src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwNone_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_ElecRFCut_VdwNone_GeomP1P1_VF_sse4_1_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_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     __m128           dummy_mask,cutoff_mask;
  77     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  78     __m128           one     = _mm_set1_ps(1.0);
  79     __m128           two     = _mm_set1_ps(2.0);
  80     x                = xx[0];
  81     f                = ff[0];
  82
  83     nri              = nlist->nri;
  84     iinr             = nlist->iinr;
  85     jindex           = nlist->jindex;
  86     jjnr             = nlist->jjnr;
  87     shiftidx         = nlist->shift;
  88     gid              = nlist->gid;
  89     shiftvec         = fr->shift_vec[0];
  90     fshift           = fr->fshift[0];
  91     facel            = _mm_set1_ps(fr->epsfac);
  92     charge           = mdatoms->chargeA;
  93     krf              = _mm_set1_ps(fr->ic->k_rf);
  94     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
  95     crf              = _mm_set1_ps(fr->ic->c_rf);
  96
  97     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
  98     rcutoff_scalar   = fr->rcoulomb;
  99     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 100     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 101
 102     /* Avoid stupid compiler warnings */
 103     jnrA = jnrB = jnrC = jnrD = 0;
 104     j_coord_offsetA = 0;
 105     j_coord_offsetB = 0;
 106     j_coord_offsetC = 0;
 107     j_coord_offsetD = 0;
 108
 109     outeriter        = 0;
 110     inneriter        = 0;
 111
 112     for(iidx=0;iidx<4*DIM;iidx++)
 113     {
 114         scratch[iidx] = 0.0;
 115     }
 116
 117     /* Start outer loop over neighborlists */
 118     for(iidx=0; iidx<nri; iidx++)
 119     {
 120         /* Load shift vector for this list */
 121         i_shift_offset   = DIM*shiftidx[iidx];
 122
 123         /* Load limits for loop over neighbors */
 124         j_index_start    = jindex[iidx];
 125         j_index_end      = jindex[iidx+1];
 126
 127         /* Get outer coordinate index */
 128         inr              = iinr[iidx];
 129         i_coord_offset   = DIM*inr;
 130
 131         /* Load i particle coords and add shift vector */
 132         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 133
 134         fix0             = _mm_setzero_ps();
 135         fiy0             = _mm_setzero_ps();
 136         fiz0             = _mm_setzero_ps();
 137
 138         /* Load parameters for i particles */
 139         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 140
 141         /* Reset potential sums */
 142         velecsum         = _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             j_coord_offsetA  = DIM*jnrA;
 154             j_coord_offsetB  = DIM*jnrB;
 155             j_coord_offsetC  = DIM*jnrC;
 156             j_coord_offsetD  = DIM*jnrD;
 157
 158             /* load j atom coordinates */
 159             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 160                                               x+j_coord_offsetC,x+j_coord_offsetD,
 161                                               &jx0,&jy0,&jz0);
 162
 163             /* Calculate displacement vector */
 164             dx00             = _mm_sub_ps(ix0,jx0);
 165             dy00             = _mm_sub_ps(iy0,jy0);
 166             dz00             = _mm_sub_ps(iz0,jz0);
 167
 168             /* Calculate squared distance and things based on it */
 169             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 170
 171             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 172
 173             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 174
 175             /* Load parameters for j particles */
 176             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 177                                                               charge+jnrC+0,charge+jnrD+0);
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             if (gmx_mm_any_lt(rsq00,rcutoff2))
 184             {
 185
 186             /* Compute parameters for interactions between i and j atoms */
 187             qq00             = _mm_mul_ps(iq0,jq0);
 188
 189             /* REACTION-FIELD ELECTROSTATICS */
 190             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 191             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 192
 193             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 194
 195             /* Update potential sum for this i atom from the interaction with this j atom. */
 196             velec            = _mm_and_ps(velec,cutoff_mask);
 197             velecsum         = _mm_add_ps(velecsum,velec);
 198
 199             fscal            = felec;
 200
 201             fscal            = _mm_and_ps(fscal,cutoff_mask);
 202
 203             /* Calculate temporary vectorial force */
 204             tx               = _mm_mul_ps(fscal,dx00);
 205             ty               = _mm_mul_ps(fscal,dy00);
 206             tz               = _mm_mul_ps(fscal,dz00);
 207
 208             /* Update vectorial force */
 209             fix0             = _mm_add_ps(fix0,tx);
 210             fiy0             = _mm_add_ps(fiy0,ty);
 211             fiz0             = _mm_add_ps(fiz0,tz);
 212
 213             fjptrA             = f+j_coord_offsetA;
 214             fjptrB             = f+j_coord_offsetB;
 215             fjptrC             = f+j_coord_offsetC;
 216             fjptrD             = f+j_coord_offsetD;
 217             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 218
 219             }
 220
 221             /* Inner loop uses 36 flops */
 222         }
 223
 224         if(jidx<j_index_end)
 225         {
 226
 227             /* Get j neighbor index, and coordinate index */
 228             jnrlistA         = jjnr[jidx];
 229             jnrlistB         = jjnr[jidx+1];
 230             jnrlistC         = jjnr[jidx+2];
 231             jnrlistD         = jjnr[jidx+3];
 232             /* Sign of each element will be negative for non-real atoms.
 233              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 234              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 235              */
 236             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 237             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 238             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 239             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 240             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 241             j_coord_offsetA  = DIM*jnrA;
 242             j_coord_offsetB  = DIM*jnrB;
 243             j_coord_offsetC  = DIM*jnrC;
 244             j_coord_offsetD  = DIM*jnrD;
 245
 246             /* load j atom coordinates */
 247             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 248                                               x+j_coord_offsetC,x+j_coord_offsetD,
 249                                               &jx0,&jy0,&jz0);
 250
 251             /* Calculate displacement vector */
 252             dx00             = _mm_sub_ps(ix0,jx0);
 253             dy00             = _mm_sub_ps(iy0,jy0);
 254             dz00             = _mm_sub_ps(iz0,jz0);
 255
 256             /* Calculate squared distance and things based on it */
 257             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 258
 259             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 260
 261             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 262
 263             /* Load parameters for j particles */
 264             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 265                                                               charge+jnrC+0,charge+jnrD+0);
 266
 267             /**************************
 268              * CALCULATE INTERACTIONS *
 269              **************************/
 270
 271             if (gmx_mm_any_lt(rsq00,rcutoff2))
 272             {
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq00             = _mm_mul_ps(iq0,jq0);
 276
 277             /* REACTION-FIELD ELECTROSTATICS */
 278             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 279             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 280
 281             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velec            = _mm_and_ps(velec,cutoff_mask);
 285             velec            = _mm_andnot_ps(dummy_mask,velec);
 286             velecsum         = _mm_add_ps(velecsum,velec);
 287
 288             fscal            = felec;
 289
 290             fscal            = _mm_and_ps(fscal,cutoff_mask);
 291
 292             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm_mul_ps(fscal,dx00);
 296             ty               = _mm_mul_ps(fscal,dy00);
 297             tz               = _mm_mul_ps(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm_add_ps(fix0,tx);
 301             fiy0             = _mm_add_ps(fiy0,ty);
 302             fiz0             = _mm_add_ps(fiz0,tz);
 303
 304             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 305             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 306             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 307             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 308             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 309
 310             }
 311
 312             /* Inner loop uses 36 flops */
 313         }
 314
 315         /* End of innermost loop */
 316
 317         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 318                                               f+i_coord_offset,fshift+i_shift_offset);
 319
 320         ggid                        = gid[iidx];
 321         /* Update potential energies */
 322         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 323
 324         /* Increment number of inner iterations */
 325         inneriter                  += j_index_end - j_index_start;
 326
 327         /* Outer loop uses 8 flops */
 328     }
 329
 330     /* Increment number of outer iterations */
 331     outeriter        += nri;
 332
 333     /* Update outer/inner flops */
 334
 335     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*36);
 336 }
 337 /*
 338  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_single
 339  * Electrostatics interaction: ReactionField
 340  * VdW interaction:            None
 341  * Geometry:                   Particle-Particle
 342  * Calculate force/pot:        Force
 343  */
 344 void
 345 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_sse4_1_single
 346                     (t_nblist * gmx_restrict                nlist,
 347                      rvec * gmx_restrict                    xx,
 348                      rvec * gmx_restrict                    ff,
 349                      t_forcerec * gmx_restrict              fr,
 350                      t_mdatoms * gmx_restrict               mdatoms,
 351                      nb_kernel_data_t * gmx_restrict        kernel_data,
 352                      t_nrnb * gmx_restrict                  nrnb)
 353 {
 354     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 355      * just 0 for non-waters.
 356      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 357      * jnr indices corresponding to data put in the four positions in the SIMD register.
 358      */
 359     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 360     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 361     int              jnrA,jnrB,jnrC,jnrD;
 362     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 363     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 364     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 365     real             rcutoff_scalar;
 366     real             *shiftvec,*fshift,*x,*f;
 367     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 368     real             scratch[4*DIM];
 369     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 370     int              vdwioffset0;
 371     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 372     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 373     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 374     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 375     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 376     real             *charge;
 377     __m128           dummy_mask,cutoff_mask;
 378     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 379     __m128           one     = _mm_set1_ps(1.0);
 380     __m128           two     = _mm_set1_ps(2.0);
 381     x                = xx[0];
 382     f                = ff[0];
 383
 384     nri              = nlist->nri;
 385     iinr             = nlist->iinr;
 386     jindex           = nlist->jindex;
 387     jjnr             = nlist->jjnr;
 388     shiftidx         = nlist->shift;
 389     gid              = nlist->gid;
 390     shiftvec         = fr->shift_vec[0];
 391     fshift           = fr->fshift[0];
 392     facel            = _mm_set1_ps(fr->epsfac);
 393     charge           = mdatoms->chargeA;
 394     krf              = _mm_set1_ps(fr->ic->k_rf);
 395     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 396     crf              = _mm_set1_ps(fr->ic->c_rf);
 397
 398     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 399     rcutoff_scalar   = fr->rcoulomb;
 400     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 401     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 402
 403     /* Avoid stupid compiler warnings */
 404     jnrA = jnrB = jnrC = jnrD = 0;
 405     j_coord_offsetA = 0;
 406     j_coord_offsetB = 0;
 407     j_coord_offsetC = 0;
 408     j_coord_offsetD = 0;
 409
 410     outeriter        = 0;
 411     inneriter        = 0;
 412
 413     for(iidx=0;iidx<4*DIM;iidx++)
 414     {
 415         scratch[iidx] = 0.0;
 416     }
 417
 418     /* Start outer loop over neighborlists */
 419     for(iidx=0; iidx<nri; iidx++)
 420     {
 421         /* Load shift vector for this list */
 422         i_shift_offset   = DIM*shiftidx[iidx];
 423
 424         /* Load limits for loop over neighbors */
 425         j_index_start    = jindex[iidx];
 426         j_index_end      = jindex[iidx+1];
 427
 428         /* Get outer coordinate index */
 429         inr              = iinr[iidx];
 430         i_coord_offset   = DIM*inr;
 431
 432         /* Load i particle coords and add shift vector */
 433         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 434
 435         fix0             = _mm_setzero_ps();
 436         fiy0             = _mm_setzero_ps();
 437         fiz0             = _mm_setzero_ps();
 438
 439         /* Load parameters for i particles */
 440         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 441
 442         /* Start inner kernel loop */
 443         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 444         {
 445
 446             /* Get j neighbor index, and coordinate index */
 447             jnrA             = jjnr[jidx];
 448             jnrB             = jjnr[jidx+1];
 449             jnrC             = jjnr[jidx+2];
 450             jnrD             = jjnr[jidx+3];
 451             j_coord_offsetA  = DIM*jnrA;
 452             j_coord_offsetB  = DIM*jnrB;
 453             j_coord_offsetC  = DIM*jnrC;
 454             j_coord_offsetD  = DIM*jnrD;
 455
 456             /* load j atom coordinates */
 457             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 458                                               x+j_coord_offsetC,x+j_coord_offsetD,
 459                                               &jx0,&jy0,&jz0);
 460
 461             /* Calculate displacement vector */
 462             dx00             = _mm_sub_ps(ix0,jx0);
 463             dy00             = _mm_sub_ps(iy0,jy0);
 464             dz00             = _mm_sub_ps(iz0,jz0);
 465
 466             /* Calculate squared distance and things based on it */
 467             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 468
 469             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 470
 471             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 472
 473             /* Load parameters for j particles */
 474             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 475                                                               charge+jnrC+0,charge+jnrD+0);
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             if (gmx_mm_any_lt(rsq00,rcutoff2))
 482             {
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq00             = _mm_mul_ps(iq0,jq0);
 486
 487             /* REACTION-FIELD ELECTROSTATICS */
 488             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 489
 490             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 491
 492             fscal            = felec;
 493
 494             fscal            = _mm_and_ps(fscal,cutoff_mask);
 495
 496             /* Calculate temporary vectorial force */
 497             tx               = _mm_mul_ps(fscal,dx00);
 498             ty               = _mm_mul_ps(fscal,dy00);
 499             tz               = _mm_mul_ps(fscal,dz00);
 500
 501             /* Update vectorial force */
 502             fix0             = _mm_add_ps(fix0,tx);
 503             fiy0             = _mm_add_ps(fiy0,ty);
 504             fiz0             = _mm_add_ps(fiz0,tz);
 505
 506             fjptrA             = f+j_coord_offsetA;
 507             fjptrB             = f+j_coord_offsetB;
 508             fjptrC             = f+j_coord_offsetC;
 509             fjptrD             = f+j_coord_offsetD;
 510             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 511
 512             }
 513
 514             /* Inner loop uses 30 flops */
 515         }
 516
 517         if(jidx<j_index_end)
 518         {
 519
 520             /* Get j neighbor index, and coordinate index */
 521             jnrlistA         = jjnr[jidx];
 522             jnrlistB         = jjnr[jidx+1];
 523             jnrlistC         = jjnr[jidx+2];
 524             jnrlistD         = jjnr[jidx+3];
 525             /* Sign of each element will be negative for non-real atoms.
 526              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 527              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 528              */
 529             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 530             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 531             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 532             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 533             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 534             j_coord_offsetA  = DIM*jnrA;
 535             j_coord_offsetB  = DIM*jnrB;
 536             j_coord_offsetC  = DIM*jnrC;
 537             j_coord_offsetD  = DIM*jnrD;
 538
 539             /* load j atom coordinates */
 540             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 541                                               x+j_coord_offsetC,x+j_coord_offsetD,
 542                                               &jx0,&jy0,&jz0);
 543
 544             /* Calculate displacement vector */
 545             dx00             = _mm_sub_ps(ix0,jx0);
 546             dy00             = _mm_sub_ps(iy0,jy0);
 547             dz00             = _mm_sub_ps(iz0,jz0);
 548
 549             /* Calculate squared distance and things based on it */
 550             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 551
 552             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 553
 554             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 555
 556             /* Load parameters for j particles */
 557             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 558                                                               charge+jnrC+0,charge+jnrD+0);
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             if (gmx_mm_any_lt(rsq00,rcutoff2))
 565             {
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq00             = _mm_mul_ps(iq0,jq0);
 569
 570             /* REACTION-FIELD ELECTROSTATICS */
 571             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 572
 573             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 574
 575             fscal            = felec;
 576
 577             fscal            = _mm_and_ps(fscal,cutoff_mask);
 578
 579             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm_mul_ps(fscal,dx00);
 583             ty               = _mm_mul_ps(fscal,dy00);
 584             tz               = _mm_mul_ps(fscal,dz00);
 585
 586             /* Update vectorial force */
 587             fix0             = _mm_add_ps(fix0,tx);
 588             fiy0             = _mm_add_ps(fiy0,ty);
 589             fiz0             = _mm_add_ps(fiz0,tz);
 590
 591             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 592             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 593             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 594             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 595             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 596
 597             }
 598
 599             /* Inner loop uses 30 flops */
 600         }
 601
 602         /* End of innermost loop */
 603
 604         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 605                                               f+i_coord_offset,fshift+i_shift_offset);
 606
 607         /* Increment number of inner iterations */
 608         inneriter                  += j_index_end - j_index_start;
 609
 610         /* Outer loop uses 7 flops */
 611     }
 612
 613     /* Increment number of outer iterations */
 614     outeriter        += nri;
 615
 616     /* Update outer/inner flops */
 617
 618     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 619 }