src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwNone_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_ElecRFCut_VdwNone_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwNone_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     __m128           dummy_mask,cutoff_mask;
  74     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  75     __m128           one     = _mm_set1_ps(1.0);
  76     __m128           two     = _mm_set1_ps(2.0);
  77     x                = xx[0];
  78     f                = ff[0];
  79
  80     nri              = nlist->nri;
  81     iinr             = nlist->iinr;
  82     jindex           = nlist->jindex;
  83     jjnr             = nlist->jjnr;
  84     shiftidx         = nlist->shift;
  85     gid              = nlist->gid;
  86     shiftvec         = fr->shift_vec[0];
  87     fshift           = fr->fshift[0];
  88     facel            = _mm_set1_ps(fr->epsfac);
  89     charge           = mdatoms->chargeA;
  90     krf              = _mm_set1_ps(fr->ic->k_rf);
  91     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
  92     crf              = _mm_set1_ps(fr->ic->c_rf);
  93
  94     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
  95     rcutoff_scalar   = fr->rcoulomb;
  96     rcutoff          = _mm_set1_ps(rcutoff_scalar);
  97     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
  98
  99     /* Avoid stupid compiler warnings */
 100     jnrA = jnrB = jnrC = jnrD = 0;
 101     j_coord_offsetA = 0;
 102     j_coord_offsetB = 0;
 103     j_coord_offsetC = 0;
 104     j_coord_offsetD = 0;
 105
 106     outeriter        = 0;
 107     inneriter        = 0;
 108
 109     /* Start outer loop over neighborlists */
 110     for(iidx=0; iidx<nri; iidx++)
 111     {
 112         /* Load shift vector for this list */
 113         i_shift_offset   = DIM*shiftidx[iidx];
 114         shX              = shiftvec[i_shift_offset+XX];
 115         shY              = shiftvec[i_shift_offset+YY];
 116         shZ              = shiftvec[i_shift_offset+ZZ];
 117
 118         /* Load limits for loop over neighbors */
 119         j_index_start    = jindex[iidx];
 120         j_index_end      = jindex[iidx+1];
 121
 122         /* Get outer coordinate index */
 123         inr              = iinr[iidx];
 124         i_coord_offset   = DIM*inr;
 125
 126         /* Load i particle coords and add shift vector */
 127         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 128         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 129         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 130
 131         fix0             = _mm_setzero_ps();
 132         fiy0             = _mm_setzero_ps();
 133         fiz0             = _mm_setzero_ps();
 134
 135         /* Load parameters for i particles */
 136         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 137
 138         /* Reset potential sums */
 139         velecsum         = _mm_setzero_ps();
 140
 141         /* Start inner kernel loop */
 142         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 143         {
 144
 145             /* Get j neighbor index, and coordinate index */
 146             jnrA             = jjnr[jidx];
 147             jnrB             = jjnr[jidx+1];
 148             jnrC             = jjnr[jidx+2];
 149             jnrD             = jjnr[jidx+3];
 150
 151             j_coord_offsetA  = DIM*jnrA;
 152             j_coord_offsetB  = DIM*jnrB;
 153             j_coord_offsetC  = DIM*jnrC;
 154             j_coord_offsetD  = DIM*jnrD;
 155
 156             /* load j atom coordinates */
 157             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 158                                               x+j_coord_offsetC,x+j_coord_offsetD,
 159                                               &jx0,&jy0,&jz0);
 160
 161             /* Calculate displacement vector */
 162             dx00             = _mm_sub_ps(ix0,jx0);
 163             dy00             = _mm_sub_ps(iy0,jy0);
 164             dz00             = _mm_sub_ps(iz0,jz0);
 165
 166             /* Calculate squared distance and things based on it */
 167             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 168
 169             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 170
 171             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 172
 173             /* Load parameters for j particles */
 174             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 175                                                               charge+jnrC+0,charge+jnrD+0);
 176
 177             /**************************
 178              * CALCULATE INTERACTIONS *
 179              **************************/
 180
 181             if (gmx_mm_any_lt(rsq00,rcutoff2))
 182             {
 183
 184             /* Compute parameters for interactions between i and j atoms */
 185             qq00             = _mm_mul_ps(iq0,jq0);
 186
 187             /* REACTION-FIELD ELECTROSTATICS */
 188             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 189             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 190
 191             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 192
 193             /* Update potential sum for this i atom from the interaction with this j atom. */
 194             velec            = _mm_and_ps(velec,cutoff_mask);
 195             velecsum         = _mm_add_ps(velecsum,velec);
 196
 197             fscal            = felec;
 198
 199             fscal            = _mm_and_ps(fscal,cutoff_mask);
 200
 201             /* Calculate temporary vectorial force */
 202             tx               = _mm_mul_ps(fscal,dx00);
 203             ty               = _mm_mul_ps(fscal,dy00);
 204             tz               = _mm_mul_ps(fscal,dz00);
 205
 206             /* Update vectorial force */
 207             fix0             = _mm_add_ps(fix0,tx);
 208             fiy0             = _mm_add_ps(fiy0,ty);
 209             fiz0             = _mm_add_ps(fiz0,tz);
 210
 211             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 212                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 213                                                    tx,ty,tz);
 214
 215             }
 216
 217             /* Inner loop uses 36 flops */
 218         }
 219
 220         if(jidx<j_index_end)
 221         {
 222
 223             /* Get j neighbor index, and coordinate index */
 224             jnrA             = jjnr[jidx];
 225             jnrB             = jjnr[jidx+1];
 226             jnrC             = jjnr[jidx+2];
 227             jnrD             = jjnr[jidx+3];
 228
 229             /* Sign of each element will be negative for non-real atoms.
 230              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 231              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 232              */
 233             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 234             jnrA       = (jnrA>=0) ? jnrA : 0;
 235             jnrB       = (jnrB>=0) ? jnrB : 0;
 236             jnrC       = (jnrC>=0) ? jnrC : 0;
 237             jnrD       = (jnrD>=0) ? jnrD : 0;
 238
 239             j_coord_offsetA  = DIM*jnrA;
 240             j_coord_offsetB  = DIM*jnrB;
 241             j_coord_offsetC  = DIM*jnrC;
 242             j_coord_offsetD  = DIM*jnrD;
 243
 244             /* load j atom coordinates */
 245             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 246                                               x+j_coord_offsetC,x+j_coord_offsetD,
 247                                               &jx0,&jy0,&jz0);
 248
 249             /* Calculate displacement vector */
 250             dx00             = _mm_sub_ps(ix0,jx0);
 251             dy00             = _mm_sub_ps(iy0,jy0);
 252             dz00             = _mm_sub_ps(iz0,jz0);
 253
 254             /* Calculate squared distance and things based on it */
 255             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 256
 257             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 258
 259             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 260
 261             /* Load parameters for j particles */
 262             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 263                                                               charge+jnrC+0,charge+jnrD+0);
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             if (gmx_mm_any_lt(rsq00,rcutoff2))
 270             {
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _mm_mul_ps(iq0,jq0);
 274
 275             /* REACTION-FIELD ELECTROSTATICS */
 276             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 277             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 278
 279             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velec            = _mm_and_ps(velec,cutoff_mask);
 283             velec            = _mm_andnot_ps(dummy_mask,velec);
 284             velecsum         = _mm_add_ps(velecsum,velec);
 285
 286             fscal            = felec;
 287
 288             fscal            = _mm_and_ps(fscal,cutoff_mask);
 289
 290             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 291
 292             /* Calculate temporary vectorial force */
 293             tx               = _mm_mul_ps(fscal,dx00);
 294             ty               = _mm_mul_ps(fscal,dy00);
 295             tz               = _mm_mul_ps(fscal,dz00);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm_add_ps(fix0,tx);
 299             fiy0             = _mm_add_ps(fiy0,ty);
 300             fiz0             = _mm_add_ps(fiz0,tz);
 301
 302             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 303                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 304                                                    tx,ty,tz);
 305
 306             }
 307
 308             /* Inner loop uses 36 flops */
 309         }
 310
 311         /* End of innermost loop */
 312
 313         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 314                                               f+i_coord_offset,fshift+i_shift_offset);
 315
 316         ggid                        = gid[iidx];
 317         /* Update potential energies */
 318         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 319
 320         /* Increment number of inner iterations */
 321         inneriter                  += j_index_end - j_index_start;
 322
 323         /* Outer loop uses 11 flops */
 324     }
 325
 326     /* Increment number of outer iterations */
 327     outeriter        += nri;
 328
 329     /* Update outer/inner flops */
 330
 331     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*11 + inneriter*36);
 332 }
 333 /*
 334  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse2_single
 335  * Electrostatics interaction: ReactionField
 336  * VdW interaction:            None
 337  * Geometry:                   Particle-Particle
 338  * Calculate force/pot:        Force
 339  */
 340 void
 341 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse2_single
 342                     (t_nblist * gmx_restrict                nlist,
 343                      rvec * gmx_restrict                    xx,
 344                      rvec * gmx_restrict                    ff,
 345                      t_forcerec * gmx_restrict              fr,
 346                      t_mdatoms * gmx_restrict               mdatoms,
 347                      nb_kernel_data_t * gmx_restrict        kernel_data,
 348                      t_nrnb * gmx_restrict                  nrnb)
 349 {
 350     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 351      * just 0 for non-waters.
 352      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 353      * jnr indices corresponding to data put in the four positions in the SIMD register.
 354      */
 355     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 356     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 357     int              jnrA,jnrB,jnrC,jnrD;
 358     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 359     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 360     real             shX,shY,shZ,rcutoff_scalar;
 361     real             *shiftvec,*fshift,*x,*f;
 362     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 363     int              vdwioffset0;
 364     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 365     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 366     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 367     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 368     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 369     real             *charge;
 370     __m128           dummy_mask,cutoff_mask;
 371     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 372     __m128           one     = _mm_set1_ps(1.0);
 373     __m128           two     = _mm_set1_ps(2.0);
 374     x                = xx[0];
 375     f                = ff[0];
 376
 377     nri              = nlist->nri;
 378     iinr             = nlist->iinr;
 379     jindex           = nlist->jindex;
 380     jjnr             = nlist->jjnr;
 381     shiftidx         = nlist->shift;
 382     gid              = nlist->gid;
 383     shiftvec         = fr->shift_vec[0];
 384     fshift           = fr->fshift[0];
 385     facel            = _mm_set1_ps(fr->epsfac);
 386     charge           = mdatoms->chargeA;
 387     krf              = _mm_set1_ps(fr->ic->k_rf);
 388     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 389     crf              = _mm_set1_ps(fr->ic->c_rf);
 390
 391     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 392     rcutoff_scalar   = fr->rcoulomb;
 393     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 394     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 395
 396     /* Avoid stupid compiler warnings */
 397     jnrA = jnrB = jnrC = jnrD = 0;
 398     j_coord_offsetA = 0;
 399     j_coord_offsetB = 0;
 400     j_coord_offsetC = 0;
 401     j_coord_offsetD = 0;
 402
 403     outeriter        = 0;
 404     inneriter        = 0;
 405
 406     /* Start outer loop over neighborlists */
 407     for(iidx=0; iidx<nri; iidx++)
 408     {
 409         /* Load shift vector for this list */
 410         i_shift_offset   = DIM*shiftidx[iidx];
 411         shX              = shiftvec[i_shift_offset+XX];
 412         shY              = shiftvec[i_shift_offset+YY];
 413         shZ              = shiftvec[i_shift_offset+ZZ];
 414
 415         /* Load limits for loop over neighbors */
 416         j_index_start    = jindex[iidx];
 417         j_index_end      = jindex[iidx+1];
 418
 419         /* Get outer coordinate index */
 420         inr              = iinr[iidx];
 421         i_coord_offset   = DIM*inr;
 422
 423         /* Load i particle coords and add shift vector */
 424         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 425         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 426         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 427
 428         fix0             = _mm_setzero_ps();
 429         fiy0             = _mm_setzero_ps();
 430         fiz0             = _mm_setzero_ps();
 431
 432         /* Load parameters for i particles */
 433         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 434
 435         /* Start inner kernel loop */
 436         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 437         {
 438
 439             /* Get j neighbor index, and coordinate index */
 440             jnrA             = jjnr[jidx];
 441             jnrB             = jjnr[jidx+1];
 442             jnrC             = jjnr[jidx+2];
 443             jnrD             = jjnr[jidx+3];
 444
 445             j_coord_offsetA  = DIM*jnrA;
 446             j_coord_offsetB  = DIM*jnrB;
 447             j_coord_offsetC  = DIM*jnrC;
 448             j_coord_offsetD  = DIM*jnrD;
 449
 450             /* load j atom coordinates */
 451             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 452                                               x+j_coord_offsetC,x+j_coord_offsetD,
 453                                               &jx0,&jy0,&jz0);
 454
 455             /* Calculate displacement vector */
 456             dx00             = _mm_sub_ps(ix0,jx0);
 457             dy00             = _mm_sub_ps(iy0,jy0);
 458             dz00             = _mm_sub_ps(iz0,jz0);
 459
 460             /* Calculate squared distance and things based on it */
 461             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 462
 463             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 464
 465             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 466
 467             /* Load parameters for j particles */
 468             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 469                                                               charge+jnrC+0,charge+jnrD+0);
 470
 471             /**************************
 472              * CALCULATE INTERACTIONS *
 473              **************************/
 474
 475             if (gmx_mm_any_lt(rsq00,rcutoff2))
 476             {
 477
 478             /* Compute parameters for interactions between i and j atoms */
 479             qq00             = _mm_mul_ps(iq0,jq0);
 480
 481             /* REACTION-FIELD ELECTROSTATICS */
 482             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 483
 484             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 485
 486             fscal            = felec;
 487
 488             fscal            = _mm_and_ps(fscal,cutoff_mask);
 489
 490             /* Calculate temporary vectorial force */
 491             tx               = _mm_mul_ps(fscal,dx00);
 492             ty               = _mm_mul_ps(fscal,dy00);
 493             tz               = _mm_mul_ps(fscal,dz00);
 494
 495             /* Update vectorial force */
 496             fix0             = _mm_add_ps(fix0,tx);
 497             fiy0             = _mm_add_ps(fiy0,ty);
 498             fiz0             = _mm_add_ps(fiz0,tz);
 499
 500             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 501                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 502                                                    tx,ty,tz);
 503
 504             }
 505
 506             /* Inner loop uses 30 flops */
 507         }
 508
 509         if(jidx<j_index_end)
 510         {
 511
 512             /* Get j neighbor index, and coordinate index */
 513             jnrA             = jjnr[jidx];
 514             jnrB             = jjnr[jidx+1];
 515             jnrC             = jjnr[jidx+2];
 516             jnrD             = jjnr[jidx+3];
 517
 518             /* Sign of each element will be negative for non-real atoms.
 519              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 520              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 521              */
 522             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 523             jnrA       = (jnrA>=0) ? jnrA : 0;
 524             jnrB       = (jnrB>=0) ? jnrB : 0;
 525             jnrC       = (jnrC>=0) ? jnrC : 0;
 526             jnrD       = (jnrD>=0) ? jnrD : 0;
 527
 528             j_coord_offsetA  = DIM*jnrA;
 529             j_coord_offsetB  = DIM*jnrB;
 530             j_coord_offsetC  = DIM*jnrC;
 531             j_coord_offsetD  = DIM*jnrD;
 532
 533             /* load j atom coordinates */
 534             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 535                                               x+j_coord_offsetC,x+j_coord_offsetD,
 536                                               &jx0,&jy0,&jz0);
 537
 538             /* Calculate displacement vector */
 539             dx00             = _mm_sub_ps(ix0,jx0);
 540             dy00             = _mm_sub_ps(iy0,jy0);
 541             dz00             = _mm_sub_ps(iz0,jz0);
 542
 543             /* Calculate squared distance and things based on it */
 544             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 545
 546             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 547
 548             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 549
 550             /* Load parameters for j particles */
 551             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 552                                                               charge+jnrC+0,charge+jnrD+0);
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm_any_lt(rsq00,rcutoff2))
 559             {
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq00             = _mm_mul_ps(iq0,jq0);
 563
 564             /* REACTION-FIELD ELECTROSTATICS */
 565             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 566
 567             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm_and_ps(fscal,cutoff_mask);
 572
 573             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_ps(fscal,dx00);
 577             ty               = _mm_mul_ps(fscal,dy00);
 578             tz               = _mm_mul_ps(fscal,dz00);
 579
 580             /* Update vectorial force */
 581             fix0             = _mm_add_ps(fix0,tx);
 582             fiy0             = _mm_add_ps(fiy0,ty);
 583             fiz0             = _mm_add_ps(fiz0,tz);
 584
 585             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 586                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 587                                                    tx,ty,tz);
 588
 589             }
 590
 591             /* Inner loop uses 30 flops */
 592         }
 593
 594         /* End of innermost loop */
 595
 596         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 597                                               f+i_coord_offset,fshift+i_shift_offset);
 598
 599         /* Increment number of inner iterations */
 600         inneriter                  += j_index_end - j_index_start;
 601
 602         /* Outer loop uses 10 flops */
 603     }
 604
 605     /* Increment number of outer iterations */
 606     outeriter        += nri;
 607
 608     /* Update outer/inner flops */
 609
 610     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*10 + inneriter*30);
 611 }