src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_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_ElecEwSh_VdwNone_GeomP1P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_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          ewitab;
  77     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  78     real             *ewtab;
  79     __m128           dummy_mask,cutoff_mask;
  80     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  81     __m128           one     = _mm_set1_ps(1.0);
  82     __m128           two     = _mm_set1_ps(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm_set1_ps(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96
  97     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
  98     ewtab            = fr->ic->tabq_coul_FDV0;
  99     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 100     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 101
 102     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 103     rcutoff_scalar   = fr->rcoulomb;
 104     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 105     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 106
 107     /* Avoid stupid compiler warnings */
 108     jnrA = jnrB = jnrC = jnrD = 0;
 109     j_coord_offsetA = 0;
 110     j_coord_offsetB = 0;
 111     j_coord_offsetC = 0;
 112     j_coord_offsetD = 0;
 113
 114     outeriter        = 0;
 115     inneriter        = 0;
 116
 117     for(iidx=0;iidx<4*DIM;iidx++)
 118     {
 119         scratch[iidx] = 0.0;
 120     }
 121
 122     /* Start outer loop over neighborlists */
 123     for(iidx=0; iidx<nri; iidx++)
 124     {
 125         /* Load shift vector for this list */
 126         i_shift_offset   = DIM*shiftidx[iidx];
 127
 128         /* Load limits for loop over neighbors */
 129         j_index_start    = jindex[iidx];
 130         j_index_end      = jindex[iidx+1];
 131
 132         /* Get outer coordinate index */
 133         inr              = iinr[iidx];
 134         i_coord_offset   = DIM*inr;
 135
 136         /* Load i particle coords and add shift vector */
 137         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 138
 139         fix0             = _mm_setzero_ps();
 140         fiy0             = _mm_setzero_ps();
 141         fiz0             = _mm_setzero_ps();
 142
 143         /* Load parameters for i particles */
 144         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 145
 146         /* Reset potential sums */
 147         velecsum         = _mm_setzero_ps();
 148
 149         /* Start inner kernel loop */
 150         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 151         {
 152
 153             /* Get j neighbor index, and coordinate index */
 154             jnrA             = jjnr[jidx];
 155             jnrB             = jjnr[jidx+1];
 156             jnrC             = jjnr[jidx+2];
 157             jnrD             = jjnr[jidx+3];
 158             j_coord_offsetA  = DIM*jnrA;
 159             j_coord_offsetB  = DIM*jnrB;
 160             j_coord_offsetC  = DIM*jnrC;
 161             j_coord_offsetD  = DIM*jnrD;
 162
 163             /* load j atom coordinates */
 164             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               x+j_coord_offsetC,x+j_coord_offsetD,
 166                                               &jx0,&jy0,&jz0);
 167
 168             /* Calculate displacement vector */
 169             dx00             = _mm_sub_ps(ix0,jx0);
 170             dy00             = _mm_sub_ps(iy0,jy0);
 171             dz00             = _mm_sub_ps(iz0,jz0);
 172
 173             /* Calculate squared distance and things based on it */
 174             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 175
 176             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 177
 178             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 179
 180             /* Load parameters for j particles */
 181             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 182                                                               charge+jnrC+0,charge+jnrD+0);
 183
 184             /**************************
 185              * CALCULATE INTERACTIONS *
 186              **************************/
 187
 188             if (gmx_mm_any_lt(rsq00,rcutoff2))
 189             {
 190
 191             r00              = _mm_mul_ps(rsq00,rinv00);
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             qq00             = _mm_mul_ps(iq0,jq0);
 195
 196             /* EWALD ELECTROSTATICS */
 197
 198             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 199             ewrt             = _mm_mul_ps(r00,ewtabscale);
 200             ewitab           = _mm_cvttps_epi32(ewrt);
 201             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 202             ewitab           = _mm_slli_epi32(ewitab,2);
 203             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 204             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 205             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 206             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 207             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 208             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 209             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 210             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 211             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 212
 213             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 214
 215             /* Update potential sum for this i atom from the interaction with this j atom. */
 216             velec            = _mm_and_ps(velec,cutoff_mask);
 217             velecsum         = _mm_add_ps(velecsum,velec);
 218
 219             fscal            = felec;
 220
 221             fscal            = _mm_and_ps(fscal,cutoff_mask);
 222
 223             /* Calculate temporary vectorial force */
 224             tx               = _mm_mul_ps(fscal,dx00);
 225             ty               = _mm_mul_ps(fscal,dy00);
 226             tz               = _mm_mul_ps(fscal,dz00);
 227
 228             /* Update vectorial force */
 229             fix0             = _mm_add_ps(fix0,tx);
 230             fiy0             = _mm_add_ps(fiy0,ty);
 231             fiz0             = _mm_add_ps(fiz0,tz);
 232
 233             fjptrA             = f+j_coord_offsetA;
 234             fjptrB             = f+j_coord_offsetB;
 235             fjptrC             = f+j_coord_offsetC;
 236             fjptrD             = f+j_coord_offsetD;
 237             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 238
 239             }
 240
 241             /* Inner loop uses 46 flops */
 242         }
 243
 244         if(jidx<j_index_end)
 245         {
 246
 247             /* Get j neighbor index, and coordinate index */
 248             jnrlistA         = jjnr[jidx];
 249             jnrlistB         = jjnr[jidx+1];
 250             jnrlistC         = jjnr[jidx+2];
 251             jnrlistD         = jjnr[jidx+3];
 252             /* Sign of each element will be negative for non-real atoms.
 253              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 254              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 255              */
 256             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 257             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 258             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 259             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 260             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 261             j_coord_offsetA  = DIM*jnrA;
 262             j_coord_offsetB  = DIM*jnrB;
 263             j_coord_offsetC  = DIM*jnrC;
 264             j_coord_offsetD  = DIM*jnrD;
 265
 266             /* load j atom coordinates */
 267             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 268                                               x+j_coord_offsetC,x+j_coord_offsetD,
 269                                               &jx0,&jy0,&jz0);
 270
 271             /* Calculate displacement vector */
 272             dx00             = _mm_sub_ps(ix0,jx0);
 273             dy00             = _mm_sub_ps(iy0,jy0);
 274             dz00             = _mm_sub_ps(iz0,jz0);
 275
 276             /* Calculate squared distance and things based on it */
 277             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 278
 279             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 280
 281             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 282
 283             /* Load parameters for j particles */
 284             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 285                                                               charge+jnrC+0,charge+jnrD+0);
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             if (gmx_mm_any_lt(rsq00,rcutoff2))
 292             {
 293
 294             r00              = _mm_mul_ps(rsq00,rinv00);
 295             r00              = _mm_andnot_ps(dummy_mask,r00);
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq00             = _mm_mul_ps(iq0,jq0);
 299
 300             /* EWALD ELECTROSTATICS */
 301
 302             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 303             ewrt             = _mm_mul_ps(r00,ewtabscale);
 304             ewitab           = _mm_cvttps_epi32(ewrt);
 305             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 306             ewitab           = _mm_slli_epi32(ewitab,2);
 307             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 308             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 309             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 310             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 311             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 312             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 313             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 314             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 315             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 316
 317             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm_and_ps(velec,cutoff_mask);
 321             velec            = _mm_andnot_ps(dummy_mask,velec);
 322             velecsum         = _mm_add_ps(velecsum,velec);
 323
 324             fscal            = felec;
 325
 326             fscal            = _mm_and_ps(fscal,cutoff_mask);
 327
 328             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm_mul_ps(fscal,dx00);
 332             ty               = _mm_mul_ps(fscal,dy00);
 333             tz               = _mm_mul_ps(fscal,dz00);
 334
 335             /* Update vectorial force */
 336             fix0             = _mm_add_ps(fix0,tx);
 337             fiy0             = _mm_add_ps(fiy0,ty);
 338             fiz0             = _mm_add_ps(fiz0,tz);
 339
 340             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 341             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 342             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 343             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 344             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 345
 346             }
 347
 348             /* Inner loop uses 47 flops */
 349         }
 350
 351         /* End of innermost loop */
 352
 353         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 354                                               f+i_coord_offset,fshift+i_shift_offset);
 355
 356         ggid                        = gid[iidx];
 357         /* Update potential energies */
 358         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 359
 360         /* Increment number of inner iterations */
 361         inneriter                  += j_index_end - j_index_start;
 362
 363         /* Outer loop uses 8 flops */
 364     }
 365
 366     /* Increment number of outer iterations */
 367     outeriter        += nri;
 368
 369     /* Update outer/inner flops */
 370
 371     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*47);
 372 }
 373 /*
 374  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_single
 375  * Electrostatics interaction: Ewald
 376  * VdW interaction:            None
 377  * Geometry:                   Particle-Particle
 378  * Calculate force/pot:        Force
 379  */
 380 void
 381 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse4_1_single
 382                     (t_nblist * gmx_restrict                nlist,
 383                      rvec * gmx_restrict                    xx,
 384                      rvec * gmx_restrict                    ff,
 385                      t_forcerec * gmx_restrict              fr,
 386                      t_mdatoms * gmx_restrict               mdatoms,
 387                      nb_kernel_data_t * gmx_restrict        kernel_data,
 388                      t_nrnb * gmx_restrict                  nrnb)
 389 {
 390     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 391      * just 0 for non-waters.
 392      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 393      * jnr indices corresponding to data put in the four positions in the SIMD register.
 394      */
 395     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 396     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 397     int              jnrA,jnrB,jnrC,jnrD;
 398     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 399     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 400     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 401     real             rcutoff_scalar;
 402     real             *shiftvec,*fshift,*x,*f;
 403     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 404     real             scratch[4*DIM];
 405     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 406     int              vdwioffset0;
 407     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 408     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 409     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 410     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 411     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 412     real             *charge;
 413     __m128i          ewitab;
 414     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 415     real             *ewtab;
 416     __m128           dummy_mask,cutoff_mask;
 417     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 418     __m128           one     = _mm_set1_ps(1.0);
 419     __m128           two     = _mm_set1_ps(2.0);
 420     x                = xx[0];
 421     f                = ff[0];
 422
 423     nri              = nlist->nri;
 424     iinr             = nlist->iinr;
 425     jindex           = nlist->jindex;
 426     jjnr             = nlist->jjnr;
 427     shiftidx         = nlist->shift;
 428     gid              = nlist->gid;
 429     shiftvec         = fr->shift_vec[0];
 430     fshift           = fr->fshift[0];
 431     facel            = _mm_set1_ps(fr->epsfac);
 432     charge           = mdatoms->chargeA;
 433
 434     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 435     ewtab            = fr->ic->tabq_coul_F;
 436     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 437     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 438
 439     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 440     rcutoff_scalar   = fr->rcoulomb;
 441     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 442     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 443
 444     /* Avoid stupid compiler warnings */
 445     jnrA = jnrB = jnrC = jnrD = 0;
 446     j_coord_offsetA = 0;
 447     j_coord_offsetB = 0;
 448     j_coord_offsetC = 0;
 449     j_coord_offsetD = 0;
 450
 451     outeriter        = 0;
 452     inneriter        = 0;
 453
 454     for(iidx=0;iidx<4*DIM;iidx++)
 455     {
 456         scratch[iidx] = 0.0;
 457     }
 458
 459     /* Start outer loop over neighborlists */
 460     for(iidx=0; iidx<nri; iidx++)
 461     {
 462         /* Load shift vector for this list */
 463         i_shift_offset   = DIM*shiftidx[iidx];
 464
 465         /* Load limits for loop over neighbors */
 466         j_index_start    = jindex[iidx];
 467         j_index_end      = jindex[iidx+1];
 468
 469         /* Get outer coordinate index */
 470         inr              = iinr[iidx];
 471         i_coord_offset   = DIM*inr;
 472
 473         /* Load i particle coords and add shift vector */
 474         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 475
 476         fix0             = _mm_setzero_ps();
 477         fiy0             = _mm_setzero_ps();
 478         fiz0             = _mm_setzero_ps();
 479
 480         /* Load parameters for i particles */
 481         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 482
 483         /* Start inner kernel loop */
 484         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 485         {
 486
 487             /* Get j neighbor index, and coordinate index */
 488             jnrA             = jjnr[jidx];
 489             jnrB             = jjnr[jidx+1];
 490             jnrC             = jjnr[jidx+2];
 491             jnrD             = jjnr[jidx+3];
 492             j_coord_offsetA  = DIM*jnrA;
 493             j_coord_offsetB  = DIM*jnrB;
 494             j_coord_offsetC  = DIM*jnrC;
 495             j_coord_offsetD  = DIM*jnrD;
 496
 497             /* load j atom coordinates */
 498             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 499                                               x+j_coord_offsetC,x+j_coord_offsetD,
 500                                               &jx0,&jy0,&jz0);
 501
 502             /* Calculate displacement vector */
 503             dx00             = _mm_sub_ps(ix0,jx0);
 504             dy00             = _mm_sub_ps(iy0,jy0);
 505             dz00             = _mm_sub_ps(iz0,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 509
 510             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 511
 512             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 513
 514             /* Load parameters for j particles */
 515             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 516                                                               charge+jnrC+0,charge+jnrD+0);
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             if (gmx_mm_any_lt(rsq00,rcutoff2))
 523             {
 524
 525             r00              = _mm_mul_ps(rsq00,rinv00);
 526
 527             /* Compute parameters for interactions between i and j atoms */
 528             qq00             = _mm_mul_ps(iq0,jq0);
 529
 530             /* EWALD ELECTROSTATICS */
 531
 532             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 533             ewrt             = _mm_mul_ps(r00,ewtabscale);
 534             ewitab           = _mm_cvttps_epi32(ewrt);
 535             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 536             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 537                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 538                                          &ewtabF,&ewtabFn);
 539             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 540             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 541
 542             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 543
 544             fscal            = felec;
 545
 546             fscal            = _mm_and_ps(fscal,cutoff_mask);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm_mul_ps(fscal,dx00);
 550             ty               = _mm_mul_ps(fscal,dy00);
 551             tz               = _mm_mul_ps(fscal,dz00);
 552
 553             /* Update vectorial force */
 554             fix0             = _mm_add_ps(fix0,tx);
 555             fiy0             = _mm_add_ps(fiy0,ty);
 556             fiz0             = _mm_add_ps(fiz0,tz);
 557
 558             fjptrA             = f+j_coord_offsetA;
 559             fjptrB             = f+j_coord_offsetB;
 560             fjptrC             = f+j_coord_offsetC;
 561             fjptrD             = f+j_coord_offsetD;
 562             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 563
 564             }
 565
 566             /* Inner loop uses 39 flops */
 567         }
 568
 569         if(jidx<j_index_end)
 570         {
 571
 572             /* Get j neighbor index, and coordinate index */
 573             jnrlistA         = jjnr[jidx];
 574             jnrlistB         = jjnr[jidx+1];
 575             jnrlistC         = jjnr[jidx+2];
 576             jnrlistD         = jjnr[jidx+3];
 577             /* Sign of each element will be negative for non-real atoms.
 578              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 579              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 580              */
 581             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 582             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 583             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 584             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 585             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 586             j_coord_offsetA  = DIM*jnrA;
 587             j_coord_offsetB  = DIM*jnrB;
 588             j_coord_offsetC  = DIM*jnrC;
 589             j_coord_offsetD  = DIM*jnrD;
 590
 591             /* load j atom coordinates */
 592             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 593                                               x+j_coord_offsetC,x+j_coord_offsetD,
 594                                               &jx0,&jy0,&jz0);
 595
 596             /* Calculate displacement vector */
 597             dx00             = _mm_sub_ps(ix0,jx0);
 598             dy00             = _mm_sub_ps(iy0,jy0);
 599             dz00             = _mm_sub_ps(iz0,jz0);
 600
 601             /* Calculate squared distance and things based on it */
 602             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 603
 604             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 605
 606             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 607
 608             /* Load parameters for j particles */
 609             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 610                                                               charge+jnrC+0,charge+jnrD+0);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             if (gmx_mm_any_lt(rsq00,rcutoff2))
 617             {
 618
 619             r00              = _mm_mul_ps(rsq00,rinv00);
 620             r00              = _mm_andnot_ps(dummy_mask,r00);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq00             = _mm_mul_ps(iq0,jq0);
 624
 625             /* EWALD ELECTROSTATICS */
 626
 627             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 628             ewrt             = _mm_mul_ps(r00,ewtabscale);
 629             ewitab           = _mm_cvttps_epi32(ewrt);
 630             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 631             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 632                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 633                                          &ewtabF,&ewtabFn);
 634             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 635             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 636
 637             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_and_ps(fscal,cutoff_mask);
 642
 643             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm_mul_ps(fscal,dx00);
 647             ty               = _mm_mul_ps(fscal,dy00);
 648             tz               = _mm_mul_ps(fscal,dz00);
 649
 650             /* Update vectorial force */
 651             fix0             = _mm_add_ps(fix0,tx);
 652             fiy0             = _mm_add_ps(fiy0,ty);
 653             fiz0             = _mm_add_ps(fiz0,tz);
 654
 655             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 656             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 657             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 658             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 659             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 660
 661             }
 662
 663             /* Inner loop uses 40 flops */
 664         }
 665
 666         /* End of innermost loop */
 667
 668         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 669                                               f+i_coord_offset,fshift+i_shift_offset);
 670
 671         /* Increment number of inner iterations */
 672         inneriter                  += j_index_end - j_index_start;
 673
 674         /* Outer loop uses 7 flops */
 675     }
 676
 677     /* Increment number of outer iterations */
 678     outeriter        += nri;
 679
 680     /* Update outer/inner flops */
 681
 682     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*40);
 683 }