src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwNone_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_VF_sse2_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_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  69     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     __m128i          ewitab;
  74     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  75     real             *ewtab;
  76     __m128           dummy_mask,cutoff_mask;
  77     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  78     __m128           one     = _mm_set1_ps(1.0);
  79     __m128           two     = _mm_set1_ps(2.0);
  80     x                = xx[0];
  81     f                = ff[0];
  82
  83     nri              = nlist->nri;
  84     iinr             = nlist->iinr;
  85     jindex           = nlist->jindex;
  86     jjnr             = nlist->jjnr;
  87     shiftidx         = nlist->shift;
  88     gid              = nlist->gid;
  89     shiftvec         = fr->shift_vec[0];
  90     fshift           = fr->fshift[0];
  91     facel            = _mm_set1_ps(fr->epsfac);
  92     charge           = mdatoms->chargeA;
  93
  94     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
  95     ewtab            = fr->ic->tabq_coul_FDV0;
  96     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
  97     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
  98
  99     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 100     rcutoff_scalar   = fr->rcoulomb;
 101     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 102     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 103
 104     /* Avoid stupid compiler warnings */
 105     jnrA = jnrB = jnrC = jnrD = 0;
 106     j_coord_offsetA = 0;
 107     j_coord_offsetB = 0;
 108     j_coord_offsetC = 0;
 109     j_coord_offsetD = 0;
 110
 111     outeriter        = 0;
 112     inneriter        = 0;
 113
 114     /* Start outer loop over neighborlists */
 115     for(iidx=0; iidx<nri; iidx++)
 116     {
 117         /* Load shift vector for this list */
 118         i_shift_offset   = DIM*shiftidx[iidx];
 119         shX              = shiftvec[i_shift_offset+XX];
 120         shY              = shiftvec[i_shift_offset+YY];
 121         shZ              = shiftvec[i_shift_offset+ZZ];
 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         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 133         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 134         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 135
 136         fix0             = _mm_setzero_ps();
 137         fiy0             = _mm_setzero_ps();
 138         fiz0             = _mm_setzero_ps();
 139
 140         /* Load parameters for i particles */
 141         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 142
 143         /* Reset potential sums */
 144         velecsum         = _mm_setzero_ps();
 145
 146         /* Start inner kernel loop */
 147         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 148         {
 149
 150             /* Get j neighbor index, and coordinate index */
 151             jnrA             = jjnr[jidx];
 152             jnrB             = jjnr[jidx+1];
 153             jnrC             = jjnr[jidx+2];
 154             jnrD             = jjnr[jidx+3];
 155
 156             j_coord_offsetA  = DIM*jnrA;
 157             j_coord_offsetB  = DIM*jnrB;
 158             j_coord_offsetC  = DIM*jnrC;
 159             j_coord_offsetD  = DIM*jnrD;
 160
 161             /* load j atom coordinates */
 162             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 163                                               x+j_coord_offsetC,x+j_coord_offsetD,
 164                                               &jx0,&jy0,&jz0);
 165
 166             /* Calculate displacement vector */
 167             dx00             = _mm_sub_ps(ix0,jx0);
 168             dy00             = _mm_sub_ps(iy0,jy0);
 169             dz00             = _mm_sub_ps(iz0,jz0);
 170
 171             /* Calculate squared distance and things based on it */
 172             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 173
 174             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 175
 176             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 177
 178             /* Load parameters for j particles */
 179             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 180                                                               charge+jnrC+0,charge+jnrD+0);
 181
 182             /**************************
 183              * CALCULATE INTERACTIONS *
 184              **************************/
 185
 186             if (gmx_mm_any_lt(rsq00,rcutoff2))
 187             {
 188
 189             r00              = _mm_mul_ps(rsq00,rinv00);
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm_mul_ps(iq0,jq0);
 193
 194             /* EWALD ELECTROSTATICS */
 195
 196             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 197             ewrt             = _mm_mul_ps(r00,ewtabscale);
 198             ewitab           = _mm_cvttps_epi32(ewrt);
 199             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 200             ewitab           = _mm_slli_epi32(ewitab,2);
 201             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 202             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 203             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 204             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 205             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 206             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 207             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 208             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 209             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 210
 211             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             velec            = _mm_and_ps(velec,cutoff_mask);
 215             velecsum         = _mm_add_ps(velecsum,velec);
 216
 217             fscal            = felec;
 218
 219             fscal            = _mm_and_ps(fscal,cutoff_mask);
 220
 221             /* Calculate temporary vectorial force */
 222             tx               = _mm_mul_ps(fscal,dx00);
 223             ty               = _mm_mul_ps(fscal,dy00);
 224             tz               = _mm_mul_ps(fscal,dz00);
 225
 226             /* Update vectorial force */
 227             fix0             = _mm_add_ps(fix0,tx);
 228             fiy0             = _mm_add_ps(fiy0,ty);
 229             fiz0             = _mm_add_ps(fiz0,tz);
 230
 231             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 232                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 233                                                    tx,ty,tz);
 234
 235             }
 236
 237             /* Inner loop uses 46 flops */
 238         }
 239
 240         if(jidx<j_index_end)
 241         {
 242
 243             /* Get j neighbor index, and coordinate index */
 244             jnrA             = jjnr[jidx];
 245             jnrB             = jjnr[jidx+1];
 246             jnrC             = jjnr[jidx+2];
 247             jnrD             = jjnr[jidx+3];
 248
 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       = (jnrA>=0) ? jnrA : 0;
 255             jnrB       = (jnrB>=0) ? jnrB : 0;
 256             jnrC       = (jnrC>=0) ? jnrC : 0;
 257             jnrD       = (jnrD>=0) ? jnrD : 0;
 258
 259             j_coord_offsetA  = DIM*jnrA;
 260             j_coord_offsetB  = DIM*jnrB;
 261             j_coord_offsetC  = DIM*jnrC;
 262             j_coord_offsetD  = DIM*jnrD;
 263
 264             /* load j atom coordinates */
 265             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 266                                               x+j_coord_offsetC,x+j_coord_offsetD,
 267                                               &jx0,&jy0,&jz0);
 268
 269             /* Calculate displacement vector */
 270             dx00             = _mm_sub_ps(ix0,jx0);
 271             dy00             = _mm_sub_ps(iy0,jy0);
 272             dz00             = _mm_sub_ps(iz0,jz0);
 273
 274             /* Calculate squared distance and things based on it */
 275             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 276
 277             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 278
 279             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 280
 281             /* Load parameters for j particles */
 282             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 283                                                               charge+jnrC+0,charge+jnrD+0);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             if (gmx_mm_any_lt(rsq00,rcutoff2))
 290             {
 291
 292             r00              = _mm_mul_ps(rsq00,rinv00);
 293             r00              = _mm_andnot_ps(dummy_mask,r00);
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             qq00             = _mm_mul_ps(iq0,jq0);
 297
 298             /* EWALD ELECTROSTATICS */
 299
 300             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 301             ewrt             = _mm_mul_ps(r00,ewtabscale);
 302             ewitab           = _mm_cvttps_epi32(ewrt);
 303             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 304             ewitab           = _mm_slli_epi32(ewitab,2);
 305             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 306             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 307             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 308             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 309             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 310             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 311             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 312             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 313             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 314
 315             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velec            = _mm_and_ps(velec,cutoff_mask);
 319             velec            = _mm_andnot_ps(dummy_mask,velec);
 320             velecsum         = _mm_add_ps(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             fscal            = _mm_and_ps(fscal,cutoff_mask);
 325
 326             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 327
 328             /* Calculate temporary vectorial force */
 329             tx               = _mm_mul_ps(fscal,dx00);
 330             ty               = _mm_mul_ps(fscal,dy00);
 331             tz               = _mm_mul_ps(fscal,dz00);
 332
 333             /* Update vectorial force */
 334             fix0             = _mm_add_ps(fix0,tx);
 335             fiy0             = _mm_add_ps(fiy0,ty);
 336             fiz0             = _mm_add_ps(fiz0,tz);
 337
 338             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 339                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 340                                                    tx,ty,tz);
 341
 342             }
 343
 344             /* Inner loop uses 47 flops */
 345         }
 346
 347         /* End of innermost loop */
 348
 349         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 350                                               f+i_coord_offset,fshift+i_shift_offset);
 351
 352         ggid                        = gid[iidx];
 353         /* Update potential energies */
 354         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 355
 356         /* Increment number of inner iterations */
 357         inneriter                  += j_index_end - j_index_start;
 358
 359         /* Outer loop uses 11 flops */
 360     }
 361
 362     /* Increment number of outer iterations */
 363     outeriter        += nri;
 364
 365     /* Update outer/inner flops */
 366
 367     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*11 + inneriter*47);
 368 }
 369 /*
 370  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_single
 371  * Electrostatics interaction: Ewald
 372  * VdW interaction:            None
 373  * Geometry:                   Particle-Particle
 374  * Calculate force/pot:        Force
 375  */
 376 void
 377 nb_kernel_ElecEwSh_VdwNone_GeomP1P1_F_sse2_single
 378                     (t_nblist * gmx_restrict                nlist,
 379                      rvec * gmx_restrict                    xx,
 380                      rvec * gmx_restrict                    ff,
 381                      t_forcerec * gmx_restrict              fr,
 382                      t_mdatoms * gmx_restrict               mdatoms,
 383                      nb_kernel_data_t * gmx_restrict        kernel_data,
 384                      t_nrnb * gmx_restrict                  nrnb)
 385 {
 386     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 387      * just 0 for non-waters.
 388      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 389      * jnr indices corresponding to data put in the four positions in the SIMD register.
 390      */
 391     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 392     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 393     int              jnrA,jnrB,jnrC,jnrD;
 394     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 395     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 396     real             shX,shY,shZ,rcutoff_scalar;
 397     real             *shiftvec,*fshift,*x,*f;
 398     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 399     int              vdwioffset0;
 400     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 401     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 402     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 403     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 404     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 405     real             *charge;
 406     __m128i          ewitab;
 407     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 408     real             *ewtab;
 409     __m128           dummy_mask,cutoff_mask;
 410     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 411     __m128           one     = _mm_set1_ps(1.0);
 412     __m128           two     = _mm_set1_ps(2.0);
 413     x                = xx[0];
 414     f                = ff[0];
 415
 416     nri              = nlist->nri;
 417     iinr             = nlist->iinr;
 418     jindex           = nlist->jindex;
 419     jjnr             = nlist->jjnr;
 420     shiftidx         = nlist->shift;
 421     gid              = nlist->gid;
 422     shiftvec         = fr->shift_vec[0];
 423     fshift           = fr->fshift[0];
 424     facel            = _mm_set1_ps(fr->epsfac);
 425     charge           = mdatoms->chargeA;
 426
 427     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 428     ewtab            = fr->ic->tabq_coul_F;
 429     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 430     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 431
 432     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 433     rcutoff_scalar   = fr->rcoulomb;
 434     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 435     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 436
 437     /* Avoid stupid compiler warnings */
 438     jnrA = jnrB = jnrC = jnrD = 0;
 439     j_coord_offsetA = 0;
 440     j_coord_offsetB = 0;
 441     j_coord_offsetC = 0;
 442     j_coord_offsetD = 0;
 443
 444     outeriter        = 0;
 445     inneriter        = 0;
 446
 447     /* Start outer loop over neighborlists */
 448     for(iidx=0; iidx<nri; iidx++)
 449     {
 450         /* Load shift vector for this list */
 451         i_shift_offset   = DIM*shiftidx[iidx];
 452         shX              = shiftvec[i_shift_offset+XX];
 453         shY              = shiftvec[i_shift_offset+YY];
 454         shZ              = shiftvec[i_shift_offset+ZZ];
 455
 456         /* Load limits for loop over neighbors */
 457         j_index_start    = jindex[iidx];
 458         j_index_end      = jindex[iidx+1];
 459
 460         /* Get outer coordinate index */
 461         inr              = iinr[iidx];
 462         i_coord_offset   = DIM*inr;
 463
 464         /* Load i particle coords and add shift vector */
 465         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 466         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 467         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 468
 469         fix0             = _mm_setzero_ps();
 470         fiy0             = _mm_setzero_ps();
 471         fiz0             = _mm_setzero_ps();
 472
 473         /* Load parameters for i particles */
 474         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 475
 476         /* Start inner kernel loop */
 477         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 478         {
 479
 480             /* Get j neighbor index, and coordinate index */
 481             jnrA             = jjnr[jidx];
 482             jnrB             = jjnr[jidx+1];
 483             jnrC             = jjnr[jidx+2];
 484             jnrD             = jjnr[jidx+3];
 485
 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             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 527             ewrt             = _mm_mul_ps(r00,ewtabscale);
 528             ewitab           = _mm_cvttps_epi32(ewrt);
 529             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 530             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 531                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 532                                          &ewtabF,&ewtabFn);
 533             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 534             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 535
 536             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 537
 538             fscal            = felec;
 539
 540             fscal            = _mm_and_ps(fscal,cutoff_mask);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_ps(fscal,dx00);
 544             ty               = _mm_mul_ps(fscal,dy00);
 545             tz               = _mm_mul_ps(fscal,dz00);
 546
 547             /* Update vectorial force */
 548             fix0             = _mm_add_ps(fix0,tx);
 549             fiy0             = _mm_add_ps(fiy0,ty);
 550             fiz0             = _mm_add_ps(fiz0,tz);
 551
 552             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 553                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 554                                                    tx,ty,tz);
 555
 556             }
 557
 558             /* Inner loop uses 39 flops */
 559         }
 560
 561         if(jidx<j_index_end)
 562         {
 563
 564             /* Get j neighbor index, and coordinate index */
 565             jnrA             = jjnr[jidx];
 566             jnrB             = jjnr[jidx+1];
 567             jnrC             = jjnr[jidx+2];
 568             jnrD             = jjnr[jidx+3];
 569
 570             /* Sign of each element will be negative for non-real atoms.
 571              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 572              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 573              */
 574             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 575             jnrA       = (jnrA>=0) ? jnrA : 0;
 576             jnrB       = (jnrB>=0) ? jnrB : 0;
 577             jnrC       = (jnrC>=0) ? jnrC : 0;
 578             jnrD       = (jnrD>=0) ? jnrD : 0;
 579
 580             j_coord_offsetA  = DIM*jnrA;
 581             j_coord_offsetB  = DIM*jnrB;
 582             j_coord_offsetC  = DIM*jnrC;
 583             j_coord_offsetD  = DIM*jnrD;
 584
 585             /* load j atom coordinates */
 586             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 587                                               x+j_coord_offsetC,x+j_coord_offsetD,
 588                                               &jx0,&jy0,&jz0);
 589
 590             /* Calculate displacement vector */
 591             dx00             = _mm_sub_ps(ix0,jx0);
 592             dy00             = _mm_sub_ps(iy0,jy0);
 593             dz00             = _mm_sub_ps(iz0,jz0);
 594
 595             /* Calculate squared distance and things based on it */
 596             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 597
 598             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 599
 600             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 601
 602             /* Load parameters for j particles */
 603             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 604                                                               charge+jnrC+0,charge+jnrD+0);
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             if (gmx_mm_any_lt(rsq00,rcutoff2))
 611             {
 612
 613             r00              = _mm_mul_ps(rsq00,rinv00);
 614             r00              = _mm_andnot_ps(dummy_mask,r00);
 615
 616             /* Compute parameters for interactions between i and j atoms */
 617             qq00             = _mm_mul_ps(iq0,jq0);
 618
 619             /* EWALD ELECTROSTATICS */
 620
 621             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 622             ewrt             = _mm_mul_ps(r00,ewtabscale);
 623             ewitab           = _mm_cvttps_epi32(ewrt);
 624             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 625             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 626                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 627                                          &ewtabF,&ewtabFn);
 628             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 629             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 630
 631             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 632
 633             fscal            = felec;
 634
 635             fscal            = _mm_and_ps(fscal,cutoff_mask);
 636
 637             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm_mul_ps(fscal,dx00);
 641             ty               = _mm_mul_ps(fscal,dy00);
 642             tz               = _mm_mul_ps(fscal,dz00);
 643
 644             /* Update vectorial force */
 645             fix0             = _mm_add_ps(fix0,tx);
 646             fiy0             = _mm_add_ps(fiy0,ty);
 647             fiz0             = _mm_add_ps(fiz0,tz);
 648
 649             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 650                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 651                                                    tx,ty,tz);
 652
 653             }
 654
 655             /* Inner loop uses 40 flops */
 656         }
 657
 658         /* End of innermost loop */
 659
 660         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 661                                               f+i_coord_offset,fshift+i_shift_offset);
 662
 663         /* Increment number of inner iterations */
 664         inneriter                  += j_index_end - j_index_start;
 665
 666         /* Outer loop uses 10 flops */
 667     }
 668
 669     /* Increment number of outer iterations */
 670     outeriter        += nri;
 671
 672     /* Update outer/inner flops */
 673
 674     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*10 + inneriter*40);
 675 }