src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_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_ElecRF_VdwLJ_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     krf              = _mm_set1_ps(fr->ic->k_rf);
  97     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
  98     crf              = _mm_set1_ps(fr->ic->c_rf);
  99     nvdwtype         = fr->ntype;
 100     vdwparam         = fr->nbfp;
 101     vdwtype          = mdatoms->typeA;
 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         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 141         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 142
 143         /* Reset potential sums */
 144         velecsum         = _mm_setzero_ps();
 145         vvdwsum          = _mm_setzero_ps();
 146
 147         /* Start inner kernel loop */
 148         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 149         {
 150
 151             /* Get j neighbor index, and coordinate index */
 152             jnrA             = jjnr[jidx];
 153             jnrB             = jjnr[jidx+1];
 154             jnrC             = jjnr[jidx+2];
 155             jnrD             = jjnr[jidx+3];
 156
 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             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 176
 177             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 178
 179             /* Load parameters for j particles */
 180             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 181                                                               charge+jnrC+0,charge+jnrD+0);
 182             vdwjidx0A        = 2*vdwtype[jnrA+0];
 183             vdwjidx0B        = 2*vdwtype[jnrB+0];
 184             vdwjidx0C        = 2*vdwtype[jnrC+0];
 185             vdwjidx0D        = 2*vdwtype[jnrD+0];
 186
 187             /**************************
 188              * CALCULATE INTERACTIONS *
 189              **************************/
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm_mul_ps(iq0,jq0);
 193             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 194                                          vdwparam+vdwioffset0+vdwjidx0B,
 195                                          vdwparam+vdwioffset0+vdwjidx0C,
 196                                          vdwparam+vdwioffset0+vdwjidx0D,
 197                                          &c6_00,&c12_00);
 198
 199             /* REACTION-FIELD ELECTROSTATICS */
 200             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 201             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 202
 203             /* LENNARD-JONES DISPERSION/REPULSION */
 204
 205             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 206             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 207             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 208             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 209             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 210
 211             /* Update potential sum for this i atom from the interaction with this j atom. */
 212             velecsum         = _mm_add_ps(velecsum,velec);
 213             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 214
 215             fscal            = _mm_add_ps(felec,fvdw);
 216
 217             /* Calculate temporary vectorial force */
 218             tx               = _mm_mul_ps(fscal,dx00);
 219             ty               = _mm_mul_ps(fscal,dy00);
 220             tz               = _mm_mul_ps(fscal,dz00);
 221
 222             /* Update vectorial force */
 223             fix0             = _mm_add_ps(fix0,tx);
 224             fiy0             = _mm_add_ps(fiy0,ty);
 225             fiz0             = _mm_add_ps(fiz0,tz);
 226
 227             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 228                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 229                                                    tx,ty,tz);
 230
 231             /* Inner loop uses 44 flops */
 232         }
 233
 234         if(jidx<j_index_end)
 235         {
 236
 237             /* Get j neighbor index, and coordinate index */
 238             jnrA             = jjnr[jidx];
 239             jnrB             = jjnr[jidx+1];
 240             jnrC             = jjnr[jidx+2];
 241             jnrD             = jjnr[jidx+3];
 242
 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       = (jnrA>=0) ? jnrA : 0;
 249             jnrB       = (jnrB>=0) ? jnrB : 0;
 250             jnrC       = (jnrC>=0) ? jnrC : 0;
 251             jnrD       = (jnrD>=0) ? jnrD : 0;
 252
 253             j_coord_offsetA  = DIM*jnrA;
 254             j_coord_offsetB  = DIM*jnrB;
 255             j_coord_offsetC  = DIM*jnrC;
 256             j_coord_offsetD  = DIM*jnrD;
 257
 258             /* load j atom coordinates */
 259             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 260                                               x+j_coord_offsetC,x+j_coord_offsetD,
 261                                               &jx0,&jy0,&jz0);
 262
 263             /* Calculate displacement vector */
 264             dx00             = _mm_sub_ps(ix0,jx0);
 265             dy00             = _mm_sub_ps(iy0,jy0);
 266             dz00             = _mm_sub_ps(iz0,jz0);
 267
 268             /* Calculate squared distance and things based on it */
 269             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 270
 271             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 272
 273             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 274
 275             /* Load parameters for j particles */
 276             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 277                                                               charge+jnrC+0,charge+jnrD+0);
 278             vdwjidx0A        = 2*vdwtype[jnrA+0];
 279             vdwjidx0B        = 2*vdwtype[jnrB+0];
 280             vdwjidx0C        = 2*vdwtype[jnrC+0];
 281             vdwjidx0D        = 2*vdwtype[jnrD+0];
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq00             = _mm_mul_ps(iq0,jq0);
 289             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 290                                          vdwparam+vdwioffset0+vdwjidx0B,
 291                                          vdwparam+vdwioffset0+vdwjidx0C,
 292                                          vdwparam+vdwioffset0+vdwjidx0D,
 293                                          &c6_00,&c12_00);
 294
 295             /* REACTION-FIELD ELECTROSTATICS */
 296             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 297             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 298
 299             /* LENNARD-JONES DISPERSION/REPULSION */
 300
 301             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 302             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 303             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 304             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 305             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _mm_andnot_ps(dummy_mask,velec);
 309             velecsum         = _mm_add_ps(velecsum,velec);
 310             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 311             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 312
 313             fscal            = _mm_add_ps(felec,fvdw);
 314
 315             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm_mul_ps(fscal,dx00);
 319             ty               = _mm_mul_ps(fscal,dy00);
 320             tz               = _mm_mul_ps(fscal,dz00);
 321
 322             /* Update vectorial force */
 323             fix0             = _mm_add_ps(fix0,tx);
 324             fiy0             = _mm_add_ps(fiy0,ty);
 325             fiz0             = _mm_add_ps(fiz0,tz);
 326
 327             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 328                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 329                                                    tx,ty,tz);
 330
 331             /* Inner loop uses 44 flops */
 332         }
 333
 334         /* End of innermost loop */
 335
 336         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 337                                               f+i_coord_offset,fshift+i_shift_offset);
 338
 339         ggid                        = gid[iidx];
 340         /* Update potential energies */
 341         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 342         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 343
 344         /* Increment number of inner iterations */
 345         inneriter                  += j_index_end - j_index_start;
 346
 347         /* Outer loop uses 12 flops */
 348     }
 349
 350     /* Increment number of outer iterations */
 351     outeriter        += nri;
 352
 353     /* Update outer/inner flops */
 354
 355     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*12 + inneriter*44);
 356 }
 357 /*
 358  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
 359  * Electrostatics interaction: ReactionField
 360  * VdW interaction:            LennardJones
 361  * Geometry:                   Particle-Particle
 362  * Calculate force/pot:        Force
 363  */
 364 void
 365 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse2_single
 366                     (t_nblist * gmx_restrict                nlist,
 367                      rvec * gmx_restrict                    xx,
 368                      rvec * gmx_restrict                    ff,
 369                      t_forcerec * gmx_restrict              fr,
 370                      t_mdatoms * gmx_restrict               mdatoms,
 371                      nb_kernel_data_t * gmx_restrict        kernel_data,
 372                      t_nrnb * gmx_restrict                  nrnb)
 373 {
 374     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 375      * just 0 for non-waters.
 376      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 377      * jnr indices corresponding to data put in the four positions in the SIMD register.
 378      */
 379     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 380     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 381     int              jnrA,jnrB,jnrC,jnrD;
 382     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 383     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 384     real             shX,shY,shZ,rcutoff_scalar;
 385     real             *shiftvec,*fshift,*x,*f;
 386     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 387     int              vdwioffset0;
 388     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 389     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 390     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 391     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 392     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 393     real             *charge;
 394     int              nvdwtype;
 395     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 396     int              *vdwtype;
 397     real             *vdwparam;
 398     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 399     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 400     __m128           dummy_mask,cutoff_mask;
 401     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 402     __m128           one     = _mm_set1_ps(1.0);
 403     __m128           two     = _mm_set1_ps(2.0);
 404     x                = xx[0];
 405     f                = ff[0];
 406
 407     nri              = nlist->nri;
 408     iinr             = nlist->iinr;
 409     jindex           = nlist->jindex;
 410     jjnr             = nlist->jjnr;
 411     shiftidx         = nlist->shift;
 412     gid              = nlist->gid;
 413     shiftvec         = fr->shift_vec[0];
 414     fshift           = fr->fshift[0];
 415     facel            = _mm_set1_ps(fr->epsfac);
 416     charge           = mdatoms->chargeA;
 417     krf              = _mm_set1_ps(fr->ic->k_rf);
 418     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 419     crf              = _mm_set1_ps(fr->ic->c_rf);
 420     nvdwtype         = fr->ntype;
 421     vdwparam         = fr->nbfp;
 422     vdwtype          = mdatoms->typeA;
 423
 424     /* Avoid stupid compiler warnings */
 425     jnrA = jnrB = jnrC = jnrD = 0;
 426     j_coord_offsetA = 0;
 427     j_coord_offsetB = 0;
 428     j_coord_offsetC = 0;
 429     j_coord_offsetD = 0;
 430
 431     outeriter        = 0;
 432     inneriter        = 0;
 433
 434     /* Start outer loop over neighborlists */
 435     for(iidx=0; iidx<nri; iidx++)
 436     {
 437         /* Load shift vector for this list */
 438         i_shift_offset   = DIM*shiftidx[iidx];
 439         shX              = shiftvec[i_shift_offset+XX];
 440         shY              = shiftvec[i_shift_offset+YY];
 441         shZ              = shiftvec[i_shift_offset+ZZ];
 442
 443         /* Load limits for loop over neighbors */
 444         j_index_start    = jindex[iidx];
 445         j_index_end      = jindex[iidx+1];
 446
 447         /* Get outer coordinate index */
 448         inr              = iinr[iidx];
 449         i_coord_offset   = DIM*inr;
 450
 451         /* Load i particle coords and add shift vector */
 452         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 453         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 454         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 455
 456         fix0             = _mm_setzero_ps();
 457         fiy0             = _mm_setzero_ps();
 458         fiz0             = _mm_setzero_ps();
 459
 460         /* Load parameters for i particles */
 461         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 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
 474             j_coord_offsetA  = DIM*jnrA;
 475             j_coord_offsetB  = DIM*jnrB;
 476             j_coord_offsetC  = DIM*jnrC;
 477             j_coord_offsetD  = DIM*jnrD;
 478
 479             /* load j atom coordinates */
 480             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 481                                               x+j_coord_offsetC,x+j_coord_offsetD,
 482                                               &jx0,&jy0,&jz0);
 483
 484             /* Calculate displacement vector */
 485             dx00             = _mm_sub_ps(ix0,jx0);
 486             dy00             = _mm_sub_ps(iy0,jy0);
 487             dz00             = _mm_sub_ps(iz0,jz0);
 488
 489             /* Calculate squared distance and things based on it */
 490             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 491
 492             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 493
 494             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 495
 496             /* Load parameters for j particles */
 497             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 498                                                               charge+jnrC+0,charge+jnrD+0);
 499             vdwjidx0A        = 2*vdwtype[jnrA+0];
 500             vdwjidx0B        = 2*vdwtype[jnrB+0];
 501             vdwjidx0C        = 2*vdwtype[jnrC+0];
 502             vdwjidx0D        = 2*vdwtype[jnrD+0];
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             /* Compute parameters for interactions between i and j atoms */
 509             qq00             = _mm_mul_ps(iq0,jq0);
 510             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 511                                          vdwparam+vdwioffset0+vdwjidx0B,
 512                                          vdwparam+vdwioffset0+vdwjidx0C,
 513                                          vdwparam+vdwioffset0+vdwjidx0D,
 514                                          &c6_00,&c12_00);
 515
 516             /* REACTION-FIELD ELECTROSTATICS */
 517             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 518
 519             /* LENNARD-JONES DISPERSION/REPULSION */
 520
 521             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 522             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 523
 524             fscal            = _mm_add_ps(felec,fvdw);
 525
 526             /* Calculate temporary vectorial force */
 527             tx               = _mm_mul_ps(fscal,dx00);
 528             ty               = _mm_mul_ps(fscal,dy00);
 529             tz               = _mm_mul_ps(fscal,dz00);
 530
 531             /* Update vectorial force */
 532             fix0             = _mm_add_ps(fix0,tx);
 533             fiy0             = _mm_add_ps(fiy0,ty);
 534             fiz0             = _mm_add_ps(fiz0,tz);
 535
 536             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 537                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 538                                                    tx,ty,tz);
 539
 540             /* Inner loop uses 34 flops */
 541         }
 542
 543         if(jidx<j_index_end)
 544         {
 545
 546             /* Get j neighbor index, and coordinate index */
 547             jnrA             = jjnr[jidx];
 548             jnrB             = jjnr[jidx+1];
 549             jnrC             = jjnr[jidx+2];
 550             jnrD             = jjnr[jidx+3];
 551
 552             /* Sign of each element will be negative for non-real atoms.
 553              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 554              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 555              */
 556             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 557             jnrA       = (jnrA>=0) ? jnrA : 0;
 558             jnrB       = (jnrB>=0) ? jnrB : 0;
 559             jnrC       = (jnrC>=0) ? jnrC : 0;
 560             jnrD       = (jnrD>=0) ? jnrD : 0;
 561
 562             j_coord_offsetA  = DIM*jnrA;
 563             j_coord_offsetB  = DIM*jnrB;
 564             j_coord_offsetC  = DIM*jnrC;
 565             j_coord_offsetD  = DIM*jnrD;
 566
 567             /* load j atom coordinates */
 568             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 569                                               x+j_coord_offsetC,x+j_coord_offsetD,
 570                                               &jx0,&jy0,&jz0);
 571
 572             /* Calculate displacement vector */
 573             dx00             = _mm_sub_ps(ix0,jx0);
 574             dy00             = _mm_sub_ps(iy0,jy0);
 575             dz00             = _mm_sub_ps(iz0,jz0);
 576
 577             /* Calculate squared distance and things based on it */
 578             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 579
 580             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 581
 582             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 583
 584             /* Load parameters for j particles */
 585             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 586                                                               charge+jnrC+0,charge+jnrD+0);
 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             /* Compute parameters for interactions between i and j atoms */
 597             qq00             = _mm_mul_ps(iq0,jq0);
 598             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 599                                          vdwparam+vdwioffset0+vdwjidx0B,
 600                                          vdwparam+vdwioffset0+vdwjidx0C,
 601                                          vdwparam+vdwioffset0+vdwjidx0D,
 602                                          &c6_00,&c12_00);
 603
 604             /* REACTION-FIELD ELECTROSTATICS */
 605             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 606
 607             /* LENNARD-JONES DISPERSION/REPULSION */
 608
 609             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 610             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 611
 612             fscal            = _mm_add_ps(felec,fvdw);
 613
 614             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 615
 616             /* Calculate temporary vectorial force */
 617             tx               = _mm_mul_ps(fscal,dx00);
 618             ty               = _mm_mul_ps(fscal,dy00);
 619             tz               = _mm_mul_ps(fscal,dz00);
 620
 621             /* Update vectorial force */
 622             fix0             = _mm_add_ps(fix0,tx);
 623             fiy0             = _mm_add_ps(fiy0,ty);
 624             fiz0             = _mm_add_ps(fiz0,tz);
 625
 626             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 627                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 628                                                    tx,ty,tz);
 629
 630             /* Inner loop uses 34 flops */
 631         }
 632
 633         /* End of innermost loop */
 634
 635         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 636                                               f+i_coord_offset,fshift+i_shift_offset);
 637
 638         /* Increment number of inner iterations */
 639         inneriter                  += j_index_end - j_index_start;
 640
 641         /* Outer loop uses 10 flops */
 642     }
 643
 644     /* Increment number of outer iterations */
 645     outeriter        += nri;
 646
 647     /* Update outer/inner flops */
 648
 649     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*10 + inneriter*34);
 650 }