src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_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_ElecEw_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_ElecEw_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     /* Avoid stupid compiler warnings */
 100     jnrA = jnrB = jnrC = jnrD = 0;
 101     j_coord_offsetA = 0;
 102     j_coord_offsetB = 0;
 103     j_coord_offsetC = 0;
 104     j_coord_offsetD = 0;
 105
 106     outeriter        = 0;
 107     inneriter        = 0;
 108
 109     /* Start outer loop over neighborlists */
 110     for(iidx=0; iidx<nri; iidx++)
 111     {
 112         /* Load shift vector for this list */
 113         i_shift_offset   = DIM*shiftidx[iidx];
 114         shX              = shiftvec[i_shift_offset+XX];
 115         shY              = shiftvec[i_shift_offset+YY];
 116         shZ              = shiftvec[i_shift_offset+ZZ];
 117
 118         /* Load limits for loop over neighbors */
 119         j_index_start    = jindex[iidx];
 120         j_index_end      = jindex[iidx+1];
 121
 122         /* Get outer coordinate index */
 123         inr              = iinr[iidx];
 124         i_coord_offset   = DIM*inr;
 125
 126         /* Load i particle coords and add shift vector */
 127         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 128         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 129         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 130
 131         fix0             = _mm_setzero_ps();
 132         fiy0             = _mm_setzero_ps();
 133         fiz0             = _mm_setzero_ps();
 134
 135         /* Load parameters for i particles */
 136         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 137
 138         /* Reset potential sums */
 139         velecsum         = _mm_setzero_ps();
 140
 141         /* Start inner kernel loop */
 142         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 143         {
 144
 145             /* Get j neighbor index, and coordinate index */
 146             jnrA             = jjnr[jidx];
 147             jnrB             = jjnr[jidx+1];
 148             jnrC             = jjnr[jidx+2];
 149             jnrD             = jjnr[jidx+3];
 150
 151             j_coord_offsetA  = DIM*jnrA;
 152             j_coord_offsetB  = DIM*jnrB;
 153             j_coord_offsetC  = DIM*jnrC;
 154             j_coord_offsetD  = DIM*jnrD;
 155
 156             /* load j atom coordinates */
 157             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 158                                               x+j_coord_offsetC,x+j_coord_offsetD,
 159                                               &jx0,&jy0,&jz0);
 160
 161             /* Calculate displacement vector */
 162             dx00             = _mm_sub_ps(ix0,jx0);
 163             dy00             = _mm_sub_ps(iy0,jy0);
 164             dz00             = _mm_sub_ps(iz0,jz0);
 165
 166             /* Calculate squared distance and things based on it */
 167             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 168
 169             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 170
 171             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 172
 173             /* Load parameters for j particles */
 174             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 175                                                               charge+jnrC+0,charge+jnrD+0);
 176
 177             /**************************
 178              * CALCULATE INTERACTIONS *
 179              **************************/
 180
 181             r00              = _mm_mul_ps(rsq00,rinv00);
 182
 183             /* Compute parameters for interactions between i and j atoms */
 184             qq00             = _mm_mul_ps(iq0,jq0);
 185
 186             /* EWALD ELECTROSTATICS */
 187
 188             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 189             ewrt             = _mm_mul_ps(r00,ewtabscale);
 190             ewitab           = _mm_cvttps_epi32(ewrt);
 191             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 192             ewitab           = _mm_slli_epi32(ewitab,2);
 193             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 194             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 195             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 196             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 197             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 198             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 199             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 200             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 201             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 202
 203             /* Update potential sum for this i atom from the interaction with this j atom. */
 204             velecsum         = _mm_add_ps(velecsum,velec);
 205
 206             fscal            = felec;
 207
 208             /* Calculate temporary vectorial force */
 209             tx               = _mm_mul_ps(fscal,dx00);
 210             ty               = _mm_mul_ps(fscal,dy00);
 211             tz               = _mm_mul_ps(fscal,dz00);
 212
 213             /* Update vectorial force */
 214             fix0             = _mm_add_ps(fix0,tx);
 215             fiy0             = _mm_add_ps(fiy0,ty);
 216             fiz0             = _mm_add_ps(fiz0,tz);
 217
 218             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 219                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 220                                                    tx,ty,tz);
 221
 222             /* Inner loop uses 41 flops */
 223         }
 224
 225         if(jidx<j_index_end)
 226         {
 227
 228             /* Get j neighbor index, and coordinate index */
 229             jnrA             = jjnr[jidx];
 230             jnrB             = jjnr[jidx+1];
 231             jnrC             = jjnr[jidx+2];
 232             jnrD             = jjnr[jidx+3];
 233
 234             /* Sign of each element will be negative for non-real atoms.
 235              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 236              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 237              */
 238             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 239             jnrA       = (jnrA>=0) ? jnrA : 0;
 240             jnrB       = (jnrB>=0) ? jnrB : 0;
 241             jnrC       = (jnrC>=0) ? jnrC : 0;
 242             jnrD       = (jnrD>=0) ? jnrD : 0;
 243
 244             j_coord_offsetA  = DIM*jnrA;
 245             j_coord_offsetB  = DIM*jnrB;
 246             j_coord_offsetC  = DIM*jnrC;
 247             j_coord_offsetD  = DIM*jnrD;
 248
 249             /* load j atom coordinates */
 250             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 251                                               x+j_coord_offsetC,x+j_coord_offsetD,
 252                                               &jx0,&jy0,&jz0);
 253
 254             /* Calculate displacement vector */
 255             dx00             = _mm_sub_ps(ix0,jx0);
 256             dy00             = _mm_sub_ps(iy0,jy0);
 257             dz00             = _mm_sub_ps(iz0,jz0);
 258
 259             /* Calculate squared distance and things based on it */
 260             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 261
 262             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 263
 264             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 265
 266             /* Load parameters for j particles */
 267             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 268                                                               charge+jnrC+0,charge+jnrD+0);
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r00              = _mm_mul_ps(rsq00,rinv00);
 275             r00              = _mm_andnot_ps(dummy_mask,r00);
 276
 277             /* Compute parameters for interactions between i and j atoms */
 278             qq00             = _mm_mul_ps(iq0,jq0);
 279
 280             /* EWALD ELECTROSTATICS */
 281
 282             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 283             ewrt             = _mm_mul_ps(r00,ewtabscale);
 284             ewitab           = _mm_cvttps_epi32(ewrt);
 285             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 286             ewitab           = _mm_slli_epi32(ewitab,2);
 287             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 288             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 289             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 290             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 291             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 292             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 293             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 294             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 295             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velec            = _mm_andnot_ps(dummy_mask,velec);
 299             velecsum         = _mm_add_ps(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm_mul_ps(fscal,dx00);
 307             ty               = _mm_mul_ps(fscal,dy00);
 308             tz               = _mm_mul_ps(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm_add_ps(fix0,tx);
 312             fiy0             = _mm_add_ps(fiy0,ty);
 313             fiz0             = _mm_add_ps(fiz0,tz);
 314
 315             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 316                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 317                                                    tx,ty,tz);
 318
 319             /* Inner loop uses 42 flops */
 320         }
 321
 322         /* End of innermost loop */
 323
 324         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 325                                               f+i_coord_offset,fshift+i_shift_offset);
 326
 327         ggid                        = gid[iidx];
 328         /* Update potential energies */
 329         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 330
 331         /* Increment number of inner iterations */
 332         inneriter                  += j_index_end - j_index_start;
 333
 334         /* Outer loop uses 11 flops */
 335     }
 336
 337     /* Increment number of outer iterations */
 338     outeriter        += nri;
 339
 340     /* Update outer/inner flops */
 341
 342     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*11 + inneriter*42);
 343 }
 344 /*
 345  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse2_single
 346  * Electrostatics interaction: Ewald
 347  * VdW interaction:            None
 348  * Geometry:                   Particle-Particle
 349  * Calculate force/pot:        Force
 350  */
 351 void
 352 nb_kernel_ElecEw_VdwNone_GeomP1P1_F_sse2_single
 353                     (t_nblist * gmx_restrict                nlist,
 354                      rvec * gmx_restrict                    xx,
 355                      rvec * gmx_restrict                    ff,
 356                      t_forcerec * gmx_restrict              fr,
 357                      t_mdatoms * gmx_restrict               mdatoms,
 358                      nb_kernel_data_t * gmx_restrict        kernel_data,
 359                      t_nrnb * gmx_restrict                  nrnb)
 360 {
 361     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 362      * just 0 for non-waters.
 363      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 364      * jnr indices corresponding to data put in the four positions in the SIMD register.
 365      */
 366     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 367     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 368     int              jnrA,jnrB,jnrC,jnrD;
 369     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 370     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 371     real             shX,shY,shZ,rcutoff_scalar;
 372     real             *shiftvec,*fshift,*x,*f;
 373     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 374     int              vdwioffset0;
 375     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 376     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 377     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 378     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 379     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 380     real             *charge;
 381     __m128i          ewitab;
 382     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 383     real             *ewtab;
 384     __m128           dummy_mask,cutoff_mask;
 385     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 386     __m128           one     = _mm_set1_ps(1.0);
 387     __m128           two     = _mm_set1_ps(2.0);
 388     x                = xx[0];
 389     f                = ff[0];
 390
 391     nri              = nlist->nri;
 392     iinr             = nlist->iinr;
 393     jindex           = nlist->jindex;
 394     jjnr             = nlist->jjnr;
 395     shiftidx         = nlist->shift;
 396     gid              = nlist->gid;
 397     shiftvec         = fr->shift_vec[0];
 398     fshift           = fr->fshift[0];
 399     facel            = _mm_set1_ps(fr->epsfac);
 400     charge           = mdatoms->chargeA;
 401
 402     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 403     ewtab            = fr->ic->tabq_coul_F;
 404     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 405     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 406
 407     /* Avoid stupid compiler warnings */
 408     jnrA = jnrB = jnrC = jnrD = 0;
 409     j_coord_offsetA = 0;
 410     j_coord_offsetB = 0;
 411     j_coord_offsetC = 0;
 412     j_coord_offsetD = 0;
 413
 414     outeriter        = 0;
 415     inneriter        = 0;
 416
 417     /* Start outer loop over neighborlists */
 418     for(iidx=0; iidx<nri; iidx++)
 419     {
 420         /* Load shift vector for this list */
 421         i_shift_offset   = DIM*shiftidx[iidx];
 422         shX              = shiftvec[i_shift_offset+XX];
 423         shY              = shiftvec[i_shift_offset+YY];
 424         shZ              = shiftvec[i_shift_offset+ZZ];
 425
 426         /* Load limits for loop over neighbors */
 427         j_index_start    = jindex[iidx];
 428         j_index_end      = jindex[iidx+1];
 429
 430         /* Get outer coordinate index */
 431         inr              = iinr[iidx];
 432         i_coord_offset   = DIM*inr;
 433
 434         /* Load i particle coords and add shift vector */
 435         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 436         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 437         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 438
 439         fix0             = _mm_setzero_ps();
 440         fiy0             = _mm_setzero_ps();
 441         fiz0             = _mm_setzero_ps();
 442
 443         /* Load parameters for i particles */
 444         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 445
 446         /* Start inner kernel loop */
 447         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 448         {
 449
 450             /* Get j neighbor index, and coordinate index */
 451             jnrA             = jjnr[jidx];
 452             jnrB             = jjnr[jidx+1];
 453             jnrC             = jjnr[jidx+2];
 454             jnrD             = jjnr[jidx+3];
 455
 456             j_coord_offsetA  = DIM*jnrA;
 457             j_coord_offsetB  = DIM*jnrB;
 458             j_coord_offsetC  = DIM*jnrC;
 459             j_coord_offsetD  = DIM*jnrD;
 460
 461             /* load j atom coordinates */
 462             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 463                                               x+j_coord_offsetC,x+j_coord_offsetD,
 464                                               &jx0,&jy0,&jz0);
 465
 466             /* Calculate displacement vector */
 467             dx00             = _mm_sub_ps(ix0,jx0);
 468             dy00             = _mm_sub_ps(iy0,jy0);
 469             dz00             = _mm_sub_ps(iz0,jz0);
 470
 471             /* Calculate squared distance and things based on it */
 472             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 473
 474             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 475
 476             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 477
 478             /* Load parameters for j particles */
 479             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 480                                                               charge+jnrC+0,charge+jnrD+0);
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             r00              = _mm_mul_ps(rsq00,rinv00);
 487
 488             /* Compute parameters for interactions between i and j atoms */
 489             qq00             = _mm_mul_ps(iq0,jq0);
 490
 491             /* EWALD ELECTROSTATICS */
 492
 493             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 494             ewrt             = _mm_mul_ps(r00,ewtabscale);
 495             ewitab           = _mm_cvttps_epi32(ewrt);
 496             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 497             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 498                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 499                                          &ewtabF,&ewtabFn);
 500             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 501             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 502
 503             fscal            = felec;
 504
 505             /* Calculate temporary vectorial force */
 506             tx               = _mm_mul_ps(fscal,dx00);
 507             ty               = _mm_mul_ps(fscal,dy00);
 508             tz               = _mm_mul_ps(fscal,dz00);
 509
 510             /* Update vectorial force */
 511             fix0             = _mm_add_ps(fix0,tx);
 512             fiy0             = _mm_add_ps(fiy0,ty);
 513             fiz0             = _mm_add_ps(fiz0,tz);
 514
 515             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 516                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 517                                                    tx,ty,tz);
 518
 519             /* Inner loop uses 36 flops */
 520         }
 521
 522         if(jidx<j_index_end)
 523         {
 524
 525             /* Get j neighbor index, and coordinate index */
 526             jnrA             = jjnr[jidx];
 527             jnrB             = jjnr[jidx+1];
 528             jnrC             = jjnr[jidx+2];
 529             jnrD             = jjnr[jidx+3];
 530
 531             /* Sign of each element will be negative for non-real atoms.
 532              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 533              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 534              */
 535             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 536             jnrA       = (jnrA>=0) ? jnrA : 0;
 537             jnrB       = (jnrB>=0) ? jnrB : 0;
 538             jnrC       = (jnrC>=0) ? jnrC : 0;
 539             jnrD       = (jnrD>=0) ? jnrD : 0;
 540
 541             j_coord_offsetA  = DIM*jnrA;
 542             j_coord_offsetB  = DIM*jnrB;
 543             j_coord_offsetC  = DIM*jnrC;
 544             j_coord_offsetD  = DIM*jnrD;
 545
 546             /* load j atom coordinates */
 547             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 548                                               x+j_coord_offsetC,x+j_coord_offsetD,
 549                                               &jx0,&jy0,&jz0);
 550
 551             /* Calculate displacement vector */
 552             dx00             = _mm_sub_ps(ix0,jx0);
 553             dy00             = _mm_sub_ps(iy0,jy0);
 554             dz00             = _mm_sub_ps(iz0,jz0);
 555
 556             /* Calculate squared distance and things based on it */
 557             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 558
 559             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 560
 561             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 562
 563             /* Load parameters for j particles */
 564             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 565                                                               charge+jnrC+0,charge+jnrD+0);
 566
 567             /**************************
 568              * CALCULATE INTERACTIONS *
 569              **************************/
 570
 571             r00              = _mm_mul_ps(rsq00,rinv00);
 572             r00              = _mm_andnot_ps(dummy_mask,r00);
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             qq00             = _mm_mul_ps(iq0,jq0);
 576
 577             /* EWALD ELECTROSTATICS */
 578
 579             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 580             ewrt             = _mm_mul_ps(r00,ewtabscale);
 581             ewitab           = _mm_cvttps_epi32(ewrt);
 582             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 583             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 584                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 585                                          &ewtabF,&ewtabFn);
 586             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 587             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm_mul_ps(fscal,dx00);
 595             ty               = _mm_mul_ps(fscal,dy00);
 596             tz               = _mm_mul_ps(fscal,dz00);
 597
 598             /* Update vectorial force */
 599             fix0             = _mm_add_ps(fix0,tx);
 600             fiy0             = _mm_add_ps(fiy0,ty);
 601             fiz0             = _mm_add_ps(fiz0,tz);
 602
 603             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 604                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 605                                                    tx,ty,tz);
 606
 607             /* Inner loop uses 37 flops */
 608         }
 609
 610         /* End of innermost loop */
 611
 612         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 613                                               f+i_coord_offset,fshift+i_shift_offset);
 614
 615         /* Increment number of inner iterations */
 616         inneriter                  += j_index_end - j_index_start;
 617
 618         /* Outer loop uses 10 flops */
 619     }
 620
 621     /* Increment number of outer iterations */
 622     outeriter        += nri;
 623
 624     /* Update outer/inner flops */
 625
 626     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*10 + inneriter*37);
 627 }