src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_avx_128_fma_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_avx_128_fma_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 AVX_128, 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           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,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  78     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  79     real             *ewtab;
  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     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
  99     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 100     beta2            = _mm_mul_ps(beta,beta);
 101     beta3            = _mm_mul_ps(beta,beta2);
 102     ewtab            = fr->ic->tabq_coul_FDV0;
 103     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 104     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 105
 106     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 107     rcutoff_scalar   = fr->rcoulomb;
 108     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 109     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = jnrC = jnrD = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115     j_coord_offsetC = 0;
 116     j_coord_offsetD = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     for(iidx=0;iidx<4*DIM;iidx++)
 122     {
 123         scratch[iidx] = 0.0;
 124     }
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _mm_setzero_ps();
 144         fiy0             = _mm_setzero_ps();
 145         fiz0             = _mm_setzero_ps();
 146
 147         /* Load parameters for i particles */
 148         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm_setzero_ps();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             jnrC             = jjnr[jidx+2];
 161             jnrD             = jjnr[jidx+3];
 162             j_coord_offsetA  = DIM*jnrA;
 163             j_coord_offsetB  = DIM*jnrB;
 164             j_coord_offsetC  = DIM*jnrC;
 165             j_coord_offsetD  = DIM*jnrD;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               x+j_coord_offsetC,x+j_coord_offsetD,
 170                                               &jx0,&jy0,&jz0);
 171
 172             /* Calculate displacement vector */
 173             dx00             = _mm_sub_ps(ix0,jx0);
 174             dy00             = _mm_sub_ps(iy0,jy0);
 175             dz00             = _mm_sub_ps(iz0,jz0);
 176
 177             /* Calculate squared distance and things based on it */
 178             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 179
 180             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 181
 182             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 183
 184             /* Load parameters for j particles */
 185             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 186                                                               charge+jnrC+0,charge+jnrD+0);
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             if (gmx_mm_any_lt(rsq00,rcutoff2))
 193             {
 194
 195             r00              = _mm_mul_ps(rsq00,rinv00);
 196
 197             /* Compute parameters for interactions between i and j atoms */
 198             qq00             = _mm_mul_ps(iq0,jq0);
 199
 200             /* EWALD ELECTROSTATICS */
 201
 202             /* Analytical PME correction */
 203             zeta2            = _mm_mul_ps(beta2,rsq00);
 204             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 205             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 206             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 207             felec            = _mm_mul_ps(qq00,felec);
 208             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 209             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 210             velec            = _mm_mul_ps(qq00,velec);
 211
 212             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 213
 214             /* Update potential sum for this i atom from the interaction with this j atom. */
 215             velec            = _mm_and_ps(velec,cutoff_mask);
 216             velecsum         = _mm_add_ps(velecsum,velec);
 217
 218             fscal            = felec;
 219
 220             fscal            = _mm_and_ps(fscal,cutoff_mask);
 221
 222              /* Update vectorial force */
 223             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 224             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 225             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 226
 227             fjptrA             = f+j_coord_offsetA;
 228             fjptrB             = f+j_coord_offsetB;
 229             fjptrC             = f+j_coord_offsetC;
 230             fjptrD             = f+j_coord_offsetD;
 231             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 232                                                    _mm_mul_ps(dx00,fscal),
 233                                                    _mm_mul_ps(dy00,fscal),
 234                                                    _mm_mul_ps(dz00,fscal));
 235
 236             }
 237
 238             /* Inner loop uses 33 flops */
 239         }
 240
 241         if(jidx<j_index_end)
 242         {
 243
 244             /* Get j neighbor index, and coordinate index */
 245             jnrlistA         = jjnr[jidx];
 246             jnrlistB         = jjnr[jidx+1];
 247             jnrlistC         = jjnr[jidx+2];
 248             jnrlistD         = jjnr[jidx+3];
 249             /* Sign of each element will be negative for non-real atoms.
 250              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 251              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 252              */
 253             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 254             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 255             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 256             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 257             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 258             j_coord_offsetA  = DIM*jnrA;
 259             j_coord_offsetB  = DIM*jnrB;
 260             j_coord_offsetC  = DIM*jnrC;
 261             j_coord_offsetD  = DIM*jnrD;
 262
 263             /* load j atom coordinates */
 264             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 265                                               x+j_coord_offsetC,x+j_coord_offsetD,
 266                                               &jx0,&jy0,&jz0);
 267
 268             /* Calculate displacement vector */
 269             dx00             = _mm_sub_ps(ix0,jx0);
 270             dy00             = _mm_sub_ps(iy0,jy0);
 271             dz00             = _mm_sub_ps(iz0,jz0);
 272
 273             /* Calculate squared distance and things based on it */
 274             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 275
 276             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 277
 278             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 279
 280             /* Load parameters for j particles */
 281             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 282                                                               charge+jnrC+0,charge+jnrD+0);
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             if (gmx_mm_any_lt(rsq00,rcutoff2))
 289             {
 290
 291             r00              = _mm_mul_ps(rsq00,rinv00);
 292             r00              = _mm_andnot_ps(dummy_mask,r00);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq00             = _mm_mul_ps(iq0,jq0);
 296
 297             /* EWALD ELECTROSTATICS */
 298
 299             /* Analytical PME correction */
 300             zeta2            = _mm_mul_ps(beta2,rsq00);
 301             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 302             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 303             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 304             felec            = _mm_mul_ps(qq00,felec);
 305             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 306             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv00,sh_ewald));
 307             velec            = _mm_mul_ps(qq00,velec);
 308
 309             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velec            = _mm_and_ps(velec,cutoff_mask);
 313             velec            = _mm_andnot_ps(dummy_mask,velec);
 314             velecsum         = _mm_add_ps(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             fscal            = _mm_and_ps(fscal,cutoff_mask);
 319
 320             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 321
 322              /* Update vectorial force */
 323             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 324             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 325             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 326
 327             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 328             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 329             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 330             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 331             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 332                                                    _mm_mul_ps(dx00,fscal),
 333                                                    _mm_mul_ps(dy00,fscal),
 334                                                    _mm_mul_ps(dz00,fscal));
 335
 336             }
 337
 338             /* Inner loop uses 34 flops */
 339         }
 340
 341         /* End of innermost loop */
 342
 343         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 344                                               f+i_coord_offset,fshift+i_shift_offset);
 345
 346         ggid                        = gid[iidx];
 347         /* Update potential energies */
 348         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 349
 350         /* Increment number of inner iterations */
 351         inneriter                  += j_index_end - j_index_start;
 352
 353         /* Outer loop uses 8 flops */
 354     }
 355
 356     /* Increment number of outer iterations */
 357     outeriter        += nri;
 358
 359     /* Update outer/inner flops */
 360
 361     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*34);
 362 }
 363 /*
 364  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_single
 365  * Electrostatics interaction: Ewald
 366  * VdW interaction:            None
 367  * Geometry:                   Particle-Particle
 368  * Calculate force/pot:        Force
 369  */
 370 void
 371 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_avx_128_fma_single
 372                     (t_nblist * gmx_restrict                nlist,
 373                      rvec * gmx_restrict                    xx,
 374                      rvec * gmx_restrict                    ff,
 375                      t_forcerec * gmx_restrict              fr,
 376                      t_mdatoms * gmx_restrict               mdatoms,
 377                      nb_kernel_data_t * gmx_restrict        kernel_data,
 378                      t_nrnb * gmx_restrict                  nrnb)
 379 {
 380     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 381      * just 0 for non-waters.
 382      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 383      * jnr indices corresponding to data put in the four positions in the SIMD register.
 384      */
 385     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 386     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 387     int              jnrA,jnrB,jnrC,jnrD;
 388     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 389     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 390     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 391     real             rcutoff_scalar;
 392     real             *shiftvec,*fshift,*x,*f;
 393     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 394     real             scratch[4*DIM];
 395     __m128           fscal,rcutoff,rcutoff2,jidxall;
 396     int              vdwioffset0;
 397     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 398     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 399     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 400     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 401     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 402     real             *charge;
 403     __m128i          ewitab;
 404     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 405     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 406     real             *ewtab;
 407     __m128           dummy_mask,cutoff_mask;
 408     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 409     __m128           one     = _mm_set1_ps(1.0);
 410     __m128           two     = _mm_set1_ps(2.0);
 411     x                = xx[0];
 412     f                = ff[0];
 413
 414     nri              = nlist->nri;
 415     iinr             = nlist->iinr;
 416     jindex           = nlist->jindex;
 417     jjnr             = nlist->jjnr;
 418     shiftidx         = nlist->shift;
 419     gid              = nlist->gid;
 420     shiftvec         = fr->shift_vec[0];
 421     fshift           = fr->fshift[0];
 422     facel            = _mm_set1_ps(fr->epsfac);
 423     charge           = mdatoms->chargeA;
 424
 425     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 426     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 427     beta2            = _mm_mul_ps(beta,beta);
 428     beta3            = _mm_mul_ps(beta,beta2);
 429     ewtab            = fr->ic->tabq_coul_F;
 430     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 431     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 432
 433     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 434     rcutoff_scalar   = fr->rcoulomb;
 435     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 436     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 437
 438     /* Avoid stupid compiler warnings */
 439     jnrA = jnrB = jnrC = jnrD = 0;
 440     j_coord_offsetA = 0;
 441     j_coord_offsetB = 0;
 442     j_coord_offsetC = 0;
 443     j_coord_offsetD = 0;
 444
 445     outeriter        = 0;
 446     inneriter        = 0;
 447
 448     for(iidx=0;iidx<4*DIM;iidx++)
 449     {
 450         scratch[iidx] = 0.0;
 451     }
 452
 453     /* Start outer loop over neighborlists */
 454     for(iidx=0; iidx<nri; iidx++)
 455     {
 456         /* Load shift vector for this list */
 457         i_shift_offset   = DIM*shiftidx[iidx];
 458
 459         /* Load limits for loop over neighbors */
 460         j_index_start    = jindex[iidx];
 461         j_index_end      = jindex[iidx+1];
 462
 463         /* Get outer coordinate index */
 464         inr              = iinr[iidx];
 465         i_coord_offset   = DIM*inr;
 466
 467         /* Load i particle coords and add shift vector */
 468         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 469
 470         fix0             = _mm_setzero_ps();
 471         fiy0             = _mm_setzero_ps();
 472         fiz0             = _mm_setzero_ps();
 473
 474         /* Load parameters for i particles */
 475         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 476
 477         /* Start inner kernel loop */
 478         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 479         {
 480
 481             /* Get j neighbor index, and coordinate index */
 482             jnrA             = jjnr[jidx];
 483             jnrB             = jjnr[jidx+1];
 484             jnrC             = jjnr[jidx+2];
 485             jnrD             = jjnr[jidx+3];
 486             j_coord_offsetA  = DIM*jnrA;
 487             j_coord_offsetB  = DIM*jnrB;
 488             j_coord_offsetC  = DIM*jnrC;
 489             j_coord_offsetD  = DIM*jnrD;
 490
 491             /* load j atom coordinates */
 492             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 493                                               x+j_coord_offsetC,x+j_coord_offsetD,
 494                                               &jx0,&jy0,&jz0);
 495
 496             /* Calculate displacement vector */
 497             dx00             = _mm_sub_ps(ix0,jx0);
 498             dy00             = _mm_sub_ps(iy0,jy0);
 499             dz00             = _mm_sub_ps(iz0,jz0);
 500
 501             /* Calculate squared distance and things based on it */
 502             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 503
 504             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 505
 506             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 507
 508             /* Load parameters for j particles */
 509             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 510                                                               charge+jnrC+0,charge+jnrD+0);
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             if (gmx_mm_any_lt(rsq00,rcutoff2))
 517             {
 518
 519             r00              = _mm_mul_ps(rsq00,rinv00);
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq00             = _mm_mul_ps(iq0,jq0);
 523
 524             /* EWALD ELECTROSTATICS */
 525
 526             /* Analytical PME correction */
 527             zeta2            = _mm_mul_ps(beta2,rsq00);
 528             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 529             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 530             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 531             felec            = _mm_mul_ps(qq00,felec);
 532
 533             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm_and_ps(fscal,cutoff_mask);
 538
 539              /* Update vectorial force */
 540             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 541             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 542             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 543
 544             fjptrA             = f+j_coord_offsetA;
 545             fjptrB             = f+j_coord_offsetB;
 546             fjptrC             = f+j_coord_offsetC;
 547             fjptrD             = f+j_coord_offsetD;
 548             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 549                                                    _mm_mul_ps(dx00,fscal),
 550                                                    _mm_mul_ps(dy00,fscal),
 551                                                    _mm_mul_ps(dz00,fscal));
 552
 553             }
 554
 555             /* Inner loop uses 31 flops */
 556         }
 557
 558         if(jidx<j_index_end)
 559         {
 560
 561             /* Get j neighbor index, and coordinate index */
 562             jnrlistA         = jjnr[jidx];
 563             jnrlistB         = jjnr[jidx+1];
 564             jnrlistC         = jjnr[jidx+2];
 565             jnrlistD         = jjnr[jidx+3];
 566             /* Sign of each element will be negative for non-real atoms.
 567              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 568              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 569              */
 570             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 571             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 572             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 573             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 574             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 575             j_coord_offsetA  = DIM*jnrA;
 576             j_coord_offsetB  = DIM*jnrB;
 577             j_coord_offsetC  = DIM*jnrC;
 578             j_coord_offsetD  = DIM*jnrD;
 579
 580             /* load j atom coordinates */
 581             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 582                                               x+j_coord_offsetC,x+j_coord_offsetD,
 583                                               &jx0,&jy0,&jz0);
 584
 585             /* Calculate displacement vector */
 586             dx00             = _mm_sub_ps(ix0,jx0);
 587             dy00             = _mm_sub_ps(iy0,jy0);
 588             dz00             = _mm_sub_ps(iz0,jz0);
 589
 590             /* Calculate squared distance and things based on it */
 591             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 592
 593             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 594
 595             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 596
 597             /* Load parameters for j particles */
 598             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 599                                                               charge+jnrC+0,charge+jnrD+0);
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             if (gmx_mm_any_lt(rsq00,rcutoff2))
 606             {
 607
 608             r00              = _mm_mul_ps(rsq00,rinv00);
 609             r00              = _mm_andnot_ps(dummy_mask,r00);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq00             = _mm_mul_ps(iq0,jq0);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Analytical PME correction */
 617             zeta2            = _mm_mul_ps(beta2,rsq00);
 618             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 619             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 620             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 621             felec            = _mm_mul_ps(qq00,felec);
 622
 623             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm_and_ps(fscal,cutoff_mask);
 628
 629             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 630
 631              /* Update vectorial force */
 632             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 633             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 634             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 635
 636             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 637             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 638             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 639             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 640             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 641                                                    _mm_mul_ps(dx00,fscal),
 642                                                    _mm_mul_ps(dy00,fscal),
 643                                                    _mm_mul_ps(dz00,fscal));
 644
 645             }
 646
 647             /* Inner loop uses 32 flops */
 648         }
 649
 650         /* End of innermost loop */
 651
 652         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 653                                               f+i_coord_offset,fshift+i_shift_offset);
 654
 655         /* Increment number of inner iterations */
 656         inneriter                  += j_index_end - j_index_start;
 657
 658         /* Outer loop uses 7 flops */
 659     }
 660
 661     /* Increment number of outer iterations */
 662     outeriter        += nri;
 663
 664     /* Update outer/inner flops */
 665
 666     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*32);
 667 }