src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_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_ElecCSTab_VdwNone_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_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     __m128i          vfitab;
  74     __m128i          ifour       = _mm_set1_epi32(4);
  75     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  76     real             *vftab;
  77     __m128           dummy_mask,cutoff_mask;
  78     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  79     __m128           one     = _mm_set1_ps(1.0);
  80     __m128           two     = _mm_set1_ps(2.0);
  81     x                = xx[0];
  82     f                = ff[0];
  83
  84     nri              = nlist->nri;
  85     iinr             = nlist->iinr;
  86     jindex           = nlist->jindex;
  87     jjnr             = nlist->jjnr;
  88     shiftidx         = nlist->shift;
  89     gid              = nlist->gid;
  90     shiftvec         = fr->shift_vec[0];
  91     fshift           = fr->fshift[0];
  92     facel            = _mm_set1_ps(fr->epsfac);
  93     charge           = mdatoms->chargeA;
  94
  95     vftab            = kernel_data->table_elec->data;
  96     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
  97
  98     /* Avoid stupid compiler warnings */
  99     jnrA = jnrB = jnrC = jnrD = 0;
 100     j_coord_offsetA = 0;
 101     j_coord_offsetB = 0;
 102     j_coord_offsetC = 0;
 103     j_coord_offsetD = 0;
 104
 105     outeriter        = 0;
 106     inneriter        = 0;
 107
 108     /* Start outer loop over neighborlists */
 109     for(iidx=0; iidx<nri; iidx++)
 110     {
 111         /* Load shift vector for this list */
 112         i_shift_offset   = DIM*shiftidx[iidx];
 113         shX              = shiftvec[i_shift_offset+XX];
 114         shY              = shiftvec[i_shift_offset+YY];
 115         shZ              = shiftvec[i_shift_offset+ZZ];
 116
 117         /* Load limits for loop over neighbors */
 118         j_index_start    = jindex[iidx];
 119         j_index_end      = jindex[iidx+1];
 120
 121         /* Get outer coordinate index */
 122         inr              = iinr[iidx];
 123         i_coord_offset   = DIM*inr;
 124
 125         /* Load i particle coords and add shift vector */
 126         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 127         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 128         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 129
 130         fix0             = _mm_setzero_ps();
 131         fiy0             = _mm_setzero_ps();
 132         fiz0             = _mm_setzero_ps();
 133
 134         /* Load parameters for i particles */
 135         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 136
 137         /* Reset potential sums */
 138         velecsum         = _mm_setzero_ps();
 139
 140         /* Start inner kernel loop */
 141         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 142         {
 143
 144             /* Get j neighbor index, and coordinate index */
 145             jnrA             = jjnr[jidx];
 146             jnrB             = jjnr[jidx+1];
 147             jnrC             = jjnr[jidx+2];
 148             jnrD             = jjnr[jidx+3];
 149
 150             j_coord_offsetA  = DIM*jnrA;
 151             j_coord_offsetB  = DIM*jnrB;
 152             j_coord_offsetC  = DIM*jnrC;
 153             j_coord_offsetD  = DIM*jnrD;
 154
 155             /* load j atom coordinates */
 156             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 157                                               x+j_coord_offsetC,x+j_coord_offsetD,
 158                                               &jx0,&jy0,&jz0);
 159
 160             /* Calculate displacement vector */
 161             dx00             = _mm_sub_ps(ix0,jx0);
 162             dy00             = _mm_sub_ps(iy0,jy0);
 163             dz00             = _mm_sub_ps(iz0,jz0);
 164
 165             /* Calculate squared distance and things based on it */
 166             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 167
 168             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 169
 170             /* Load parameters for j particles */
 171             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 172                                                               charge+jnrC+0,charge+jnrD+0);
 173
 174             /**************************
 175              * CALCULATE INTERACTIONS *
 176              **************************/
 177
 178             r00              = _mm_mul_ps(rsq00,rinv00);
 179
 180             /* Compute parameters for interactions between i and j atoms */
 181             qq00             = _mm_mul_ps(iq0,jq0);
 182
 183             /* Calculate table index by multiplying r with table scale and truncate to integer */
 184             rt               = _mm_mul_ps(r00,vftabscale);
 185             vfitab           = _mm_cvttps_epi32(rt);
 186             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 187             vfitab           = _mm_slli_epi32(vfitab,2);
 188
 189             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 190             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 191             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 192             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 193             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 194             _MM_TRANSPOSE4_PS(Y,F,G,H);
 195             Heps             = _mm_mul_ps(vfeps,H);
 196             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 197             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 198             velec            = _mm_mul_ps(qq00,VV);
 199             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 200             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 201
 202             /* Update potential sum for this i atom from the interaction with this j atom. */
 203             velecsum         = _mm_add_ps(velecsum,velec);
 204
 205             fscal            = felec;
 206
 207             /* Calculate temporary vectorial force */
 208             tx               = _mm_mul_ps(fscal,dx00);
 209             ty               = _mm_mul_ps(fscal,dy00);
 210             tz               = _mm_mul_ps(fscal,dz00);
 211
 212             /* Update vectorial force */
 213             fix0             = _mm_add_ps(fix0,tx);
 214             fiy0             = _mm_add_ps(fiy0,ty);
 215             fiz0             = _mm_add_ps(fiz0,tz);
 216
 217             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 218                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 219                                                    tx,ty,tz);
 220
 221             /* Inner loop uses 43 flops */
 222         }
 223
 224         if(jidx<j_index_end)
 225         {
 226
 227             /* Get j neighbor index, and coordinate index */
 228             jnrA             = jjnr[jidx];
 229             jnrB             = jjnr[jidx+1];
 230             jnrC             = jjnr[jidx+2];
 231             jnrD             = jjnr[jidx+3];
 232
 233             /* Sign of each element will be negative for non-real atoms.
 234              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 235              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 236              */
 237             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 238             jnrA       = (jnrA>=0) ? jnrA : 0;
 239             jnrB       = (jnrB>=0) ? jnrB : 0;
 240             jnrC       = (jnrC>=0) ? jnrC : 0;
 241             jnrD       = (jnrD>=0) ? jnrD : 0;
 242
 243             j_coord_offsetA  = DIM*jnrA;
 244             j_coord_offsetB  = DIM*jnrB;
 245             j_coord_offsetC  = DIM*jnrC;
 246             j_coord_offsetD  = DIM*jnrD;
 247
 248             /* load j atom coordinates */
 249             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 250                                               x+j_coord_offsetC,x+j_coord_offsetD,
 251                                               &jx0,&jy0,&jz0);
 252
 253             /* Calculate displacement vector */
 254             dx00             = _mm_sub_ps(ix0,jx0);
 255             dy00             = _mm_sub_ps(iy0,jy0);
 256             dz00             = _mm_sub_ps(iz0,jz0);
 257
 258             /* Calculate squared distance and things based on it */
 259             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 260
 261             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 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             r00              = _mm_mul_ps(rsq00,rinv00);
 272             r00              = _mm_andnot_ps(dummy_mask,r00);
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq00             = _mm_mul_ps(iq0,jq0);
 276
 277             /* Calculate table index by multiplying r with table scale and truncate to integer */
 278             rt               = _mm_mul_ps(r00,vftabscale);
 279             vfitab           = _mm_cvttps_epi32(rt);
 280             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 281             vfitab           = _mm_slli_epi32(vfitab,2);
 282
 283             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 284             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 285             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 286             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 287             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 288             _MM_TRANSPOSE4_PS(Y,F,G,H);
 289             Heps             = _mm_mul_ps(vfeps,H);
 290             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 291             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 292             velec            = _mm_mul_ps(qq00,VV);
 293             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 294             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velec            = _mm_andnot_ps(dummy_mask,velec);
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm_mul_ps(fscal,dx00);
 306             ty               = _mm_mul_ps(fscal,dy00);
 307             tz               = _mm_mul_ps(fscal,dz00);
 308
 309             /* Update vectorial force */
 310             fix0             = _mm_add_ps(fix0,tx);
 311             fiy0             = _mm_add_ps(fiy0,ty);
 312             fiz0             = _mm_add_ps(fiz0,tz);
 313
 314             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 315                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 316                                                    tx,ty,tz);
 317
 318             /* Inner loop uses 44 flops */
 319         }
 320
 321         /* End of innermost loop */
 322
 323         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 324                                               f+i_coord_offset,fshift+i_shift_offset);
 325
 326         ggid                        = gid[iidx];
 327         /* Update potential energies */
 328         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 329
 330         /* Increment number of inner iterations */
 331         inneriter                  += j_index_end - j_index_start;
 332
 333         /* Outer loop uses 11 flops */
 334     }
 335
 336     /* Increment number of outer iterations */
 337     outeriter        += nri;
 338
 339     /* Update outer/inner flops */
 340
 341     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*11 + inneriter*44);
 342 }
 343 /*
 344  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_single
 345  * Electrostatics interaction: CubicSplineTable
 346  * VdW interaction:            None
 347  * Geometry:                   Particle-Particle
 348  * Calculate force/pot:        Force
 349  */
 350 void
 351 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse2_single
 352                     (t_nblist * gmx_restrict                nlist,
 353                      rvec * gmx_restrict                    xx,
 354                      rvec * gmx_restrict                    ff,
 355                      t_forcerec * gmx_restrict              fr,
 356                      t_mdatoms * gmx_restrict               mdatoms,
 357                      nb_kernel_data_t * gmx_restrict        kernel_data,
 358                      t_nrnb * gmx_restrict                  nrnb)
 359 {
 360     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 361      * just 0 for non-waters.
 362      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 363      * jnr indices corresponding to data put in the four positions in the SIMD register.
 364      */
 365     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 366     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 367     int              jnrA,jnrB,jnrC,jnrD;
 368     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 369     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 370     real             shX,shY,shZ,rcutoff_scalar;
 371     real             *shiftvec,*fshift,*x,*f;
 372     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 373     int              vdwioffset0;
 374     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 375     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 376     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 377     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 378     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 379     real             *charge;
 380     __m128i          vfitab;
 381     __m128i          ifour       = _mm_set1_epi32(4);
 382     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 383     real             *vftab;
 384     __m128           dummy_mask,cutoff_mask;
 385     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 386     __m128           one     = _mm_set1_ps(1.0);
 387     __m128           two     = _mm_set1_ps(2.0);
 388     x                = xx[0];
 389     f                = ff[0];
 390
 391     nri              = nlist->nri;
 392     iinr             = nlist->iinr;
 393     jindex           = nlist->jindex;
 394     jjnr             = nlist->jjnr;
 395     shiftidx         = nlist->shift;
 396     gid              = nlist->gid;
 397     shiftvec         = fr->shift_vec[0];
 398     fshift           = fr->fshift[0];
 399     facel            = _mm_set1_ps(fr->epsfac);
 400     charge           = mdatoms->chargeA;
 401
 402     vftab            = kernel_data->table_elec->data;
 403     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 404
 405     /* Avoid stupid compiler warnings */
 406     jnrA = jnrB = jnrC = jnrD = 0;
 407     j_coord_offsetA = 0;
 408     j_coord_offsetB = 0;
 409     j_coord_offsetC = 0;
 410     j_coord_offsetD = 0;
 411
 412     outeriter        = 0;
 413     inneriter        = 0;
 414
 415     /* Start outer loop over neighborlists */
 416     for(iidx=0; iidx<nri; iidx++)
 417     {
 418         /* Load shift vector for this list */
 419         i_shift_offset   = DIM*shiftidx[iidx];
 420         shX              = shiftvec[i_shift_offset+XX];
 421         shY              = shiftvec[i_shift_offset+YY];
 422         shZ              = shiftvec[i_shift_offset+ZZ];
 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         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 434         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 435         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 436
 437         fix0             = _mm_setzero_ps();
 438         fiy0             = _mm_setzero_ps();
 439         fiz0             = _mm_setzero_ps();
 440
 441         /* Load parameters for i particles */
 442         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 443
 444         /* Start inner kernel loop */
 445         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 446         {
 447
 448             /* Get j neighbor index, and coordinate index */
 449             jnrA             = jjnr[jidx];
 450             jnrB             = jjnr[jidx+1];
 451             jnrC             = jjnr[jidx+2];
 452             jnrD             = jjnr[jidx+3];
 453
 454             j_coord_offsetA  = DIM*jnrA;
 455             j_coord_offsetB  = DIM*jnrB;
 456             j_coord_offsetC  = DIM*jnrC;
 457             j_coord_offsetD  = DIM*jnrD;
 458
 459             /* load j atom coordinates */
 460             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 461                                               x+j_coord_offsetC,x+j_coord_offsetD,
 462                                               &jx0,&jy0,&jz0);
 463
 464             /* Calculate displacement vector */
 465             dx00             = _mm_sub_ps(ix0,jx0);
 466             dy00             = _mm_sub_ps(iy0,jy0);
 467             dz00             = _mm_sub_ps(iz0,jz0);
 468
 469             /* Calculate squared distance and things based on it */
 470             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 471
 472             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 473
 474             /* Load parameters for j particles */
 475             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 476                                                               charge+jnrC+0,charge+jnrD+0);
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             r00              = _mm_mul_ps(rsq00,rinv00);
 483
 484             /* Compute parameters for interactions between i and j atoms */
 485             qq00             = _mm_mul_ps(iq0,jq0);
 486
 487             /* Calculate table index by multiplying r with table scale and truncate to integer */
 488             rt               = _mm_mul_ps(r00,vftabscale);
 489             vfitab           = _mm_cvttps_epi32(rt);
 490             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 491             vfitab           = _mm_slli_epi32(vfitab,2);
 492
 493             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 494             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 495             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 496             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 497             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 498             _MM_TRANSPOSE4_PS(Y,F,G,H);
 499             Heps             = _mm_mul_ps(vfeps,H);
 500             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 501             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 502             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 503
 504             fscal            = felec;
 505
 506             /* Calculate temporary vectorial force */
 507             tx               = _mm_mul_ps(fscal,dx00);
 508             ty               = _mm_mul_ps(fscal,dy00);
 509             tz               = _mm_mul_ps(fscal,dz00);
 510
 511             /* Update vectorial force */
 512             fix0             = _mm_add_ps(fix0,tx);
 513             fiy0             = _mm_add_ps(fiy0,ty);
 514             fiz0             = _mm_add_ps(fiz0,tz);
 515
 516             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 517                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 518                                                    tx,ty,tz);
 519
 520             /* Inner loop uses 39 flops */
 521         }
 522
 523         if(jidx<j_index_end)
 524         {
 525
 526             /* Get j neighbor index, and coordinate index */
 527             jnrA             = jjnr[jidx];
 528             jnrB             = jjnr[jidx+1];
 529             jnrC             = jjnr[jidx+2];
 530             jnrD             = jjnr[jidx+3];
 531
 532             /* Sign of each element will be negative for non-real atoms.
 533              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 534              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 535              */
 536             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 537             jnrA       = (jnrA>=0) ? jnrA : 0;
 538             jnrB       = (jnrB>=0) ? jnrB : 0;
 539             jnrC       = (jnrC>=0) ? jnrC : 0;
 540             jnrD       = (jnrD>=0) ? jnrD : 0;
 541
 542             j_coord_offsetA  = DIM*jnrA;
 543             j_coord_offsetB  = DIM*jnrB;
 544             j_coord_offsetC  = DIM*jnrC;
 545             j_coord_offsetD  = DIM*jnrD;
 546
 547             /* load j atom coordinates */
 548             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 549                                               x+j_coord_offsetC,x+j_coord_offsetD,
 550                                               &jx0,&jy0,&jz0);
 551
 552             /* Calculate displacement vector */
 553             dx00             = _mm_sub_ps(ix0,jx0);
 554             dy00             = _mm_sub_ps(iy0,jy0);
 555             dz00             = _mm_sub_ps(iz0,jz0);
 556
 557             /* Calculate squared distance and things based on it */
 558             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 559
 560             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 561
 562             /* Load parameters for j particles */
 563             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 564                                                               charge+jnrC+0,charge+jnrD+0);
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             r00              = _mm_mul_ps(rsq00,rinv00);
 571             r00              = _mm_andnot_ps(dummy_mask,r00);
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq00             = _mm_mul_ps(iq0,jq0);
 575
 576             /* Calculate table index by multiplying r with table scale and truncate to integer */
 577             rt               = _mm_mul_ps(r00,vftabscale);
 578             vfitab           = _mm_cvttps_epi32(rt);
 579             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 580             vfitab           = _mm_slli_epi32(vfitab,2);
 581
 582             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 583             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 584             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 585             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 586             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 587             _MM_TRANSPOSE4_PS(Y,F,G,H);
 588             Heps             = _mm_mul_ps(vfeps,H);
 589             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 590             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 591             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 592
 593             fscal            = felec;
 594
 595             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 596
 597             /* Calculate temporary vectorial force */
 598             tx               = _mm_mul_ps(fscal,dx00);
 599             ty               = _mm_mul_ps(fscal,dy00);
 600             tz               = _mm_mul_ps(fscal,dz00);
 601
 602             /* Update vectorial force */
 603             fix0             = _mm_add_ps(fix0,tx);
 604             fiy0             = _mm_add_ps(fiy0,ty);
 605             fiz0             = _mm_add_ps(fiz0,tz);
 606
 607             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 608                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 609                                                    tx,ty,tz);
 610
 611             /* Inner loop uses 40 flops */
 612         }
 613
 614         /* End of innermost loop */
 615
 616         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 617                                               f+i_coord_offset,fshift+i_shift_offset);
 618
 619         /* Increment number of inner iterations */
 620         inneriter                  += j_index_end - j_index_start;
 621
 622         /* Outer loop uses 10 flops */
 623     }
 624
 625     /* Increment number of outer iterations */
 626     outeriter        += nri;
 627
 628     /* Update outer/inner flops */
 629
 630     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*10 + inneriter*40);
 631 }