src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_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_ElecCSTab_VdwNone_GeomP1P1_VF_sse4_1_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_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     __m128i          vfitab;
  77     __m128i          ifour       = _mm_set1_epi32(4);
  78     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  79     real             *vftab;
  80     __m128           dummy_mask,cutoff_mask;
  81     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  82     __m128           one     = _mm_set1_ps(1.0);
  83     __m128           two     = _mm_set1_ps(2.0);
  84     x                = xx[0];
  85     f                = ff[0];
  86
  87     nri              = nlist->nri;
  88     iinr             = nlist->iinr;
  89     jindex           = nlist->jindex;
  90     jjnr             = nlist->jjnr;
  91     shiftidx         = nlist->shift;
  92     gid              = nlist->gid;
  93     shiftvec         = fr->shift_vec[0];
  94     fshift           = fr->fshift[0];
  95     facel            = _mm_set1_ps(fr->epsfac);
  96     charge           = mdatoms->chargeA;
  97
  98     vftab            = kernel_data->table_elec->data;
  99     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 100
 101     /* Avoid stupid compiler warnings */
 102     jnrA = jnrB = jnrC = jnrD = 0;
 103     j_coord_offsetA = 0;
 104     j_coord_offsetB = 0;
 105     j_coord_offsetC = 0;
 106     j_coord_offsetD = 0;
 107
 108     outeriter        = 0;
 109     inneriter        = 0;
 110
 111     for(iidx=0;iidx<4*DIM;iidx++)
 112     {
 113         scratch[iidx] = 0.0;
 114     }
 115
 116     /* Start outer loop over neighborlists */
 117     for(iidx=0; iidx<nri; iidx++)
 118     {
 119         /* Load shift vector for this list */
 120         i_shift_offset   = DIM*shiftidx[iidx];
 121
 122         /* Load limits for loop over neighbors */
 123         j_index_start    = jindex[iidx];
 124         j_index_end      = jindex[iidx+1];
 125
 126         /* Get outer coordinate index */
 127         inr              = iinr[iidx];
 128         i_coord_offset   = DIM*inr;
 129
 130         /* Load i particle coords and add shift vector */
 131         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 132
 133         fix0             = _mm_setzero_ps();
 134         fiy0             = _mm_setzero_ps();
 135         fiz0             = _mm_setzero_ps();
 136
 137         /* Load parameters for i particles */
 138         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 139
 140         /* Reset potential sums */
 141         velecsum         = _mm_setzero_ps();
 142
 143         /* Start inner kernel loop */
 144         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 145         {
 146
 147             /* Get j neighbor index, and coordinate index */
 148             jnrA             = jjnr[jidx];
 149             jnrB             = jjnr[jidx+1];
 150             jnrC             = jjnr[jidx+2];
 151             jnrD             = jjnr[jidx+3];
 152             j_coord_offsetA  = DIM*jnrA;
 153             j_coord_offsetB  = DIM*jnrB;
 154             j_coord_offsetC  = DIM*jnrC;
 155             j_coord_offsetD  = DIM*jnrD;
 156
 157             /* load j atom coordinates */
 158             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 159                                               x+j_coord_offsetC,x+j_coord_offsetD,
 160                                               &jx0,&jy0,&jz0);
 161
 162             /* Calculate displacement vector */
 163             dx00             = _mm_sub_ps(ix0,jx0);
 164             dy00             = _mm_sub_ps(iy0,jy0);
 165             dz00             = _mm_sub_ps(iz0,jz0);
 166
 167             /* Calculate squared distance and things based on it */
 168             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 169
 170             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 171
 172             /* Load parameters for j particles */
 173             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 174                                                               charge+jnrC+0,charge+jnrD+0);
 175
 176             /**************************
 177              * CALCULATE INTERACTIONS *
 178              **************************/
 179
 180             r00              = _mm_mul_ps(rsq00,rinv00);
 181
 182             /* Compute parameters for interactions between i and j atoms */
 183             qq00             = _mm_mul_ps(iq0,jq0);
 184
 185             /* Calculate table index by multiplying r with table scale and truncate to integer */
 186             rt               = _mm_mul_ps(r00,vftabscale);
 187             vfitab           = _mm_cvttps_epi32(rt);
 188             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 189             vfitab           = _mm_slli_epi32(vfitab,2);
 190
 191             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 192             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 193             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 194             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 195             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 196             _MM_TRANSPOSE4_PS(Y,F,G,H);
 197             Heps             = _mm_mul_ps(vfeps,H);
 198             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 199             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 200             velec            = _mm_mul_ps(qq00,VV);
 201             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 202             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 203
 204             /* Update potential sum for this i atom from the interaction with this j atom. */
 205             velecsum         = _mm_add_ps(velecsum,velec);
 206
 207             fscal            = felec;
 208
 209             /* Calculate temporary vectorial force */
 210             tx               = _mm_mul_ps(fscal,dx00);
 211             ty               = _mm_mul_ps(fscal,dy00);
 212             tz               = _mm_mul_ps(fscal,dz00);
 213
 214             /* Update vectorial force */
 215             fix0             = _mm_add_ps(fix0,tx);
 216             fiy0             = _mm_add_ps(fiy0,ty);
 217             fiz0             = _mm_add_ps(fiz0,tz);
 218
 219             fjptrA             = f+j_coord_offsetA;
 220             fjptrB             = f+j_coord_offsetB;
 221             fjptrC             = f+j_coord_offsetC;
 222             fjptrD             = f+j_coord_offsetD;
 223             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 224
 225             /* Inner loop uses 43 flops */
 226         }
 227
 228         if(jidx<j_index_end)
 229         {
 230
 231             /* Get j neighbor index, and coordinate index */
 232             jnrlistA         = jjnr[jidx];
 233             jnrlistB         = jjnr[jidx+1];
 234             jnrlistC         = jjnr[jidx+2];
 235             jnrlistD         = jjnr[jidx+3];
 236             /* Sign of each element will be negative for non-real atoms.
 237              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 238              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 239              */
 240             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 241             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 242             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 243             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 244             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 245             j_coord_offsetA  = DIM*jnrA;
 246             j_coord_offsetB  = DIM*jnrB;
 247             j_coord_offsetC  = DIM*jnrC;
 248             j_coord_offsetD  = DIM*jnrD;
 249
 250             /* load j atom coordinates */
 251             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 252                                               x+j_coord_offsetC,x+j_coord_offsetD,
 253                                               &jx0,&jy0,&jz0);
 254
 255             /* Calculate displacement vector */
 256             dx00             = _mm_sub_ps(ix0,jx0);
 257             dy00             = _mm_sub_ps(iy0,jy0);
 258             dz00             = _mm_sub_ps(iz0,jz0);
 259
 260             /* Calculate squared distance and things based on it */
 261             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 262
 263             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 264
 265             /* Load parameters for j particles */
 266             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 267                                                               charge+jnrC+0,charge+jnrD+0);
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             r00              = _mm_mul_ps(rsq00,rinv00);
 274             r00              = _mm_andnot_ps(dummy_mask,r00);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq00             = _mm_mul_ps(iq0,jq0);
 278
 279             /* Calculate table index by multiplying r with table scale and truncate to integer */
 280             rt               = _mm_mul_ps(r00,vftabscale);
 281             vfitab           = _mm_cvttps_epi32(rt);
 282             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 283             vfitab           = _mm_slli_epi32(vfitab,2);
 284
 285             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 286             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 287             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 288             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 289             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 290             _MM_TRANSPOSE4_PS(Y,F,G,H);
 291             Heps             = _mm_mul_ps(vfeps,H);
 292             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 293             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 294             velec            = _mm_mul_ps(qq00,VV);
 295             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 296             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm_andnot_ps(dummy_mask,velec);
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_ps(fscal,dx00);
 308             ty               = _mm_mul_ps(fscal,dy00);
 309             tz               = _mm_mul_ps(fscal,dz00);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm_add_ps(fix0,tx);
 313             fiy0             = _mm_add_ps(fiy0,ty);
 314             fiz0             = _mm_add_ps(fiz0,tz);
 315
 316             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 317             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 318             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 319             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 320             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 321
 322             /* Inner loop uses 44 flops */
 323         }
 324
 325         /* End of innermost loop */
 326
 327         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 328                                               f+i_coord_offset,fshift+i_shift_offset);
 329
 330         ggid                        = gid[iidx];
 331         /* Update potential energies */
 332         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 333
 334         /* Increment number of inner iterations */
 335         inneriter                  += j_index_end - j_index_start;
 336
 337         /* Outer loop uses 8 flops */
 338     }
 339
 340     /* Increment number of outer iterations */
 341     outeriter        += nri;
 342
 343     /* Update outer/inner flops */
 344
 345     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*44);
 346 }
 347 /*
 348  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse4_1_single
 349  * Electrostatics interaction: CubicSplineTable
 350  * VdW interaction:            None
 351  * Geometry:                   Particle-Particle
 352  * Calculate force/pot:        Force
 353  */
 354 void
 355 nb_kernel_ElecCSTab_VdwNone_GeomP1P1_F_sse4_1_single
 356                     (t_nblist * gmx_restrict                nlist,
 357                      rvec * gmx_restrict                    xx,
 358                      rvec * gmx_restrict                    ff,
 359                      t_forcerec * gmx_restrict              fr,
 360                      t_mdatoms * gmx_restrict               mdatoms,
 361                      nb_kernel_data_t * gmx_restrict        kernel_data,
 362                      t_nrnb * gmx_restrict                  nrnb)
 363 {
 364     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 365      * just 0 for non-waters.
 366      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 367      * jnr indices corresponding to data put in the four positions in the SIMD register.
 368      */
 369     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 370     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 371     int              jnrA,jnrB,jnrC,jnrD;
 372     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 373     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 374     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 375     real             rcutoff_scalar;
 376     real             *shiftvec,*fshift,*x,*f;
 377     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 378     real             scratch[4*DIM];
 379     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 380     int              vdwioffset0;
 381     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 382     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 383     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 384     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 385     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 386     real             *charge;
 387     __m128i          vfitab;
 388     __m128i          ifour       = _mm_set1_epi32(4);
 389     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 390     real             *vftab;
 391     __m128           dummy_mask,cutoff_mask;
 392     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 393     __m128           one     = _mm_set1_ps(1.0);
 394     __m128           two     = _mm_set1_ps(2.0);
 395     x                = xx[0];
 396     f                = ff[0];
 397
 398     nri              = nlist->nri;
 399     iinr             = nlist->iinr;
 400     jindex           = nlist->jindex;
 401     jjnr             = nlist->jjnr;
 402     shiftidx         = nlist->shift;
 403     gid              = nlist->gid;
 404     shiftvec         = fr->shift_vec[0];
 405     fshift           = fr->fshift[0];
 406     facel            = _mm_set1_ps(fr->epsfac);
 407     charge           = mdatoms->chargeA;
 408
 409     vftab            = kernel_data->table_elec->data;
 410     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 411
 412     /* Avoid stupid compiler warnings */
 413     jnrA = jnrB = jnrC = jnrD = 0;
 414     j_coord_offsetA = 0;
 415     j_coord_offsetB = 0;
 416     j_coord_offsetC = 0;
 417     j_coord_offsetD = 0;
 418
 419     outeriter        = 0;
 420     inneriter        = 0;
 421
 422     for(iidx=0;iidx<4*DIM;iidx++)
 423     {
 424         scratch[iidx] = 0.0;
 425     }
 426
 427     /* Start outer loop over neighborlists */
 428     for(iidx=0; iidx<nri; iidx++)
 429     {
 430         /* Load shift vector for this list */
 431         i_shift_offset   = DIM*shiftidx[iidx];
 432
 433         /* Load limits for loop over neighbors */
 434         j_index_start    = jindex[iidx];
 435         j_index_end      = jindex[iidx+1];
 436
 437         /* Get outer coordinate index */
 438         inr              = iinr[iidx];
 439         i_coord_offset   = DIM*inr;
 440
 441         /* Load i particle coords and add shift vector */
 442         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 443
 444         fix0             = _mm_setzero_ps();
 445         fiy0             = _mm_setzero_ps();
 446         fiz0             = _mm_setzero_ps();
 447
 448         /* Load parameters for i particles */
 449         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 450
 451         /* Start inner kernel loop */
 452         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 453         {
 454
 455             /* Get j neighbor index, and coordinate index */
 456             jnrA             = jjnr[jidx];
 457             jnrB             = jjnr[jidx+1];
 458             jnrC             = jjnr[jidx+2];
 459             jnrD             = jjnr[jidx+3];
 460             j_coord_offsetA  = DIM*jnrA;
 461             j_coord_offsetB  = DIM*jnrB;
 462             j_coord_offsetC  = DIM*jnrC;
 463             j_coord_offsetD  = DIM*jnrD;
 464
 465             /* load j atom coordinates */
 466             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 467                                               x+j_coord_offsetC,x+j_coord_offsetD,
 468                                               &jx0,&jy0,&jz0);
 469
 470             /* Calculate displacement vector */
 471             dx00             = _mm_sub_ps(ix0,jx0);
 472             dy00             = _mm_sub_ps(iy0,jy0);
 473             dz00             = _mm_sub_ps(iz0,jz0);
 474
 475             /* Calculate squared distance and things based on it */
 476             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 477
 478             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 479
 480             /* Load parameters for j particles */
 481             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 482                                                               charge+jnrC+0,charge+jnrD+0);
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             r00              = _mm_mul_ps(rsq00,rinv00);
 489
 490             /* Compute parameters for interactions between i and j atoms */
 491             qq00             = _mm_mul_ps(iq0,jq0);
 492
 493             /* Calculate table index by multiplying r with table scale and truncate to integer */
 494             rt               = _mm_mul_ps(r00,vftabscale);
 495             vfitab           = _mm_cvttps_epi32(rt);
 496             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 497             vfitab           = _mm_slli_epi32(vfitab,2);
 498
 499             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 500             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 501             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 502             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 503             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 504             _MM_TRANSPOSE4_PS(Y,F,G,H);
 505             Heps             = _mm_mul_ps(vfeps,H);
 506             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 507             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 508             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 509
 510             fscal            = felec;
 511
 512             /* Calculate temporary vectorial force */
 513             tx               = _mm_mul_ps(fscal,dx00);
 514             ty               = _mm_mul_ps(fscal,dy00);
 515             tz               = _mm_mul_ps(fscal,dz00);
 516
 517             /* Update vectorial force */
 518             fix0             = _mm_add_ps(fix0,tx);
 519             fiy0             = _mm_add_ps(fiy0,ty);
 520             fiz0             = _mm_add_ps(fiz0,tz);
 521
 522             fjptrA             = f+j_coord_offsetA;
 523             fjptrB             = f+j_coord_offsetB;
 524             fjptrC             = f+j_coord_offsetC;
 525             fjptrD             = f+j_coord_offsetD;
 526             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 527
 528             /* Inner loop uses 39 flops */
 529         }
 530
 531         if(jidx<j_index_end)
 532         {
 533
 534             /* Get j neighbor index, and coordinate index */
 535             jnrlistA         = jjnr[jidx];
 536             jnrlistB         = jjnr[jidx+1];
 537             jnrlistC         = jjnr[jidx+2];
 538             jnrlistD         = jjnr[jidx+3];
 539             /* Sign of each element will be negative for non-real atoms.
 540              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 541              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 542              */
 543             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 544             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 545             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 546             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 547             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 548             j_coord_offsetA  = DIM*jnrA;
 549             j_coord_offsetB  = DIM*jnrB;
 550             j_coord_offsetC  = DIM*jnrC;
 551             j_coord_offsetD  = DIM*jnrD;
 552
 553             /* load j atom coordinates */
 554             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 555                                               x+j_coord_offsetC,x+j_coord_offsetD,
 556                                               &jx0,&jy0,&jz0);
 557
 558             /* Calculate displacement vector */
 559             dx00             = _mm_sub_ps(ix0,jx0);
 560             dy00             = _mm_sub_ps(iy0,jy0);
 561             dz00             = _mm_sub_ps(iz0,jz0);
 562
 563             /* Calculate squared distance and things based on it */
 564             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 565
 566             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 567
 568             /* Load parameters for j particles */
 569             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 570                                                               charge+jnrC+0,charge+jnrD+0);
 571
 572             /**************************
 573              * CALCULATE INTERACTIONS *
 574              **************************/
 575
 576             r00              = _mm_mul_ps(rsq00,rinv00);
 577             r00              = _mm_andnot_ps(dummy_mask,r00);
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             qq00             = _mm_mul_ps(iq0,jq0);
 581
 582             /* Calculate table index by multiplying r with table scale and truncate to integer */
 583             rt               = _mm_mul_ps(r00,vftabscale);
 584             vfitab           = _mm_cvttps_epi32(rt);
 585             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 586             vfitab           = _mm_slli_epi32(vfitab,2);
 587
 588             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 589             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 590             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 591             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 592             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 593             _MM_TRANSPOSE4_PS(Y,F,G,H);
 594             Heps             = _mm_mul_ps(vfeps,H);
 595             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 596             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 597             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 598
 599             fscal            = felec;
 600
 601             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 602
 603             /* Calculate temporary vectorial force */
 604             tx               = _mm_mul_ps(fscal,dx00);
 605             ty               = _mm_mul_ps(fscal,dy00);
 606             tz               = _mm_mul_ps(fscal,dz00);
 607
 608             /* Update vectorial force */
 609             fix0             = _mm_add_ps(fix0,tx);
 610             fiy0             = _mm_add_ps(fiy0,ty);
 611             fiz0             = _mm_add_ps(fiz0,tz);
 612
 613             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 614             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 615             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 616             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 617             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 618
 619             /* Inner loop uses 40 flops */
 620         }
 621
 622         /* End of innermost loop */
 623
 624         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 625                                               f+i_coord_offset,fshift+i_shift_offset);
 626
 627         /* Increment number of inner iterations */
 628         inneriter                  += j_index_end - j_index_start;
 629
 630         /* Outer loop uses 7 flops */
 631     }
 632
 633     /* Increment number of outer iterations */
 634     outeriter        += nri;
 635
 636     /* Update outer/inner flops */
 637
 638     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*40);
 639 }