src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwLJ_GeomP1P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_VF_sse4_1_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_sse4_1_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     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     int              nvdwtype;
  77     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  78     int              *vdwtype;
  79     real             *vdwparam;
  80     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  81     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_ps(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99     krf              = _mm_set1_ps(fr->ic->k_rf);
 100     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 101     crf              = _mm_set1_ps(fr->ic->c_rf);
 102     nvdwtype         = fr->ntype;
 103     vdwparam         = fr->nbfp;
 104     vdwtype          = mdatoms->typeA;
 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         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 144         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146         /* Reset potential sums */
 147         velecsum         = _mm_setzero_ps();
 148         vvdwsum          = _mm_setzero_ps();
 149
 150         /* Start inner kernel loop */
 151         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 152         {
 153
 154             /* Get j neighbor index, and coordinate index */
 155             jnrA             = jjnr[jidx];
 156             jnrB             = jjnr[jidx+1];
 157             jnrC             = jjnr[jidx+2];
 158             jnrD             = jjnr[jidx+3];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161             j_coord_offsetC  = DIM*jnrC;
 162             j_coord_offsetD  = DIM*jnrD;
 163
 164             /* load j atom coordinates */
 165             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 166                                               x+j_coord_offsetC,x+j_coord_offsetD,
 167                                               &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx00             = _mm_sub_ps(ix0,jx0);
 171             dy00             = _mm_sub_ps(iy0,jy0);
 172             dz00             = _mm_sub_ps(iz0,jz0);
 173
 174             /* Calculate squared distance and things based on it */
 175             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 176
 177             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 178
 179             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 183                                                               charge+jnrC+0,charge+jnrD+0);
 184             vdwjidx0A        = 2*vdwtype[jnrA+0];
 185             vdwjidx0B        = 2*vdwtype[jnrB+0];
 186             vdwjidx0C        = 2*vdwtype[jnrC+0];
 187             vdwjidx0D        = 2*vdwtype[jnrD+0];
 188
 189             /**************************
 190              * CALCULATE INTERACTIONS *
 191              **************************/
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm_mul_ps(iq0,jq0);
 195             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 196                                          vdwparam+vdwioffset0+vdwjidx0B,
 197                                          vdwparam+vdwioffset0+vdwjidx0C,
 198                                          vdwparam+vdwioffset0+vdwjidx0D,
 199                                          &c6_00,&c12_00);
 200
 201             /* REACTION-FIELD ELECTROSTATICS */
 202             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 203             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 204
 205             /* LENNARD-JONES DISPERSION/REPULSION */
 206
 207             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 208             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 209             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 210             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 211             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             velecsum         = _mm_add_ps(velecsum,velec);
 215             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 216
 217             fscal            = _mm_add_ps(felec,fvdw);
 218
 219             /* Calculate temporary vectorial force */
 220             tx               = _mm_mul_ps(fscal,dx00);
 221             ty               = _mm_mul_ps(fscal,dy00);
 222             tz               = _mm_mul_ps(fscal,dz00);
 223
 224             /* Update vectorial force */
 225             fix0             = _mm_add_ps(fix0,tx);
 226             fiy0             = _mm_add_ps(fiy0,ty);
 227             fiz0             = _mm_add_ps(fiz0,tz);
 228
 229             fjptrA             = f+j_coord_offsetA;
 230             fjptrB             = f+j_coord_offsetB;
 231             fjptrC             = f+j_coord_offsetC;
 232             fjptrD             = f+j_coord_offsetD;
 233             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 234
 235             /* Inner loop uses 44 flops */
 236         }
 237
 238         if(jidx<j_index_end)
 239         {
 240
 241             /* Get j neighbor index, and coordinate index */
 242             jnrlistA         = jjnr[jidx];
 243             jnrlistB         = jjnr[jidx+1];
 244             jnrlistC         = jjnr[jidx+2];
 245             jnrlistD         = jjnr[jidx+3];
 246             /* Sign of each element will be negative for non-real atoms.
 247              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 248              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 249              */
 250             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 251             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 252             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 253             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 254             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 255             j_coord_offsetA  = DIM*jnrA;
 256             j_coord_offsetB  = DIM*jnrB;
 257             j_coord_offsetC  = DIM*jnrC;
 258             j_coord_offsetD  = DIM*jnrD;
 259
 260             /* load j atom coordinates */
 261             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 262                                               x+j_coord_offsetC,x+j_coord_offsetD,
 263                                               &jx0,&jy0,&jz0);
 264
 265             /* Calculate displacement vector */
 266             dx00             = _mm_sub_ps(ix0,jx0);
 267             dy00             = _mm_sub_ps(iy0,jy0);
 268             dz00             = _mm_sub_ps(iz0,jz0);
 269
 270             /* Calculate squared distance and things based on it */
 271             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 272
 273             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 274
 275             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 276
 277             /* Load parameters for j particles */
 278             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 279                                                               charge+jnrC+0,charge+jnrD+0);
 280             vdwjidx0A        = 2*vdwtype[jnrA+0];
 281             vdwjidx0B        = 2*vdwtype[jnrB+0];
 282             vdwjidx0C        = 2*vdwtype[jnrC+0];
 283             vdwjidx0D        = 2*vdwtype[jnrD+0];
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq00             = _mm_mul_ps(iq0,jq0);
 291             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 292                                          vdwparam+vdwioffset0+vdwjidx0B,
 293                                          vdwparam+vdwioffset0+vdwjidx0C,
 294                                          vdwparam+vdwioffset0+vdwjidx0D,
 295                                          &c6_00,&c12_00);
 296
 297             /* REACTION-FIELD ELECTROSTATICS */
 298             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 299             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 300
 301             /* LENNARD-JONES DISPERSION/REPULSION */
 302
 303             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 304             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 305             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 306             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 307             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velec            = _mm_andnot_ps(dummy_mask,velec);
 311             velecsum         = _mm_add_ps(velecsum,velec);
 312             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 313             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 314
 315             fscal            = _mm_add_ps(felec,fvdw);
 316
 317             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 318
 319             /* Calculate temporary vectorial force */
 320             tx               = _mm_mul_ps(fscal,dx00);
 321             ty               = _mm_mul_ps(fscal,dy00);
 322             tz               = _mm_mul_ps(fscal,dz00);
 323
 324             /* Update vectorial force */
 325             fix0             = _mm_add_ps(fix0,tx);
 326             fiy0             = _mm_add_ps(fiy0,ty);
 327             fiz0             = _mm_add_ps(fiz0,tz);
 328
 329             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 330             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 331             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 332             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 333             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 334
 335             /* Inner loop uses 44 flops */
 336         }
 337
 338         /* End of innermost loop */
 339
 340         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 341                                               f+i_coord_offset,fshift+i_shift_offset);
 342
 343         ggid                        = gid[iidx];
 344         /* Update potential energies */
 345         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 346         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 347
 348         /* Increment number of inner iterations */
 349         inneriter                  += j_index_end - j_index_start;
 350
 351         /* Outer loop uses 9 flops */
 352     }
 353
 354     /* Increment number of outer iterations */
 355     outeriter        += nri;
 356
 357     /* Update outer/inner flops */
 358
 359     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*44);
 360 }
 361 /*
 362  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse4_1_single
 363  * Electrostatics interaction: ReactionField
 364  * VdW interaction:            LennardJones
 365  * Geometry:                   Particle-Particle
 366  * Calculate force/pot:        Force
 367  */
 368 void
 369 nb_kernel_ElecRF_VdwLJ_GeomP1P1_F_sse4_1_single
 370                     (t_nblist * gmx_restrict                nlist,
 371                      rvec * gmx_restrict                    xx,
 372                      rvec * gmx_restrict                    ff,
 373                      t_forcerec * gmx_restrict              fr,
 374                      t_mdatoms * gmx_restrict               mdatoms,
 375                      nb_kernel_data_t * gmx_restrict        kernel_data,
 376                      t_nrnb * gmx_restrict                  nrnb)
 377 {
 378     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 379      * just 0 for non-waters.
 380      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 381      * jnr indices corresponding to data put in the four positions in the SIMD register.
 382      */
 383     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 384     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 385     int              jnrA,jnrB,jnrC,jnrD;
 386     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 387     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 388     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 389     real             rcutoff_scalar;
 390     real             *shiftvec,*fshift,*x,*f;
 391     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 392     real             scratch[4*DIM];
 393     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 394     int              vdwioffset0;
 395     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 396     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 397     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 398     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 399     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 400     real             *charge;
 401     int              nvdwtype;
 402     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 403     int              *vdwtype;
 404     real             *vdwparam;
 405     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 406     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 407     __m128           dummy_mask,cutoff_mask;
 408     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 409     __m128           one     = _mm_set1_ps(1.0);
 410     __m128           two     = _mm_set1_ps(2.0);
 411     x                = xx[0];
 412     f                = ff[0];
 413
 414     nri              = nlist->nri;
 415     iinr             = nlist->iinr;
 416     jindex           = nlist->jindex;
 417     jjnr             = nlist->jjnr;
 418     shiftidx         = nlist->shift;
 419     gid              = nlist->gid;
 420     shiftvec         = fr->shift_vec[0];
 421     fshift           = fr->fshift[0];
 422     facel            = _mm_set1_ps(fr->epsfac);
 423     charge           = mdatoms->chargeA;
 424     krf              = _mm_set1_ps(fr->ic->k_rf);
 425     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 426     crf              = _mm_set1_ps(fr->ic->c_rf);
 427     nvdwtype         = fr->ntype;
 428     vdwparam         = fr->nbfp;
 429     vdwtype          = mdatoms->typeA;
 430
 431     /* Avoid stupid compiler warnings */
 432     jnrA = jnrB = jnrC = jnrD = 0;
 433     j_coord_offsetA = 0;
 434     j_coord_offsetB = 0;
 435     j_coord_offsetC = 0;
 436     j_coord_offsetD = 0;
 437
 438     outeriter        = 0;
 439     inneriter        = 0;
 440
 441     for(iidx=0;iidx<4*DIM;iidx++)
 442     {
 443         scratch[iidx] = 0.0;
 444     }
 445
 446     /* Start outer loop over neighborlists */
 447     for(iidx=0; iidx<nri; iidx++)
 448     {
 449         /* Load shift vector for this list */
 450         i_shift_offset   = DIM*shiftidx[iidx];
 451
 452         /* Load limits for loop over neighbors */
 453         j_index_start    = jindex[iidx];
 454         j_index_end      = jindex[iidx+1];
 455
 456         /* Get outer coordinate index */
 457         inr              = iinr[iidx];
 458         i_coord_offset   = DIM*inr;
 459
 460         /* Load i particle coords and add shift vector */
 461         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 462
 463         fix0             = _mm_setzero_ps();
 464         fiy0             = _mm_setzero_ps();
 465         fiz0             = _mm_setzero_ps();
 466
 467         /* Load parameters for i particles */
 468         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 469         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 470
 471         /* Start inner kernel loop */
 472         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 473         {
 474
 475             /* Get j neighbor index, and coordinate index */
 476             jnrA             = jjnr[jidx];
 477             jnrB             = jjnr[jidx+1];
 478             jnrC             = jjnr[jidx+2];
 479             jnrD             = jjnr[jidx+3];
 480             j_coord_offsetA  = DIM*jnrA;
 481             j_coord_offsetB  = DIM*jnrB;
 482             j_coord_offsetC  = DIM*jnrC;
 483             j_coord_offsetD  = DIM*jnrD;
 484
 485             /* load j atom coordinates */
 486             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 487                                               x+j_coord_offsetC,x+j_coord_offsetD,
 488                                               &jx0,&jy0,&jz0);
 489
 490             /* Calculate displacement vector */
 491             dx00             = _mm_sub_ps(ix0,jx0);
 492             dy00             = _mm_sub_ps(iy0,jy0);
 493             dz00             = _mm_sub_ps(iz0,jz0);
 494
 495             /* Calculate squared distance and things based on it */
 496             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 497
 498             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 499
 500             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 501
 502             /* Load parameters for j particles */
 503             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 504                                                               charge+jnrC+0,charge+jnrD+0);
 505             vdwjidx0A        = 2*vdwtype[jnrA+0];
 506             vdwjidx0B        = 2*vdwtype[jnrB+0];
 507             vdwjidx0C        = 2*vdwtype[jnrC+0];
 508             vdwjidx0D        = 2*vdwtype[jnrD+0];
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq00             = _mm_mul_ps(iq0,jq0);
 516             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 517                                          vdwparam+vdwioffset0+vdwjidx0B,
 518                                          vdwparam+vdwioffset0+vdwjidx0C,
 519                                          vdwparam+vdwioffset0+vdwjidx0D,
 520                                          &c6_00,&c12_00);
 521
 522             /* REACTION-FIELD ELECTROSTATICS */
 523             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 524
 525             /* LENNARD-JONES DISPERSION/REPULSION */
 526
 527             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 528             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 529
 530             fscal            = _mm_add_ps(felec,fvdw);
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm_mul_ps(fscal,dx00);
 534             ty               = _mm_mul_ps(fscal,dy00);
 535             tz               = _mm_mul_ps(fscal,dz00);
 536
 537             /* Update vectorial force */
 538             fix0             = _mm_add_ps(fix0,tx);
 539             fiy0             = _mm_add_ps(fiy0,ty);
 540             fiz0             = _mm_add_ps(fiz0,tz);
 541
 542             fjptrA             = f+j_coord_offsetA;
 543             fjptrB             = f+j_coord_offsetB;
 544             fjptrC             = f+j_coord_offsetC;
 545             fjptrD             = f+j_coord_offsetD;
 546             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 547
 548             /* Inner loop uses 34 flops */
 549         }
 550
 551         if(jidx<j_index_end)
 552         {
 553
 554             /* Get j neighbor index, and coordinate index */
 555             jnrlistA         = jjnr[jidx];
 556             jnrlistB         = jjnr[jidx+1];
 557             jnrlistC         = jjnr[jidx+2];
 558             jnrlistD         = jjnr[jidx+3];
 559             /* Sign of each element will be negative for non-real atoms.
 560              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 561              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 562              */
 563             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 564             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 565             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 566             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 567             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 568             j_coord_offsetA  = DIM*jnrA;
 569             j_coord_offsetB  = DIM*jnrB;
 570             j_coord_offsetC  = DIM*jnrC;
 571             j_coord_offsetD  = DIM*jnrD;
 572
 573             /* load j atom coordinates */
 574             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 575                                               x+j_coord_offsetC,x+j_coord_offsetD,
 576                                               &jx0,&jy0,&jz0);
 577
 578             /* Calculate displacement vector */
 579             dx00             = _mm_sub_ps(ix0,jx0);
 580             dy00             = _mm_sub_ps(iy0,jy0);
 581             dz00             = _mm_sub_ps(iz0,jz0);
 582
 583             /* Calculate squared distance and things based on it */
 584             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 585
 586             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 587
 588             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 589
 590             /* Load parameters for j particles */
 591             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 592                                                               charge+jnrC+0,charge+jnrD+0);
 593             vdwjidx0A        = 2*vdwtype[jnrA+0];
 594             vdwjidx0B        = 2*vdwtype[jnrB+0];
 595             vdwjidx0C        = 2*vdwtype[jnrC+0];
 596             vdwjidx0D        = 2*vdwtype[jnrD+0];
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             /* Compute parameters for interactions between i and j atoms */
 603             qq00             = _mm_mul_ps(iq0,jq0);
 604             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 605                                          vdwparam+vdwioffset0+vdwjidx0B,
 606                                          vdwparam+vdwioffset0+vdwjidx0C,
 607                                          vdwparam+vdwioffset0+vdwjidx0D,
 608                                          &c6_00,&c12_00);
 609
 610             /* REACTION-FIELD ELECTROSTATICS */
 611             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 612
 613             /* LENNARD-JONES DISPERSION/REPULSION */
 614
 615             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 616             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 617
 618             fscal            = _mm_add_ps(felec,fvdw);
 619
 620             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 621
 622             /* Calculate temporary vectorial force */
 623             tx               = _mm_mul_ps(fscal,dx00);
 624             ty               = _mm_mul_ps(fscal,dy00);
 625             tz               = _mm_mul_ps(fscal,dz00);
 626
 627             /* Update vectorial force */
 628             fix0             = _mm_add_ps(fix0,tx);
 629             fiy0             = _mm_add_ps(fiy0,ty);
 630             fiz0             = _mm_add_ps(fiz0,tz);
 631
 632             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 633             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 634             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 635             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 636             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 637
 638             /* Inner loop uses 34 flops */
 639         }
 640
 641         /* End of innermost loop */
 642
 643         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 644                                               f+i_coord_offset,fshift+i_shift_offset);
 645
 646         /* Increment number of inner iterations */
 647         inneriter                  += j_index_end - j_index_start;
 648
 649         /* Outer loop uses 7 flops */
 650     }
 651
 652     /* Increment number of outer iterations */
 653     outeriter        += nri;
 654
 655     /* Update outer/inner flops */
 656
 657     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*34);
 658 }