index_end      = jindex[iidx+1];
 440
 441         /* Get outer coordinate index */
 442         inr              = iinr[iidx];
 443         i_coord_offset   = DIM*inr;
 444
 445         /* Load i particle coords and add shift vector */
 446         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 447         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 448         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 449
 450         fix0             = _mm_setzero_ps();
 451         fiy0             = _mm_setzero_ps();
 452         fiz0             = _mm_setzero_ps();
 453
 454         /* Load parameters for i particles */
 455         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 456         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 457
 458         /* Start inner kernel loop */
 459         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 460         {
 461
 462             /* Get j neighbor index, and coordinate index */
 463             jnrA             = jjnr[jidx];
 464             jnrB             = jjnr[jidx+1];
 465             jnrC             = jjnr[jidx+2];
 466             jnrD             = jjnr[jidx+3];
 467
 468             j_coord_offsetA  = DIM*jnrA;
 469             j_coord_offsetB  = DIM*jnrB;
 470             j_coord_offsetC  = DIM*jnrC;
 471             j_coord_offsetD  = DIM*jnrD;
 472
 473             /* load j atom coordinates */
 474             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 475                                               x+j_coord_offsetC,x+j_coord_offsetD,
 476                                               &jx0,&jy0,&jz0);
 477
 478             /* Calculate displacement vector */
 479             dx00             = _mm_sub_ps(ix0,jx0);
 480             dy00             = _mm_sub_ps(iy0,jy0);
 481             dz00             = _mm_sub_ps(iz0,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 485
 486             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 487
 488             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 489
 490             /* Load parameters for j particles */
 491             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 492                                                               charge+jnrC+0,charge+jnrD+0);
 493             vdwjidx0A        = 2*vdwtype[jnrA+0];
 494             vdwjidx0B        = 2*vdwtype[jnrB+0];
 495             vdwjidx0C        = 2*vdwtype[jnrC+0];
 496             vdwjidx0D        = 2*vdwtype[jnrD+0];
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq00             = _mm_mul_ps(iq0,jq0);
 504             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 505                                          vdwparam+vdwioffset0+vdwjidx0B,
 506                                          vdwparam+vdwioffset0+vdwjidx0C,
 507                                          vdwparam+vdwioffset0+vdwjidx0D,
 508                                          &c6_00,&c12_00);
 509
 510             /* COULOMB ELECTROSTATICS */
 511             velec            = _mm_mul_ps(qq00,rinv00);
 512             felec            = _mm_mul_ps(velec,rinvsq00);
 513
 514             /* LENNARD-JONES DISPERSION/REPULSION */
 515
 516             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 517             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 518
 519             fscal            = _mm_add_ps(felec,fvdw);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm_mul_ps(fscal,dx00);
 523             ty               = _mm_mul_ps(fscal,dy00);
 524             tz               = _mm_mul_ps(fscal,dz00);
 525
 526             /* Update vectorial force */
 527             fix0             = _mm_add_ps(fix0,tx);
 528             fiy0             = _mm_add_ps(fiy0,ty);
 529             fiz0             = _mm_add_ps(fiz0,tz);
 530
 531             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 532                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 533                                                    tx,ty,tz);
 534
 535             /* Inner loop uses 34 flops */
 536         }
 537
 538         if(jidx<j_index_end)
 539         {
 540
 541             /* Get j neighbor index, and coordinate index */
 542             jnrA             = jjnr[jidx];
 543             jnrB             = jjnr[jidx+1];
 544             jnrC             = jjnr[jidx+2];
 545             jnrD             = jjnr[jidx+3];
 546
 547             /* Sign of each element will be negative for non-real atoms.
 548              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 549              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 550              */
 551             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 552             jnrA       = (jnrA>=0) ? jnrA : 0;
 553             jnrB       = (jnrB>=0) ? jnrB : 0;
 554             jnrC       = (jnrC>=0) ? jnrC : 0;
 555             jnrD       = (jnrD>=0) ? jnrD : 0;
 556
 557             j_coord_offsetA  = DIM*jnrA;
 558             j_coord_offsetB  = DIM*jnrB;
 559             j_coord_offsetC  = DIM*jnrC;
 560             j_coord_offsetD  = DIM*jnrD;
 561
 562             /* load j atom coordinates */
 563             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 564                                               x+j_coord_offsetC,x+j_coord_offsetD,
 565                                               &jx0,&jy0,&jz0);
 566
 567             /* Calculate displacement vector */
 568             dx00             = _mm_sub_ps(ix0,jx0);
 569             dy00             = _mm_sub_ps(iy0,jy0);
 570             dz00             = _mm_sub_ps(iz0,jz0);
 571
 572             /* Calculate squared distance and things based on it */
 573             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 574
 575             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 576
 577             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 578
 579             /* Load parameters for j particles */
 580             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 581                                                               charge+jnrC+0,charge+jnrD+0);
 582             vdwjidx0A        = 2*vdwtype[jnrA+0];
 583             vdwjidx0B        = 2*vdwtype[jnrB+0];
 584             vdwjidx0C        = 2*vdwtype[jnrC+0];
 585             vdwjidx0D        = 2*vdwtype[jnrD+0];
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq00             = _mm_mul_ps(iq0,jq0);
 593             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 594                                          vdwparam+vdwioffset0+vdwjidx0B,
 595                                          vdwparam+vdwioffset0+vdwjidx0C,
 596                                          vdwparam+vdwioffset0+vdwjidx0D,
 597                                          &c6_00,&c12_00);
 598
 599             /* COULOMB ELECTROSTATICS */
 600             velec            = _mm_mul_ps(qq00,rinv00);
 601             felec            = _mm_mul_ps(velec,rinvsq00);
 602
 603             /* LENNARD-JONES DISPERSION/REPULSION */
 604
 605             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 606             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 607
 608             fscal            = _mm_add_ps(felec,fvdw);
 609
 610             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm_mul_ps(fscal,dx00);
 614             ty               = _mm_mul_ps(fscal,dy00);
 615             tz               = _mm_mul_ps(fscal,dz00);
 616
 617             /* Update vectorial force */
 618             fix0             = _mm_add_ps(fix0,tx);
 619             fiy0             = _mm_add_ps(fiy0,ty);
 620             fiz0             = _mm_add_ps(fiz0,tz);
 621
 622             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 623                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 624                                                    tx,ty,tz);
 625
 626             /* Inner loop uses 34 flops */
 627         }
 628
 629         /* End of innermost loop */
 630
 631         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 632                                               f+i_coord_offset,fshift+i_shift_offset);
 633
 634         /* Increment number of inner iterations */
 635         inneriter                  += j_index_end - j_index_start;
 636
 637         /* Outer loop uses 10 flops */
 638     }
 639
 640     /* Increment number of outer iterations */
 641     outeriter        += nri;
 642
 643     /* Update outer/inner flops */
 644
 645     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*34);
 646 }