src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwNone_GeomW4P1_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_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwNone_GeomW4P1_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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset1;
  70     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  71     int              vdwioffset2;
  72     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  73     int              vdwioffset3;
  74     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     __m128i          ewitab;
  83     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  84     real             *ewtab;
  85     __m128           dummy_mask,cutoff_mask;
  86     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  87     __m128           one     = _mm_set1_ps(1.0);
  88     __m128           two     = _mm_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102
 103     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 104     ewtab            = fr->ic->tabq_coul_FDV0;
 105     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 106     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 107
 108     /* Setup water-specific parameters */
 109     inr              = nlist->iinr[0];
 110     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 111     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 112     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = jnrC = jnrD = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118     j_coord_offsetC = 0;
 119     j_coord_offsetD = 0;
 120
 121     outeriter        = 0;
 122     inneriter        = 0;
 123
 124     for(iidx=0;iidx<4*DIM;iidx++)
 125     {
 126         scratch[iidx] = 0.0;
 127     }
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134
 135         /* Load limits for loop over neighbors */
 136         j_index_start    = jindex[iidx];
 137         j_index_end      = jindex[iidx+1];
 138
 139         /* Get outer coordinate index */
 140         inr              = iinr[iidx];
 141         i_coord_offset   = DIM*inr;
 142
 143         /* Load i particle coords and add shift vector */
 144         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 145                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 146
 147         fix1             = _mm_setzero_ps();
 148         fiy1             = _mm_setzero_ps();
 149         fiz1             = _mm_setzero_ps();
 150         fix2             = _mm_setzero_ps();
 151         fiy2             = _mm_setzero_ps();
 152         fiz2             = _mm_setzero_ps();
 153         fix3             = _mm_setzero_ps();
 154         fiy3             = _mm_setzero_ps();
 155         fiz3             = _mm_setzero_ps();
 156
 157         /* Reset potential sums */
 158         velecsum         = _mm_setzero_ps();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             jnrC             = jjnr[jidx+2];
 168             jnrD             = jjnr[jidx+3];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171             j_coord_offsetC  = DIM*jnrC;
 172             j_coord_offsetD  = DIM*jnrD;
 173
 174             /* load j atom coordinates */
 175             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               x+j_coord_offsetC,x+j_coord_offsetD,
 177                                               &jx0,&jy0,&jz0);
 178
 179             /* Calculate displacement vector */
 180             dx10             = _mm_sub_ps(ix1,jx0);
 181             dy10             = _mm_sub_ps(iy1,jy0);
 182             dz10             = _mm_sub_ps(iz1,jz0);
 183             dx20             = _mm_sub_ps(ix2,jx0);
 184             dy20             = _mm_sub_ps(iy2,jy0);
 185             dz20             = _mm_sub_ps(iz2,jz0);
 186             dx30             = _mm_sub_ps(ix3,jx0);
 187             dy30             = _mm_sub_ps(iy3,jy0);
 188             dz30             = _mm_sub_ps(iz3,jz0);
 189
 190             /* Calculate squared distance and things based on it */
 191             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 192             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 193             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 194
 195             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 196             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 197             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 198
 199             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 200             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 201             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 202
 203             /* Load parameters for j particles */
 204             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 205                                                               charge+jnrC+0,charge+jnrD+0);
 206
 207             fjx0             = _mm_setzero_ps();
 208             fjy0             = _mm_setzero_ps();
 209             fjz0             = _mm_setzero_ps();
 210
 211             /**************************
 212              * CALCULATE INTERACTIONS *
 213              **************************/
 214
 215             r10              = _mm_mul_ps(rsq10,rinv10);
 216
 217             /* Compute parameters for interactions between i and j atoms */
 218             qq10             = _mm_mul_ps(iq1,jq0);
 219
 220             /* EWALD ELECTROSTATICS */
 221
 222             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 223             ewrt             = _mm_mul_ps(r10,ewtabscale);
 224             ewitab           = _mm_cvttps_epi32(ewrt);
 225             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 226             ewitab           = _mm_slli_epi32(ewitab,2);
 227             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 228             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 229             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 230             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 231             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 232             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 233             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 234             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 235             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 236
 237             /* Update potential sum for this i atom from the interaction with this j atom. */
 238             velecsum         = _mm_add_ps(velecsum,velec);
 239
 240             fscal            = felec;
 241
 242             /* Calculate temporary vectorial force */
 243             tx               = _mm_mul_ps(fscal,dx10);
 244             ty               = _mm_mul_ps(fscal,dy10);
 245             tz               = _mm_mul_ps(fscal,dz10);
 246
 247             /* Update vectorial force */
 248             fix1             = _mm_add_ps(fix1,tx);
 249             fiy1             = _mm_add_ps(fiy1,ty);
 250             fiz1             = _mm_add_ps(fiz1,tz);
 251
 252             fjx0             = _mm_add_ps(fjx0,tx);
 253             fjy0             = _mm_add_ps(fjy0,ty);
 254             fjz0             = _mm_add_ps(fjz0,tz);
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             r20              = _mm_mul_ps(rsq20,rinv20);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             qq20             = _mm_mul_ps(iq2,jq0);
 264
 265             /* EWALD ELECTROSTATICS */
 266
 267             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 268             ewrt             = _mm_mul_ps(r20,ewtabscale);
 269             ewitab           = _mm_cvttps_epi32(ewrt);
 270             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 271             ewitab           = _mm_slli_epi32(ewitab,2);
 272             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 273             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 274             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 275             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 276             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 277             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 278             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 279             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 280             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 281
 282             /* Update potential sum for this i atom from the interaction with this j atom. */
 283             velecsum         = _mm_add_ps(velecsum,velec);
 284
 285             fscal            = felec;
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_ps(fscal,dx20);
 289             ty               = _mm_mul_ps(fscal,dy20);
 290             tz               = _mm_mul_ps(fscal,dz20);
 291
 292             /* Update vectorial force */
 293             fix2             = _mm_add_ps(fix2,tx);
 294             fiy2             = _mm_add_ps(fiy2,ty);
 295             fiz2             = _mm_add_ps(fiz2,tz);
 296
 297             fjx0             = _mm_add_ps(fjx0,tx);
 298             fjy0             = _mm_add_ps(fjy0,ty);
 299             fjz0             = _mm_add_ps(fjz0,tz);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             r30              = _mm_mul_ps(rsq30,rinv30);
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq30             = _mm_mul_ps(iq3,jq0);
 309
 310             /* EWALD ELECTROSTATICS */
 311
 312             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 313             ewrt             = _mm_mul_ps(r30,ewtabscale);
 314             ewitab           = _mm_cvttps_epi32(ewrt);
 315             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 316             ewitab           = _mm_slli_epi32(ewitab,2);
 317             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 318             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 319             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 320             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 321             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 322             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 323             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 324             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 325             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm_add_ps(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm_mul_ps(fscal,dx30);
 334             ty               = _mm_mul_ps(fscal,dy30);
 335             tz               = _mm_mul_ps(fscal,dz30);
 336
 337             /* Update vectorial force */
 338             fix3             = _mm_add_ps(fix3,tx);
 339             fiy3             = _mm_add_ps(fiy3,ty);
 340             fiz3             = _mm_add_ps(fiz3,tz);
 341
 342             fjx0             = _mm_add_ps(fjx0,tx);
 343             fjy0             = _mm_add_ps(fjy0,ty);
 344             fjz0             = _mm_add_ps(fjz0,tz);
 345
 346             fjptrA             = f+j_coord_offsetA;
 347             fjptrB             = f+j_coord_offsetB;
 348             fjptrC             = f+j_coord_offsetC;
 349             fjptrD             = f+j_coord_offsetD;
 350
 351             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 352
 353             /* Inner loop uses 123 flops */
 354         }
 355
 356         if(jidx<j_index_end)
 357         {
 358
 359             /* Get j neighbor index, and coordinate index */
 360             jnrlistA         = jjnr[jidx];
 361             jnrlistB         = jjnr[jidx+1];
 362             jnrlistC         = jjnr[jidx+2];
 363             jnrlistD         = jjnr[jidx+3];
 364             /* Sign of each element will be negative for non-real atoms.
 365              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 366              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 367              */
 368             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 369             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 370             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 371             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 372             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 373             j_coord_offsetA  = DIM*jnrA;
 374             j_coord_offsetB  = DIM*jnrB;
 375             j_coord_offsetC  = DIM*jnrC;
 376             j_coord_offsetD  = DIM*jnrD;
 377
 378             /* load j atom coordinates */
 379             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 380                                               x+j_coord_offsetC,x+j_coord_offsetD,
 381                                               &jx0,&jy0,&jz0);
 382
 383             /* Calculate displacement vector */
 384             dx10             = _mm_sub_ps(ix1,jx0);
 385             dy10             = _mm_sub_ps(iy1,jy0);
 386             dz10             = _mm_sub_ps(iz1,jz0);
 387             dx20             = _mm_sub_ps(ix2,jx0);
 388             dy20             = _mm_sub_ps(iy2,jy0);
 389             dz20             = _mm_sub_ps(iz2,jz0);
 390             dx30             = _mm_sub_ps(ix3,jx0);
 391             dy30             = _mm_sub_ps(iy3,jy0);
 392             dz30             = _mm_sub_ps(iz3,jz0);
 393
 394             /* Calculate squared distance and things based on it */
 395             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 396             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 397             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 398
 399             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 400             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 401             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 402
 403             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 404             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 405             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 406
 407             /* Load parameters for j particles */
 408             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 409                                                               charge+jnrC+0,charge+jnrD+0);
 410
 411             fjx0             = _mm_setzero_ps();
 412             fjy0             = _mm_setzero_ps();
 413             fjz0             = _mm_setzero_ps();
 414
 415             /**************************
 416              * CALCULATE INTERACTIONS *
 417              **************************/
 418
 419             r10              = _mm_mul_ps(rsq10,rinv10);
 420             r10              = _mm_andnot_ps(dummy_mask,r10);
 421
 422             /* Compute parameters for interactions between i and j atoms */
 423             qq10             = _mm_mul_ps(iq1,jq0);
 424
 425             /* EWALD ELECTROSTATICS */
 426
 427             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 428             ewrt             = _mm_mul_ps(r10,ewtabscale);
 429             ewitab           = _mm_cvttps_epi32(ewrt);
 430             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 431             ewitab           = _mm_slli_epi32(ewitab,2);
 432             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 433             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 434             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 435             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 436             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 437             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 438             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 439             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 440             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velec            = _mm_andnot_ps(dummy_mask,velec);
 444             velecsum         = _mm_add_ps(velecsum,velec);
 445
 446             fscal            = felec;
 447
 448             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 449
 450             /* Calculate temporary vectorial force */
 451             tx               = _mm_mul_ps(fscal,dx10);
 452             ty               = _mm_mul_ps(fscal,dy10);
 453             tz               = _mm_mul_ps(fscal,dz10);
 454
 455             /* Update vectorial force */
 456             fix1             = _mm_add_ps(fix1,tx);
 457             fiy1             = _mm_add_ps(fiy1,ty);
 458             fiz1             = _mm_add_ps(fiz1,tz);
 459
 460             fjx0             = _mm_add_ps(fjx0,tx);
 461             fjy0             = _mm_add_ps(fjy0,ty);
 462             fjz0             = _mm_add_ps(fjz0,tz);
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r20              = _mm_mul_ps(rsq20,rinv20);
 469             r20              = _mm_andnot_ps(dummy_mask,r20);
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq20             = _mm_mul_ps(iq2,jq0);
 473
 474             /* EWALD ELECTROSTATICS */
 475
 476             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 477             ewrt             = _mm_mul_ps(r20,ewtabscale);
 478             ewitab           = _mm_cvttps_epi32(ewrt);
 479             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 480             ewitab           = _mm_slli_epi32(ewitab,2);
 481             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 482             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 483             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 484             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 485             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 486             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 487             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 488             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 489             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 490
 491             /* Update potential sum for this i atom from the interaction with this j atom. */
 492             velec            = _mm_andnot_ps(dummy_mask,velec);
 493             velecsum         = _mm_add_ps(velecsum,velec);
 494
 495             fscal            = felec;
 496
 497             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 498
 499             /* Calculate temporary vectorial force */
 500             tx               = _mm_mul_ps(fscal,dx20);
 501             ty               = _mm_mul_ps(fscal,dy20);
 502             tz               = _mm_mul_ps(fscal,dz20);
 503
 504             /* Update vectorial force */
 505             fix2             = _mm_add_ps(fix2,tx);
 506             fiy2             = _mm_add_ps(fiy2,ty);
 507             fiz2             = _mm_add_ps(fiz2,tz);
 508
 509             fjx0             = _mm_add_ps(fjx0,tx);
 510             fjy0             = _mm_add_ps(fjy0,ty);
 511             fjz0             = _mm_add_ps(fjz0,tz);
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             r30              = _mm_mul_ps(rsq30,rinv30);
 518             r30              = _mm_andnot_ps(dummy_mask,r30);
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq30             = _mm_mul_ps(iq3,jq0);
 522
 523             /* EWALD ELECTROSTATICS */
 524
 525             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 526             ewrt             = _mm_mul_ps(r30,ewtabscale);
 527             ewitab           = _mm_cvttps_epi32(ewrt);
 528             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 529             ewitab           = _mm_slli_epi32(ewitab,2);
 530             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 531             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 532             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 533             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 534             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 535             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 536             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 537             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 538             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velec            = _mm_andnot_ps(dummy_mask,velec);
 542             velecsum         = _mm_add_ps(velecsum,velec);
 543
 544             fscal            = felec;
 545
 546             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm_mul_ps(fscal,dx30);
 550             ty               = _mm_mul_ps(fscal,dy30);
 551             tz               = _mm_mul_ps(fscal,dz30);
 552
 553             /* Update vectorial force */
 554             fix3             = _mm_add_ps(fix3,tx);
 555             fiy3             = _mm_add_ps(fiy3,ty);
 556             fiz3             = _mm_add_ps(fiz3,tz);
 557
 558             fjx0             = _mm_add_ps(fjx0,tx);
 559             fjy0             = _mm_add_ps(fjy0,ty);
 560             fjz0             = _mm_add_ps(fjz0,tz);
 561
 562             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 563             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 564             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 565             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 566
 567             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 568
 569             /* Inner loop uses 126 flops */
 570         }
 571
 572         /* End of innermost loop */
 573
 574         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 575                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 576
 577         ggid                        = gid[iidx];
 578         /* Update potential energies */
 579         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 580
 581         /* Increment number of inner iterations */
 582         inneriter                  += j_index_end - j_index_start;
 583
 584         /* Outer loop uses 19 flops */
 585     }
 586
 587     /* Increment number of outer iterations */
 588     outeriter        += nri;
 589
 590     /* Update outer/inner flops */
 591
 592     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*126);
 593 }
 594 /*
 595  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_single
 596  * Electrostatics interaction: Ewald
 597  * VdW interaction:            None
 598  * Geometry:                   Water4-Particle
 599  * Calculate force/pot:        Force
 600  */
 601 void
 602 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_single
 603                     (t_nblist * gmx_restrict                nlist,
 604                      rvec * gmx_restrict                    xx,
 605                      rvec * gmx_restrict                    ff,
 606                      t_forcerec * gmx_restrict              fr,
 607                      t_mdatoms * gmx_restrict               mdatoms,
 608                      nb_kernel_data_t * gmx_restrict        kernel_data,
 609                      t_nrnb * gmx_restrict                  nrnb)
 610 {
 611     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 612      * just 0 for non-waters.
 613      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 614      * jnr indices corresponding to data put in the four positions in the SIMD register.
 615      */
 616     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 617     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 618     int              jnrA,jnrB,jnrC,jnrD;
 619     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 620     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 621     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 622     real             rcutoff_scalar;
 623     real             *shiftvec,*fshift,*x,*f;
 624     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 625     real             scratch[4*DIM];
 626     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 627     int              vdwioffset1;
 628     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 629     int              vdwioffset2;
 630     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 631     int              vdwioffset3;
 632     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 633     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 634     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 635     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 636     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 637     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 638     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 639     real             *charge;
 640     __m128i          ewitab;
 641     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 642     real             *ewtab;
 643     __m128           dummy_mask,cutoff_mask;
 644     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 645     __m128           one     = _mm_set1_ps(1.0);
 646     __m128           two     = _mm_set1_ps(2.0);
 647     x                = xx[0];
 648     f                = ff[0];
 649
 650     nri              = nlist->nri;
 651     iinr             = nlist->iinr;
 652     jindex           = nlist->jindex;
 653     jjnr             = nlist->jjnr;
 654     shiftidx         = nlist->shift;
 655     gid              = nlist->gid;
 656     shiftvec         = fr->shift_vec[0];
 657     fshift           = fr->fshift[0];
 658     facel            = _mm_set1_ps(fr->epsfac);
 659     charge           = mdatoms->chargeA;
 660
 661     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 662     ewtab            = fr->ic->tabq_coul_F;
 663     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 664     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 665
 666     /* Setup water-specific parameters */
 667     inr              = nlist->iinr[0];
 668     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 669     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 670     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 671
 672     /* Avoid stupid compiler warnings */
 673     jnrA = jnrB = jnrC = jnrD = 0;
 674     j_coord_offsetA = 0;
 675     j_coord_offsetB = 0;
 676     j_coord_offsetC = 0;
 677     j_coord_offsetD = 0;
 678
 679     outeriter        = 0;
 680     inneriter        = 0;
 681
 682     for(iidx=0;iidx<4*DIM;iidx++)
 683     {
 684         scratch[iidx] = 0.0;
 685     }
 686
 687     /* Start outer loop over neighborlists */
 688     for(iidx=0; iidx<nri; iidx++)
 689     {
 690         /* Load shift vector for this list */
 691         i_shift_offset   = DIM*shiftidx[iidx];
 692
 693         /* Load limits for loop over neighbors */
 694         j_index_start    = jindex[iidx];
 695         j_index_end      = jindex[iidx+1];
 696
 697         /* Get outer coordinate index */
 698         inr              = iinr[iidx];
 699         i_coord_offset   = DIM*inr;
 700
 701         /* Load i particle coords and add shift vector */
 702         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 703                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 704
 705         fix1             = _mm_setzero_ps();
 706         fiy1             = _mm_setzero_ps();
 707         fiz1             = _mm_setzero_ps();
 708         fix2             = _mm_setzero_ps();
 709         fiy2             = _mm_setzero_ps();
 710         fiz2             = _mm_setzero_ps();
 711         fix3             = _mm_setzero_ps();
 712         fiy3             = _mm_setzero_ps();
 713         fiz3             = _mm_setzero_ps();
 714
 715         /* Start inner kernel loop */
 716         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 717         {
 718
 719             /* Get j neighbor index, and coordinate index */
 720             jnrA             = jjnr[jidx];
 721             jnrB             = jjnr[jidx+1];
 722             jnrC             = jjnr[jidx+2];
 723             jnrD             = jjnr[jidx+3];
 724             j_coord_offsetA  = DIM*jnrA;
 725             j_coord_offsetB  = DIM*jnrB;
 726             j_coord_offsetC  = DIM*jnrC;
 727             j_coord_offsetD  = DIM*jnrD;
 728
 729             /* load j atom coordinates */
 730             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 731                                               x+j_coord_offsetC,x+j_coord_offsetD,
 732                                               &jx0,&jy0,&jz0);
 733
 734             /* Calculate displacement vector */
 735             dx10             = _mm_sub_ps(ix1,jx0);
 736             dy10             = _mm_sub_ps(iy1,jy0);
 737             dz10             = _mm_sub_ps(iz1,jz0);
 738             dx20             = _mm_sub_ps(ix2,jx0);
 739             dy20             = _mm_sub_ps(iy2,jy0);
 740             dz20             = _mm_sub_ps(iz2,jz0);
 741             dx30             = _mm_sub_ps(ix3,jx0);
 742             dy30             = _mm_sub_ps(iy3,jy0);
 743             dz30             = _mm_sub_ps(iz3,jz0);
 744
 745             /* Calculate squared distance and things based on it */
 746             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 747             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 748             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 749
 750             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 751             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 752             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 753
 754             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 755             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 756             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 757
 758             /* Load parameters for j particles */
 759             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 760                                                               charge+jnrC+0,charge+jnrD+0);
 761
 762             fjx0             = _mm_setzero_ps();
 763             fjy0             = _mm_setzero_ps();
 764             fjz0             = _mm_setzero_ps();
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             r10              = _mm_mul_ps(rsq10,rinv10);
 771
 772             /* Compute parameters for interactions between i and j atoms */
 773             qq10             = _mm_mul_ps(iq1,jq0);
 774
 775             /* EWALD ELECTROSTATICS */
 776
 777             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 778             ewrt             = _mm_mul_ps(r10,ewtabscale);
 779             ewitab           = _mm_cvttps_epi32(ewrt);
 780             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 781             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 782                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 783                                          &ewtabF,&ewtabFn);
 784             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 785             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 786
 787             fscal            = felec;
 788
 789             /* Calculate temporary vectorial force */
 790             tx               = _mm_mul_ps(fscal,dx10);
 791             ty               = _mm_mul_ps(fscal,dy10);
 792             tz               = _mm_mul_ps(fscal,dz10);
 793
 794             /* Update vectorial force */
 795             fix1             = _mm_add_ps(fix1,tx);
 796             fiy1             = _mm_add_ps(fiy1,ty);
 797             fiz1             = _mm_add_ps(fiz1,tz);
 798
 799             fjx0             = _mm_add_ps(fjx0,tx);
 800             fjy0             = _mm_add_ps(fjy0,ty);
 801             fjz0             = _mm_add_ps(fjz0,tz);
 802
 803             /**************************
 804              * CALCULATE INTERACTIONS *
 805              **************************/
 806
 807             r20              = _mm_mul_ps(rsq20,rinv20);
 808
 809             /* Compute parameters for interactions between i and j atoms */
 810             qq20             = _mm_mul_ps(iq2,jq0);
 811
 812             /* EWALD ELECTROSTATICS */
 813
 814             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 815             ewrt             = _mm_mul_ps(r20,ewtabscale);
 816             ewitab           = _mm_cvttps_epi32(ewrt);
 817             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 818             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 819                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 820                                          &ewtabF,&ewtabFn);
 821             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 822             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 823
 824             fscal            = felec;
 825
 826             /* Calculate temporary vectorial force */
 827             tx               = _mm_mul_ps(fscal,dx20);
 828             ty               = _mm_mul_ps(fscal,dy20);
 829             tz               = _mm_mul_ps(fscal,dz20);
 830
 831             /* Update vectorial force */
 832             fix2             = _mm_add_ps(fix2,tx);
 833             fiy2             = _mm_add_ps(fiy2,ty);
 834             fiz2             = _mm_add_ps(fiz2,tz);
 835
 836             fjx0             = _mm_add_ps(fjx0,tx);
 837             fjy0             = _mm_add_ps(fjy0,ty);
 838             fjz0             = _mm_add_ps(fjz0,tz);
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             r30              = _mm_mul_ps(rsq30,rinv30);
 845
 846             /* Compute parameters for interactions between i and j atoms */
 847             qq30             = _mm_mul_ps(iq3,jq0);
 848
 849             /* EWALD ELECTROSTATICS */
 850
 851             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 852             ewrt             = _mm_mul_ps(r30,ewtabscale);
 853             ewitab           = _mm_cvttps_epi32(ewrt);
 854             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 855             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 856                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 857                                          &ewtabF,&ewtabFn);
 858             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 859             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 860
 861             fscal            = felec;
 862
 863             /* Calculate temporary vectorial force */
 864             tx               = _mm_mul_ps(fscal,dx30);
 865             ty               = _mm_mul_ps(fscal,dy30);
 866             tz               = _mm_mul_ps(fscal,dz30);
 867
 868             /* Update vectorial force */
 869             fix3             = _mm_add_ps(fix3,tx);
 870             fiy3             = _mm_add_ps(fiy3,ty);
 871             fiz3             = _mm_add_ps(fiz3,tz);
 872
 873             fjx0             = _mm_add_ps(fjx0,tx);
 874             fjy0             = _mm_add_ps(fjy0,ty);
 875             fjz0             = _mm_add_ps(fjz0,tz);
 876
 877             fjptrA             = f+j_coord_offsetA;
 878             fjptrB             = f+j_coord_offsetB;
 879             fjptrC             = f+j_coord_offsetC;
 880             fjptrD             = f+j_coord_offsetD;
 881
 882             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 883
 884             /* Inner loop uses 108 flops */
 885         }
 886
 887         if(jidx<j_index_end)
 888         {
 889
 890             /* Get j neighbor index, and coordinate index */
 891             jnrlistA         = jjnr[jidx];
 892             jnrlistB         = jjnr[jidx+1];
 893             jnrlistC         = jjnr[jidx+2];
 894             jnrlistD         = jjnr[jidx+3];
 895             /* Sign of each element will be negative for non-real atoms.
 896              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 897              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 898              */
 899             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 900             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 901             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 902             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 903             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 904             j_coord_offsetA  = DIM*jnrA;
 905             j_coord_offsetB  = DIM*jnrB;
 906             j_coord_offsetC  = DIM*jnrC;
 907             j_coord_offsetD  = DIM*jnrD;
 908
 909             /* load j atom coordinates */
 910             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 911                                               x+j_coord_offsetC,x+j_coord_offsetD,
 912                                               &jx0,&jy0,&jz0);
 913
 914             /* Calculate displacement vector */
 915             dx10             = _mm_sub_ps(ix1,jx0);
 916             dy10             = _mm_sub_ps(iy1,jy0);
 917             dz10             = _mm_sub_ps(iz1,jz0);
 918             dx20             = _mm_sub_ps(ix2,jx0);
 919             dy20             = _mm_sub_ps(iy2,jy0);
 920             dz20             = _mm_sub_ps(iz2,jz0);
 921             dx30             = _mm_sub_ps(ix3,jx0);
 922             dy30             = _mm_sub_ps(iy3,jy0);
 923             dz30             = _mm_sub_ps(iz3,jz0);
 924
 925             /* Calculate squared distance and things based on it */
 926             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 927             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 928             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 929
 930             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 931             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 932             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 933
 934             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 935             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 936             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 937
 938             /* Load parameters for j particles */
 939             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 940                                                               charge+jnrC+0,charge+jnrD+0);
 941
 942             fjx0             = _mm_setzero_ps();
 943             fjy0             = _mm_setzero_ps();
 944             fjz0             = _mm_setzero_ps();
 945
 946             /**************************
 947              * CALCULATE INTERACTIONS *
 948              **************************/
 949
 950             r10              = _mm_mul_ps(rsq10,rinv10);
 951             r10              = _mm_andnot_ps(dummy_mask,r10);
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq10             = _mm_mul_ps(iq1,jq0);
 955
 956             /* EWALD ELECTROSTATICS */
 957
 958             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 959             ewrt             = _mm_mul_ps(r10,ewtabscale);
 960             ewitab           = _mm_cvttps_epi32(ewrt);
 961             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 962             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 963                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 964                                          &ewtabF,&ewtabFn);
 965             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 966             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 967
 968             fscal            = felec;
 969
 970             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 971
 972             /* Calculate temporary vectorial force */
 973             tx               = _mm_mul_ps(fscal,dx10);
 974             ty               = _mm_mul_ps(fscal,dy10);
 975             tz               = _mm_mul_ps(fscal,dz10);
 976
 977             /* Update vectorial force */
 978             fix1             = _mm_add_ps(fix1,tx);
 979             fiy1             = _mm_add_ps(fiy1,ty);
 980             fiz1             = _mm_add_ps(fiz1,tz);
 981
 982             fjx0             = _mm_add_ps(fjx0,tx);
 983             fjy0             = _mm_add_ps(fjy0,ty);
 984             fjz0             = _mm_add_ps(fjz0,tz);
 985
 986             /**************************
 987              * CALCULATE INTERACTIONS *
 988              **************************/
 989
 990             r20              = _mm_mul_ps(rsq20,rinv20);
 991             r20              = _mm_andnot_ps(dummy_mask,r20);
 992
 993             /* Compute parameters for interactions between i and j atoms */
 994             qq20             = _mm_mul_ps(iq2,jq0);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 999             ewrt             = _mm_mul_ps(r20,ewtabscale);
1000             ewitab           = _mm_cvttps_epi32(ewrt);
1001             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1002             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1003                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1004                                          &ewtabF,&ewtabFn);
1005             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1006             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1007
1008             fscal            = felec;
1009
1010             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1011
1012             /* Calculate temporary vectorial force */
1013             tx               = _mm_mul_ps(fscal,dx20);
1014             ty               = _mm_mul_ps(fscal,dy20);
1015             tz               = _mm_mul_ps(fscal,dz20);
1016
1017             /* Update vectorial force */
1018             fix2             = _mm_add_ps(fix2,tx);
1019             fiy2             = _mm_add_ps(fiy2,ty);
1020             fiz2             = _mm_add_ps(fiz2,tz);
1021
1022             fjx0             = _mm_add_ps(fjx0,tx);
1023             fjy0             = _mm_add_ps(fjy0,ty);
1024             fjz0             = _mm_add_ps(fjz0,tz);
1025
1026             /**************************
1027              * CALCULATE INTERACTIONS *
1028              **************************/
1029
1030             r30              = _mm_mul_ps(rsq30,rinv30);
1031             r30              = _mm_andnot_ps(dummy_mask,r30);
1032
1033             /* Compute parameters for interactions between i and j atoms */
1034             qq30             = _mm_mul_ps(iq3,jq0);
1035
1036             /* EWALD ELECTROSTATICS */
1037
1038             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1039             ewrt             = _mm_mul_ps(r30,ewtabscale);
1040             ewitab           = _mm_cvttps_epi32(ewrt);
1041             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1042             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1043                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1044                                          &ewtabF,&ewtabFn);
1045             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1046             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1051
1052             /* Calculate temporary vectorial force */
1053             tx               = _mm_mul_ps(fscal,dx30);
1054             ty               = _mm_mul_ps(fscal,dy30);
1055             tz               = _mm_mul_ps(fscal,dz30);
1056
1057             /* Update vectorial force */
1058             fix3             = _mm_add_ps(fix3,tx);
1059             fiy3             = _mm_add_ps(fiy3,ty);
1060             fiz3             = _mm_add_ps(fiz3,tz);
1061
1062             fjx0             = _mm_add_ps(fjx0,tx);
1063             fjy0             = _mm_add_ps(fjy0,ty);
1064             fjz0             = _mm_add_ps(fjz0,tz);
1065
1066             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1067             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1068             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1069             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1070
1071             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1072
1073             /* Inner loop uses 111 flops */
1074         }
1075
1076         /* End of innermost loop */
1077
1078         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1079                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1080
1081         /* Increment number of inner iterations */
1082         inneriter                  += j_index_end - j_index_start;
1083
1084         /* Outer loop uses 18 flops */
1085     }
1086
1087     /* Increment number of outer iterations */
1088     outeriter        += nri;
1089
1090     /* Update outer/inner flops */
1091
1092     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*111);
1093 }