src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_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_ElecEw_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_ElecEw_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     /* Avoid stupid compiler warnings */
 103     jnrA = jnrB = jnrC = jnrD = 0;
 104     j_coord_offsetA = 0;
 105     j_coord_offsetB = 0;
 106     j_coord_offsetC = 0;
 107     j_coord_offsetD = 0;
 108
 109     outeriter        = 0;
 110     inneriter        = 0;
 111
 112     for(iidx=0;iidx<4*DIM;iidx++)
 113     {
 114         scratch[iidx] = 0.0;
 115     }
 116
 117     /* Start outer loop over neighborlists */
 118     for(iidx=0; iidx<nri; iidx++)
 119     {
 120         /* Load shift vector for this list */
 121         i_shift_offset   = DIM*shiftidx[iidx];
 122
 123         /* Load limits for loop over neighbors */
 124         j_index_start    = jindex[iidx];
 125         j_index_end      = jindex[iidx+1];
 126
 127         /* Get outer coordinate index */
 128         inr              = iinr[iidx];
 129         i_coord_offset   = DIM*inr;
 130
 131         /* Load i particle coords and add shift vector */
 132         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 133
 134         fix0             = _mm_setzero_ps();
 135         fiy0             = _mm_setzero_ps();
 136         fiz0             = _mm_setzero_ps();
 137
 138         /* Load parameters for i particles */
 139         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 140
 141         /* Reset potential sums */
 142         velecsum         = _mm_setzero_ps();
 143
 144         /* Start inner kernel loop */
 145         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 146         {
 147
 148             /* Get j neighbor index, and coordinate index */
 149             jnrA             = jjnr[jidx];
 150             jnrB             = jjnr[jidx+1];
 151             jnrC             = jjnr[jidx+2];
 152             jnrD             = jjnr[jidx+3];
 153             j_coord_offsetA  = DIM*jnrA;
 154             j_coord_offsetB  = DIM*jnrB;
 155             j_coord_offsetC  = DIM*jnrC;
 156             j_coord_offsetD  = DIM*jnrD;
 157
 158             /* load j atom coordinates */
 159             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 160                                               x+j_coord_offsetC,x+j_coord_offsetD,
 161                                               &jx0,&jy0,&jz0);
 162
 163             /* Calculate displacement vector */
 164             dx00             = _mm_sub_ps(ix0,jx0);
 165             dy00             = _mm_sub_ps(iy0,jy0);
 166             dz00             = _mm_sub_ps(iz0,jz0);
 167
 168             /* Calculate squared distance and things based on it */
 169             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 170
 171             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 172
 173             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 174
 175             /* Load parameters for j particles */
 176             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 177                                                               charge+jnrC+0,charge+jnrD+0);
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             r00              = _mm_mul_ps(rsq00,rinv00);
 184
 185             /* Compute parameters for interactions between i and j atoms */
 186             qq00             = _mm_mul_ps(iq0,jq0);
 187
 188             /* EWALD ELECTROSTATICS */
 189
 190             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 191             ewrt             = _mm_mul_ps(r00,ewtabscale);
 192             ewitab           = _mm_cvttps_epi32(ewrt);
 193             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 194             ewitab           = _mm_slli_epi32(ewitab,2);
 195             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 196             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 197             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 198             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 199             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 200             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 201             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 202             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 203             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 204
 205             /* Update potential sum for this i atom from the interaction with this j atom. */
 206             velecsum         = _mm_add_ps(velecsum,velec);
 207
 208             fscal            = felec;
 209
 210             /* Calculate temporary vectorial force */
 211             tx               = _mm_mul_ps(fscal,dx00);
 212             ty               = _mm_mul_ps(fscal,dy00);
 213             tz               = _mm_mul_ps(fscal,dz00);
 214
 215             /* Update vectorial force */
 216             fix0             = _mm_add_ps(fix0,tx);
 217             fiy0             = _mm_add_ps(fiy0,ty);
 218             fiz0             = _mm_add_ps(fiz0,tz);
 219
 220             fjptrA             = f+j_coord_offsetA;
 221             fjptrB             = f+j_coord_offsetB;
 222             fjptrC             = f+j_coord_offsetC;
 223             fjptrD             = f+j_coord_offsetD;
 224             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 225
 226             /* Inner loop uses 41 flops */
 227         }
 228
 229         if(jidx<j_index_end)
 230         {
 231
 232             /* Get j neighbor index, and coordinate index */
 233             jnrlistA         = jjnr[jidx];
 234             jnrlistB         = jjnr[jidx+1];
 235             jnrlistC         = jjnr[jidx+2];
 236             jnrlistD         = jjnr[jidx+3];
 237             /* Sign of each element will be negative for non-real atoms.
 238              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 239              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 240              */
 241             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 242             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 243             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 244             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 245             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 246             j_coord_offsetA  = DIM*jnrA;
 247             j_coord_offsetB  = DIM*jnrB;
 248             j_coord_offsetC  = DIM*jnrC;
 249             j_coord_offsetD  = DIM*jnrD;
 250
 251             /* load j atom coordinates */
 252             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 253                                               x+j_coord_offsetC,x+j_coord_offsetD,
 254                                               &jx0,&jy0,&jz0);
 255
 256             /* Calculate displacement vector */
 257             dx00             = _mm_sub_ps(ix0,jx0);
 258             dy00             = _mm_sub_ps(iy0,jy0);
 259             dz00             = _mm_sub_ps(iz0,jz0);
 260
 261             /* Calculate squared distance and things based on it */
 262             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 263
 264             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 265
 266             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 267
 268             /* Load parameters for j particles */
 269             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 270                                                               charge+jnrC+0,charge+jnrD+0);
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             r00              = _mm_mul_ps(rsq00,rinv00);
 277             r00              = _mm_andnot_ps(dummy_mask,r00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq00             = _mm_mul_ps(iq0,jq0);
 281
 282             /* EWALD ELECTROSTATICS */
 283
 284             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 285             ewrt             = _mm_mul_ps(r00,ewtabscale);
 286             ewitab           = _mm_cvttps_epi32(ewrt);
 287             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 288             ewitab           = _mm_slli_epi32(ewitab,2);
 289             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 290             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 291             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 292             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 293             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 294             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 295             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 296             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 297             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velec            = _mm_andnot_ps(dummy_mask,velec);
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302
 303             fscal            = felec;
 304
 305             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 306
 307             /* Calculate temporary vectorial force */
 308             tx               = _mm_mul_ps(fscal,dx00);
 309             ty               = _mm_mul_ps(fscal,dy00);
 310             tz               = _mm_mul_ps(fscal,dz00);
 311
 312             /* Update vectorial force */
 313             fix0             = _mm_add_ps(fix0,tx);
 314             fiy0             = _mm_add_ps(fiy0,ty);
 315             fiz0             = _mm_add_ps(fiz0,tz);
 316
 317             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 318             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 319             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 320             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 321             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 322
 323             /* Inner loop uses 42 flops */
 324         }
 325
 326         /* End of innermost loop */
 327
 328         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 329                                               f+i_coord_offset,fshift+i_shift_offset);
 330
 331         ggid                        = gid[iidx];
 332         /* Update potential energies */
 333         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 334
 335         /* Increment number of inner iterations */
 336         inneriter                  += j_index_end - j_index_start;
 337
 338         /* Outer loop uses 8 flops */
 339     }
 340
 341     /* Increment number of outer iterations */
 342     outeriter        += nri;
 343
 344     /* Update outer/inner flops */
 345
 346     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*42);
 347 }
 348 /*
 349  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_single
 350  * Electrostatics interaction: Ewald
 351  * VdW interaction:            None
 352  * Geometry:                   Particle-Particle
 353  * Calculate force/pot:        Force
 354  */
 355 void
 356 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse4_1_single
 357                     (t_nblist * gmx_restrict                nlist,
 358                      rvec * gmx_restrict                    xx,
 359                      rvec * gmx_restrict                    ff,
 360                      t_forcerec * gmx_restrict              fr,
 361                      t_mdatoms * gmx_restrict               mdatoms,
 362                      nb_kernel_data_t * gmx_restrict        kernel_data,
 363                      t_nrnb * gmx_restrict                  nrnb)
 364 {
 365     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 366      * just 0 for non-waters.
 367      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 368      * jnr indices corresponding to data put in the four positions in the SIMD register.
 369      */
 370     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 371     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 372     int              jnrA,jnrB,jnrC,jnrD;
 373     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 374     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 375     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 376     real             rcutoff_scalar;
 377     real             *shiftvec,*fshift,*x,*f;
 378     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 379     real             scratch[4*DIM];
 380     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 381     int              vdwioffset0;
 382     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 383     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 384     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 385     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 386     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 387     real             *charge;
 388     __m128i          ewitab;
 389     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 390     real             *ewtab;
 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     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 410     ewtab            = fr->ic->tabq_coul_F;
 411     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 412     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 413
 414     /* Avoid stupid compiler warnings */
 415     jnrA = jnrB = jnrC = jnrD = 0;
 416     j_coord_offsetA = 0;
 417     j_coord_offsetB = 0;
 418     j_coord_offsetC = 0;
 419     j_coord_offsetD = 0;
 420
 421     outeriter        = 0;
 422     inneriter        = 0;
 423
 424     for(iidx=0;iidx<4*DIM;iidx++)
 425     {
 426         scratch[iidx] = 0.0;
 427     }
 428
 429     /* Start outer loop over neighborlists */
 430     for(iidx=0; iidx<nri; iidx++)
 431     {
 432         /* Load shift vector for this list */
 433         i_shift_offset   = DIM*shiftidx[iidx];
 434
 435         /* Load limits for loop over neighbors */
 436         j_index_start    = jindex[iidx];
 437         j_index_end      = jindex[iidx+1];
 438
 439         /* Get outer coordinate index */
 440         inr              = iinr[iidx];
 441         i_coord_offset   = DIM*inr;
 442
 443         /* Load i particle coords and add shift vector */
 444         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 445
 446         fix0             = _mm_setzero_ps();
 447         fiy0             = _mm_setzero_ps();
 448         fiz0             = _mm_setzero_ps();
 449
 450         /* Load parameters for i particles */
 451         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 452
 453         /* Start inner kernel loop */
 454         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 455         {
 456
 457             /* Get j neighbor index, and coordinate index */
 458             jnrA             = jjnr[jidx];
 459             jnrB             = jjnr[jidx+1];
 460             jnrC             = jjnr[jidx+2];
 461             jnrD             = jjnr[jidx+3];
 462             j_coord_offsetA  = DIM*jnrA;
 463             j_coord_offsetB  = DIM*jnrB;
 464             j_coord_offsetC  = DIM*jnrC;
 465             j_coord_offsetD  = DIM*jnrD;
 466
 467             /* load j atom coordinates */
 468             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 469                                               x+j_coord_offsetC,x+j_coord_offsetD,
 470                                               &jx0,&jy0,&jz0);
 471
 472             /* Calculate displacement vector */
 473             dx00             = _mm_sub_ps(ix0,jx0);
 474             dy00             = _mm_sub_ps(iy0,jy0);
 475             dz00             = _mm_sub_ps(iz0,jz0);
 476
 477             /* Calculate squared distance and things based on it */
 478             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 479
 480             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 481
 482             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 483
 484             /* Load parameters for j particles */
 485             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 486                                                               charge+jnrC+0,charge+jnrD+0);
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             r00              = _mm_mul_ps(rsq00,rinv00);
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             qq00             = _mm_mul_ps(iq0,jq0);
 496
 497             /* EWALD ELECTROSTATICS */
 498
 499             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 500             ewrt             = _mm_mul_ps(r00,ewtabscale);
 501             ewitab           = _mm_cvttps_epi32(ewrt);
 502             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 503             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 504                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 505                                          &ewtabF,&ewtabFn);
 506             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 507             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 508
 509             fscal            = felec;
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm_mul_ps(fscal,dx00);
 513             ty               = _mm_mul_ps(fscal,dy00);
 514             tz               = _mm_mul_ps(fscal,dz00);
 515
 516             /* Update vectorial force */
 517             fix0             = _mm_add_ps(fix0,tx);
 518             fiy0             = _mm_add_ps(fiy0,ty);
 519             fiz0             = _mm_add_ps(fiz0,tz);
 520
 521             fjptrA             = f+j_coord_offsetA;
 522             fjptrB             = f+j_coord_offsetB;
 523             fjptrC             = f+j_coord_offsetC;
 524             fjptrD             = f+j_coord_offsetD;
 525             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 526
 527             /* Inner loop uses 36 flops */
 528         }
 529
 530         if(jidx<j_index_end)
 531         {
 532
 533             /* Get j neighbor index, and coordinate index */
 534             jnrlistA         = jjnr[jidx];
 535             jnrlistB         = jjnr[jidx+1];
 536             jnrlistC         = jjnr[jidx+2];
 537             jnrlistD         = jjnr[jidx+3];
 538             /* Sign of each element will be negative for non-real atoms.
 539              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 540              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 541              */
 542             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 543             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 544             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 545             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 546             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 547             j_coord_offsetA  = DIM*jnrA;
 548             j_coord_offsetB  = DIM*jnrB;
 549             j_coord_offsetC  = DIM*jnrC;
 550             j_coord_offsetD  = DIM*jnrD;
 551
 552             /* load j atom coordinates */
 553             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 554                                               x+j_coord_offsetC,x+j_coord_offsetD,
 555                                               &jx0,&jy0,&jz0);
 556
 557             /* Calculate displacement vector */
 558             dx00             = _mm_sub_ps(ix0,jx0);
 559             dy00             = _mm_sub_ps(iy0,jy0);
 560             dz00             = _mm_sub_ps(iz0,jz0);
 561
 562             /* Calculate squared distance and things based on it */
 563             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 564
 565             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 566
 567             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 568
 569             /* Load parameters for j particles */
 570             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 571                                                               charge+jnrC+0,charge+jnrD+0);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             r00              = _mm_mul_ps(rsq00,rinv00);
 578             r00              = _mm_andnot_ps(dummy_mask,r00);
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq00             = _mm_mul_ps(iq0,jq0);
 582
 583             /* EWALD ELECTROSTATICS */
 584
 585             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 586             ewrt             = _mm_mul_ps(r00,ewtabscale);
 587             ewitab           = _mm_cvttps_epi32(ewrt);
 588             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 589             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 590                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 591                                          &ewtabF,&ewtabFn);
 592             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 593             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 594
 595             fscal            = felec;
 596
 597             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm_mul_ps(fscal,dx00);
 601             ty               = _mm_mul_ps(fscal,dy00);
 602             tz               = _mm_mul_ps(fscal,dz00);
 603
 604             /* Update vectorial force */
 605             fix0             = _mm_add_ps(fix0,tx);
 606             fiy0             = _mm_add_ps(fiy0,ty);
 607             fiz0             = _mm_add_ps(fiz0,tz);
 608
 609             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 610             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 611             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 612             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 613             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 614
 615             /* Inner loop uses 37 flops */
 616         }
 617
 618         /* End of innermost loop */
 619
 620         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 621                                               f+i_coord_offset,fshift+i_shift_offset);
 622
 623         /* Increment number of inner iterations */
 624         inneriter                  += j_index_end - j_index_start;
 625
 626         /* Outer loop uses 7 flops */
 627     }
 628
 629     /* Increment number of outer iterations */
 630     outeriter        += nri;
 631
 632     /* Update outer/inner flops */
 633
 634     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*37);
 635 }