src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_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_ElecEwSh_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_ElecEwSh_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              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              vdwioffset1;
  67     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  75     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  76     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128i          ewitab;
  80     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_ps(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99
 100     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 101     ewtab            = fr->ic->tabq_coul_FDV0;
 102     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 103     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 104
 105     /* Setup water-specific parameters */
 106     inr              = nlist->iinr[0];
 107     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 108     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 109     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 110
 111     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 112     rcutoff_scalar   = fr->rcoulomb;
 113     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 114     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 115
 116     /* Avoid stupid compiler warnings */
 117     jnrA = jnrB = jnrC = jnrD = 0;
 118     j_coord_offsetA = 0;
 119     j_coord_offsetB = 0;
 120     j_coord_offsetC = 0;
 121     j_coord_offsetD = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131         shX              = shiftvec[i_shift_offset+XX];
 132         shY              = shiftvec[i_shift_offset+YY];
 133         shZ              = shiftvec[i_shift_offset+ZZ];
 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         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 145         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 146         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 147         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 148         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 149         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 150         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 151         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 152         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 153
 154         fix1             = _mm_setzero_ps();
 155         fiy1             = _mm_setzero_ps();
 156         fiz1             = _mm_setzero_ps();
 157         fix2             = _mm_setzero_ps();
 158         fiy2             = _mm_setzero_ps();
 159         fiz2             = _mm_setzero_ps();
 160         fix3             = _mm_setzero_ps();
 161         fiy3             = _mm_setzero_ps();
 162         fiz3             = _mm_setzero_ps();
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_ps();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             jnrC             = jjnr[jidx+2];
 175             jnrD             = jjnr[jidx+3];
 176
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179             j_coord_offsetC  = DIM*jnrC;
 180             j_coord_offsetD  = DIM*jnrD;
 181
 182             /* load j atom coordinates */
 183             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 184                                               x+j_coord_offsetC,x+j_coord_offsetD,
 185                                               &jx0,&jy0,&jz0);
 186
 187             /* Calculate displacement vector */
 188             dx10             = _mm_sub_ps(ix1,jx0);
 189             dy10             = _mm_sub_ps(iy1,jy0);
 190             dz10             = _mm_sub_ps(iz1,jz0);
 191             dx20             = _mm_sub_ps(ix2,jx0);
 192             dy20             = _mm_sub_ps(iy2,jy0);
 193             dz20             = _mm_sub_ps(iz2,jz0);
 194             dx30             = _mm_sub_ps(ix3,jx0);
 195             dy30             = _mm_sub_ps(iy3,jy0);
 196             dz30             = _mm_sub_ps(iz3,jz0);
 197
 198             /* Calculate squared distance and things based on it */
 199             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 200             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 201             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 202
 203             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 204             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 205             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 206
 207             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 208             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 209             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                               charge+jnrC+0,charge+jnrD+0);
 214
 215             /**************************
 216              * CALCULATE INTERACTIONS *
 217              **************************/
 218
 219             if (gmx_mm_any_lt(rsq10,rcutoff2))
 220             {
 221
 222             r10              = _mm_mul_ps(rsq10,rinv10);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq10             = _mm_mul_ps(iq1,jq0);
 226
 227             /* EWALD ELECTROSTATICS */
 228
 229             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 230             ewrt             = _mm_mul_ps(r10,ewtabscale);
 231             ewitab           = _mm_cvttps_epi32(ewrt);
 232             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 233             ewitab           = _mm_slli_epi32(ewitab,2);
 234             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 235             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 236             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 237             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 238             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 239             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 240             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 241             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 242             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 243
 244             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velec            = _mm_and_ps(velec,cutoff_mask);
 248             velecsum         = _mm_add_ps(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             fscal            = _mm_and_ps(fscal,cutoff_mask);
 253
 254             /* Calculate temporary vectorial force */
 255             tx               = _mm_mul_ps(fscal,dx10);
 256             ty               = _mm_mul_ps(fscal,dy10);
 257             tz               = _mm_mul_ps(fscal,dz10);
 258
 259             /* Update vectorial force */
 260             fix1             = _mm_add_ps(fix1,tx);
 261             fiy1             = _mm_add_ps(fiy1,ty);
 262             fiz1             = _mm_add_ps(fiz1,tz);
 263
 264             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 265                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 266                                                    tx,ty,tz);
 267
 268             }
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             if (gmx_mm_any_lt(rsq20,rcutoff2))
 275             {
 276
 277             r20              = _mm_mul_ps(rsq20,rinv20);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             qq20             = _mm_mul_ps(iq2,jq0);
 281
 282             /* EWALD ELECTROSTATICS */
 283
 284             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 285             ewrt             = _mm_mul_ps(r20,ewtabscale);
 286             ewitab           = _mm_cvttps_epi32(ewrt);
 287             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 288             ewitab           = _mm_slli_epi32(ewitab,2);
 289             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 290             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 291             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 292             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 293             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 294             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 295             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 296             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 297             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 298
 299             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velec            = _mm_and_ps(velec,cutoff_mask);
 303             velecsum         = _mm_add_ps(velecsum,velec);
 304
 305             fscal            = felec;
 306
 307             fscal            = _mm_and_ps(fscal,cutoff_mask);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm_mul_ps(fscal,dx20);
 311             ty               = _mm_mul_ps(fscal,dy20);
 312             tz               = _mm_mul_ps(fscal,dz20);
 313
 314             /* Update vectorial force */
 315             fix2             = _mm_add_ps(fix2,tx);
 316             fiy2             = _mm_add_ps(fiy2,ty);
 317             fiz2             = _mm_add_ps(fiz2,tz);
 318
 319             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 320                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 321                                                    tx,ty,tz);
 322
 323             }
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             if (gmx_mm_any_lt(rsq30,rcutoff2))
 330             {
 331
 332             r30              = _mm_mul_ps(rsq30,rinv30);
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq30             = _mm_mul_ps(iq3,jq0);
 336
 337             /* EWALD ELECTROSTATICS */
 338
 339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 340             ewrt             = _mm_mul_ps(r30,ewtabscale);
 341             ewitab           = _mm_cvttps_epi32(ewrt);
 342             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 343             ewitab           = _mm_slli_epi32(ewitab,2);
 344             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 345             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 346             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 347             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 348             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 349             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 350             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 351             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 352             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 353
 354             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velec            = _mm_and_ps(velec,cutoff_mask);
 358             velecsum         = _mm_add_ps(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             fscal            = _mm_and_ps(fscal,cutoff_mask);
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm_mul_ps(fscal,dx30);
 366             ty               = _mm_mul_ps(fscal,dy30);
 367             tz               = _mm_mul_ps(fscal,dz30);
 368
 369             /* Update vectorial force */
 370             fix3             = _mm_add_ps(fix3,tx);
 371             fiy3             = _mm_add_ps(fiy3,ty);
 372             fiz3             = _mm_add_ps(fiz3,tz);
 373
 374             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 375                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 376                                                    tx,ty,tz);
 377
 378             }
 379
 380             /* Inner loop uses 138 flops */
 381         }
 382
 383         if(jidx<j_index_end)
 384         {
 385
 386             /* Get j neighbor index, and coordinate index */
 387             jnrA             = jjnr[jidx];
 388             jnrB             = jjnr[jidx+1];
 389             jnrC             = jjnr[jidx+2];
 390             jnrD             = jjnr[jidx+3];
 391
 392             /* Sign of each element will be negative for non-real atoms.
 393              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 394              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 395              */
 396             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 397             jnrA       = (jnrA>=0) ? jnrA : 0;
 398             jnrB       = (jnrB>=0) ? jnrB : 0;
 399             jnrC       = (jnrC>=0) ? jnrC : 0;
 400             jnrD       = (jnrD>=0) ? jnrD : 0;
 401
 402             j_coord_offsetA  = DIM*jnrA;
 403             j_coord_offsetB  = DIM*jnrB;
 404             j_coord_offsetC  = DIM*jnrC;
 405             j_coord_offsetD  = DIM*jnrD;
 406
 407             /* load j atom coordinates */
 408             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 409                                               x+j_coord_offsetC,x+j_coord_offsetD,
 410                                               &jx0,&jy0,&jz0);
 411
 412             /* Calculate displacement vector */
 413             dx10             = _mm_sub_ps(ix1,jx0);
 414             dy10             = _mm_sub_ps(iy1,jy0);
 415             dz10             = _mm_sub_ps(iz1,jz0);
 416             dx20             = _mm_sub_ps(ix2,jx0);
 417             dy20             = _mm_sub_ps(iy2,jy0);
 418             dz20             = _mm_sub_ps(iz2,jz0);
 419             dx30             = _mm_sub_ps(ix3,jx0);
 420             dy30             = _mm_sub_ps(iy3,jy0);
 421             dz30             = _mm_sub_ps(iz3,jz0);
 422
 423             /* Calculate squared distance and things based on it */
 424             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 425             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 426             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 427
 428             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 429             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 430             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 431
 432             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 433             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 434             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 435
 436             /* Load parameters for j particles */
 437             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 438                                                               charge+jnrC+0,charge+jnrD+0);
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             if (gmx_mm_any_lt(rsq10,rcutoff2))
 445             {
 446
 447             r10              = _mm_mul_ps(rsq10,rinv10);
 448             r10              = _mm_andnot_ps(dummy_mask,r10);
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq10             = _mm_mul_ps(iq1,jq0);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 456             ewrt             = _mm_mul_ps(r10,ewtabscale);
 457             ewitab           = _mm_cvttps_epi32(ewrt);
 458             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 459             ewitab           = _mm_slli_epi32(ewitab,2);
 460             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 461             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 462             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 463             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 464             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 465             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 466             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 467             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 468             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 469
 470             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             velec            = _mm_and_ps(velec,cutoff_mask);
 474             velec            = _mm_andnot_ps(dummy_mask,velec);
 475             velecsum         = _mm_add_ps(velecsum,velec);
 476
 477             fscal            = felec;
 478
 479             fscal            = _mm_and_ps(fscal,cutoff_mask);
 480
 481             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 482
 483             /* Calculate temporary vectorial force */
 484             tx               = _mm_mul_ps(fscal,dx10);
 485             ty               = _mm_mul_ps(fscal,dy10);
 486             tz               = _mm_mul_ps(fscal,dz10);
 487
 488             /* Update vectorial force */
 489             fix1             = _mm_add_ps(fix1,tx);
 490             fiy1             = _mm_add_ps(fiy1,ty);
 491             fiz1             = _mm_add_ps(fiz1,tz);
 492
 493             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 494                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 495                                                    tx,ty,tz);
 496
 497             }
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             if (gmx_mm_any_lt(rsq20,rcutoff2))
 504             {
 505
 506             r20              = _mm_mul_ps(rsq20,rinv20);
 507             r20              = _mm_andnot_ps(dummy_mask,r20);
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq20             = _mm_mul_ps(iq2,jq0);
 511
 512             /* EWALD ELECTROSTATICS */
 513
 514             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 515             ewrt             = _mm_mul_ps(r20,ewtabscale);
 516             ewitab           = _mm_cvttps_epi32(ewrt);
 517             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 518             ewitab           = _mm_slli_epi32(ewitab,2);
 519             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 520             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 521             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 522             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 523             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 524             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 525             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 526             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 527             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 528
 529             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 530
 531             /* Update potential sum for this i atom from the interaction with this j atom. */
 532             velec            = _mm_and_ps(velec,cutoff_mask);
 533             velec            = _mm_andnot_ps(dummy_mask,velec);
 534             velecsum         = _mm_add_ps(velecsum,velec);
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm_and_ps(fscal,cutoff_mask);
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_ps(fscal,dx20);
 544             ty               = _mm_mul_ps(fscal,dy20);
 545             tz               = _mm_mul_ps(fscal,dz20);
 546
 547             /* Update vectorial force */
 548             fix2             = _mm_add_ps(fix2,tx);
 549             fiy2             = _mm_add_ps(fiy2,ty);
 550             fiz2             = _mm_add_ps(fiz2,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             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             if (gmx_mm_any_lt(rsq30,rcutoff2))
 563             {
 564
 565             r30              = _mm_mul_ps(rsq30,rinv30);
 566             r30              = _mm_andnot_ps(dummy_mask,r30);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq30             = _mm_mul_ps(iq3,jq0);
 570
 571             /* EWALD ELECTROSTATICS */
 572
 573             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 574             ewrt             = _mm_mul_ps(r30,ewtabscale);
 575             ewitab           = _mm_cvttps_epi32(ewrt);
 576             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 577             ewitab           = _mm_slli_epi32(ewitab,2);
 578             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 579             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 580             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 581             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 582             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 583             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 584             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 585             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 586             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 587
 588             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velec            = _mm_and_ps(velec,cutoff_mask);
 592             velec            = _mm_andnot_ps(dummy_mask,velec);
 593             velecsum         = _mm_add_ps(velecsum,velec);
 594
 595             fscal            = felec;
 596
 597             fscal            = _mm_and_ps(fscal,cutoff_mask);
 598
 599             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm_mul_ps(fscal,dx30);
 603             ty               = _mm_mul_ps(fscal,dy30);
 604             tz               = _mm_mul_ps(fscal,dz30);
 605
 606             /* Update vectorial force */
 607             fix3             = _mm_add_ps(fix3,tx);
 608             fiy3             = _mm_add_ps(fiy3,ty);
 609             fiz3             = _mm_add_ps(fiz3,tz);
 610
 611             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 612                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 613                                                    tx,ty,tz);
 614
 615             }
 616
 617             /* Inner loop uses 141 flops */
 618         }
 619
 620         /* End of innermost loop */
 621
 622         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 623                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 624
 625         ggid                        = gid[iidx];
 626         /* Update potential energies */
 627         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 628
 629         /* Increment number of inner iterations */
 630         inneriter                  += j_index_end - j_index_start;
 631
 632         /* Outer loop uses 28 flops */
 633     }
 634
 635     /* Increment number of outer iterations */
 636     outeriter        += nri;
 637
 638     /* Update outer/inner flops */
 639
 640     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*28 + inneriter*141);
 641 }
 642 /*
 643  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse2_single
 644  * Electrostatics interaction: Ewald
 645  * VdW interaction:            None
 646  * Geometry:                   Water4-Particle
 647  * Calculate force/pot:        Force
 648  */
 649 void
 650 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse2_single
 651                     (t_nblist * gmx_restrict                nlist,
 652                      rvec * gmx_restrict                    xx,
 653                      rvec * gmx_restrict                    ff,
 654                      t_forcerec * gmx_restrict              fr,
 655                      t_mdatoms * gmx_restrict               mdatoms,
 656                      nb_kernel_data_t * gmx_restrict        kernel_data,
 657                      t_nrnb * gmx_restrict                  nrnb)
 658 {
 659     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 660      * just 0 for non-waters.
 661      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 662      * jnr indices corresponding to data put in the four positions in the SIMD register.
 663      */
 664     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 665     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 666     int              jnrA,jnrB,jnrC,jnrD;
 667     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 668     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 669     real             shX,shY,shZ,rcutoff_scalar;
 670     real             *shiftvec,*fshift,*x,*f;
 671     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 672     int              vdwioffset1;
 673     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 674     int              vdwioffset2;
 675     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 676     int              vdwioffset3;
 677     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 678     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 679     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 680     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 681     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 682     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 683     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 684     real             *charge;
 685     __m128i          ewitab;
 686     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 687     real             *ewtab;
 688     __m128           dummy_mask,cutoff_mask;
 689     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 690     __m128           one     = _mm_set1_ps(1.0);
 691     __m128           two     = _mm_set1_ps(2.0);
 692     x                = xx[0];
 693     f                = ff[0];
 694
 695     nri              = nlist->nri;
 696     iinr             = nlist->iinr;
 697     jindex           = nlist->jindex;
 698     jjnr             = nlist->jjnr;
 699     shiftidx         = nlist->shift;
 700     gid              = nlist->gid;
 701     shiftvec         = fr->shift_vec[0];
 702     fshift           = fr->fshift[0];
 703     facel            = _mm_set1_ps(fr->epsfac);
 704     charge           = mdatoms->chargeA;
 705
 706     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 707     ewtab            = fr->ic->tabq_coul_F;
 708     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 709     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 710
 711     /* Setup water-specific parameters */
 712     inr              = nlist->iinr[0];
 713     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 714     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 715     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 716
 717     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 718     rcutoff_scalar   = fr->rcoulomb;
 719     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 720     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 721
 722     /* Avoid stupid compiler warnings */
 723     jnrA = jnrB = jnrC = jnrD = 0;
 724     j_coord_offsetA = 0;
 725     j_coord_offsetB = 0;
 726     j_coord_offsetC = 0;
 727     j_coord_offsetD = 0;
 728
 729     outeriter        = 0;
 730     inneriter        = 0;
 731
 732     /* Start outer loop over neighborlists */
 733     for(iidx=0; iidx<nri; iidx++)
 734     {
 735         /* Load shift vector for this list */
 736         i_shift_offset   = DIM*shiftidx[iidx];
 737         shX              = shiftvec[i_shift_offset+XX];
 738         shY              = shiftvec[i_shift_offset+YY];
 739         shZ              = shiftvec[i_shift_offset+ZZ];
 740
 741         /* Load limits for loop over neighbors */
 742         j_index_start    = jindex[iidx];
 743         j_index_end      = jindex[iidx+1];
 744
 745         /* Get outer coordinate index */
 746         inr              = iinr[iidx];
 747         i_coord_offset   = DIM*inr;
 748
 749         /* Load i particle coords and add shift vector */
 750         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 751         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 752         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 753         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 754         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 755         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 756         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 757         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 758         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 759
 760         fix1             = _mm_setzero_ps();
 761         fiy1             = _mm_setzero_ps();
 762         fiz1             = _mm_setzero_ps();
 763         fix2             = _mm_setzero_ps();
 764         fiy2             = _mm_setzero_ps();
 765         fiz2             = _mm_setzero_ps();
 766         fix3             = _mm_setzero_ps();
 767         fiy3             = _mm_setzero_ps();
 768         fiz3             = _mm_setzero_ps();
 769
 770         /* Start inner kernel loop */
 771         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 772         {
 773
 774             /* Get j neighbor index, and coordinate index */
 775             jnrA             = jjnr[jidx];
 776             jnrB             = jjnr[jidx+1];
 777             jnrC             = jjnr[jidx+2];
 778             jnrD             = jjnr[jidx+3];
 779
 780             j_coord_offsetA  = DIM*jnrA;
 781             j_coord_offsetB  = DIM*jnrB;
 782             j_coord_offsetC  = DIM*jnrC;
 783             j_coord_offsetD  = DIM*jnrD;
 784
 785             /* load j atom coordinates */
 786             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 787                                               x+j_coord_offsetC,x+j_coord_offsetD,
 788                                               &jx0,&jy0,&jz0);
 789
 790             /* Calculate displacement vector */
 791             dx10             = _mm_sub_ps(ix1,jx0);
 792             dy10             = _mm_sub_ps(iy1,jy0);
 793             dz10             = _mm_sub_ps(iz1,jz0);
 794             dx20             = _mm_sub_ps(ix2,jx0);
 795             dy20             = _mm_sub_ps(iy2,jy0);
 796             dz20             = _mm_sub_ps(iz2,jz0);
 797             dx30             = _mm_sub_ps(ix3,jx0);
 798             dy30             = _mm_sub_ps(iy3,jy0);
 799             dz30             = _mm_sub_ps(iz3,jz0);
 800
 801             /* Calculate squared distance and things based on it */
 802             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 803             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 804             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 805
 806             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 807             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 808             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 809
 810             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 811             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 812             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 813
 814             /* Load parameters for j particles */
 815             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 816                                                               charge+jnrC+0,charge+jnrD+0);
 817
 818             /**************************
 819              * CALCULATE INTERACTIONS *
 820              **************************/
 821
 822             if (gmx_mm_any_lt(rsq10,rcutoff2))
 823             {
 824
 825             r10              = _mm_mul_ps(rsq10,rinv10);
 826
 827             /* Compute parameters for interactions between i and j atoms */
 828             qq10             = _mm_mul_ps(iq1,jq0);
 829
 830             /* EWALD ELECTROSTATICS */
 831
 832             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 833             ewrt             = _mm_mul_ps(r10,ewtabscale);
 834             ewitab           = _mm_cvttps_epi32(ewrt);
 835             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 836             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 837                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 838                                          &ewtabF,&ewtabFn);
 839             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 840             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 841
 842             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 843
 844             fscal            = felec;
 845
 846             fscal            = _mm_and_ps(fscal,cutoff_mask);
 847
 848             /* Calculate temporary vectorial force */
 849             tx               = _mm_mul_ps(fscal,dx10);
 850             ty               = _mm_mul_ps(fscal,dy10);
 851             tz               = _mm_mul_ps(fscal,dz10);
 852
 853             /* Update vectorial force */
 854             fix1             = _mm_add_ps(fix1,tx);
 855             fiy1             = _mm_add_ps(fiy1,ty);
 856             fiz1             = _mm_add_ps(fiz1,tz);
 857
 858             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 859                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 860                                                    tx,ty,tz);
 861
 862             }
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             if (gmx_mm_any_lt(rsq20,rcutoff2))
 869             {
 870
 871             r20              = _mm_mul_ps(rsq20,rinv20);
 872
 873             /* Compute parameters for interactions between i and j atoms */
 874             qq20             = _mm_mul_ps(iq2,jq0);
 875
 876             /* EWALD ELECTROSTATICS */
 877
 878             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 879             ewrt             = _mm_mul_ps(r20,ewtabscale);
 880             ewitab           = _mm_cvttps_epi32(ewrt);
 881             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 882             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 883                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 884                                          &ewtabF,&ewtabFn);
 885             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 886             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 887
 888             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 889
 890             fscal            = felec;
 891
 892             fscal            = _mm_and_ps(fscal,cutoff_mask);
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm_mul_ps(fscal,dx20);
 896             ty               = _mm_mul_ps(fscal,dy20);
 897             tz               = _mm_mul_ps(fscal,dz20);
 898
 899             /* Update vectorial force */
 900             fix2             = _mm_add_ps(fix2,tx);
 901             fiy2             = _mm_add_ps(fiy2,ty);
 902             fiz2             = _mm_add_ps(fiz2,tz);
 903
 904             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 905                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 906                                                    tx,ty,tz);
 907
 908             }
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             if (gmx_mm_any_lt(rsq30,rcutoff2))
 915             {
 916
 917             r30              = _mm_mul_ps(rsq30,rinv30);
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq30             = _mm_mul_ps(iq3,jq0);
 921
 922             /* EWALD ELECTROSTATICS */
 923
 924             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 925             ewrt             = _mm_mul_ps(r30,ewtabscale);
 926             ewitab           = _mm_cvttps_epi32(ewrt);
 927             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 928             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 929                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 930                                          &ewtabF,&ewtabFn);
 931             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 932             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 933
 934             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 935
 936             fscal            = felec;
 937
 938             fscal            = _mm_and_ps(fscal,cutoff_mask);
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm_mul_ps(fscal,dx30);
 942             ty               = _mm_mul_ps(fscal,dy30);
 943             tz               = _mm_mul_ps(fscal,dz30);
 944
 945             /* Update vectorial force */
 946             fix3             = _mm_add_ps(fix3,tx);
 947             fiy3             = _mm_add_ps(fiy3,ty);
 948             fiz3             = _mm_add_ps(fiz3,tz);
 949
 950             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 951                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 952                                                    tx,ty,tz);
 953
 954             }
 955
 956             /* Inner loop uses 117 flops */
 957         }
 958
 959         if(jidx<j_index_end)
 960         {
 961
 962             /* Get j neighbor index, and coordinate index */
 963             jnrA             = jjnr[jidx];
 964             jnrB             = jjnr[jidx+1];
 965             jnrC             = jjnr[jidx+2];
 966             jnrD             = jjnr[jidx+3];
 967
 968             /* Sign of each element will be negative for non-real atoms.
 969              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 970              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 971              */
 972             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 973             jnrA       = (jnrA>=0) ? jnrA : 0;
 974             jnrB       = (jnrB>=0) ? jnrB : 0;
 975             jnrC       = (jnrC>=0) ? jnrC : 0;
 976             jnrD       = (jnrD>=0) ? jnrD : 0;
 977
 978             j_coord_offsetA  = DIM*jnrA;
 979             j_coord_offsetB  = DIM*jnrB;
 980             j_coord_offsetC  = DIM*jnrC;
 981             j_coord_offsetD  = DIM*jnrD;
 982
 983             /* load j atom coordinates */
 984             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 985                                               x+j_coord_offsetC,x+j_coord_offsetD,
 986                                               &jx0,&jy0,&jz0);
 987
 988             /* Calculate displacement vector */
 989             dx10             = _mm_sub_ps(ix1,jx0);
 990             dy10             = _mm_sub_ps(iy1,jy0);
 991             dz10             = _mm_sub_ps(iz1,jz0);
 992             dx20             = _mm_sub_ps(ix2,jx0);
 993             dy20             = _mm_sub_ps(iy2,jy0);
 994             dz20             = _mm_sub_ps(iz2,jz0);
 995             dx30             = _mm_sub_ps(ix3,jx0);
 996             dy30             = _mm_sub_ps(iy3,jy0);
 997             dz30             = _mm_sub_ps(iz3,jz0);
 998
 999             /* Calculate squared distance and things based on it */
1000             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1001             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1002             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1003
1004             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1005             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1006             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1007
1008             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1009             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1010             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1011
1012             /* Load parameters for j particles */
1013             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1014                                                               charge+jnrC+0,charge+jnrD+0);
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             if (gmx_mm_any_lt(rsq10,rcutoff2))
1021             {
1022
1023             r10              = _mm_mul_ps(rsq10,rinv10);
1024             r10              = _mm_andnot_ps(dummy_mask,r10);
1025
1026             /* Compute parameters for interactions between i and j atoms */
1027             qq10             = _mm_mul_ps(iq1,jq0);
1028
1029             /* EWALD ELECTROSTATICS */
1030
1031             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1032             ewrt             = _mm_mul_ps(r10,ewtabscale);
1033             ewitab           = _mm_cvttps_epi32(ewrt);
1034             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1035             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1036                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1037                                          &ewtabF,&ewtabFn);
1038             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1039             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1040
1041             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1042
1043             fscal            = felec;
1044
1045             fscal            = _mm_and_ps(fscal,cutoff_mask);
1046
1047             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1048
1049             /* Calculate temporary vectorial force */
1050             tx               = _mm_mul_ps(fscal,dx10);
1051             ty               = _mm_mul_ps(fscal,dy10);
1052             tz               = _mm_mul_ps(fscal,dz10);
1053
1054             /* Update vectorial force */
1055             fix1             = _mm_add_ps(fix1,tx);
1056             fiy1             = _mm_add_ps(fiy1,ty);
1057             fiz1             = _mm_add_ps(fiz1,tz);
1058
1059             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1060                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1061                                                    tx,ty,tz);
1062
1063             }
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             if (gmx_mm_any_lt(rsq20,rcutoff2))
1070             {
1071
1072             r20              = _mm_mul_ps(rsq20,rinv20);
1073             r20              = _mm_andnot_ps(dummy_mask,r20);
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq20             = _mm_mul_ps(iq2,jq0);
1077
1078             /* EWALD ELECTROSTATICS */
1079
1080             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1081             ewrt             = _mm_mul_ps(r20,ewtabscale);
1082             ewitab           = _mm_cvttps_epi32(ewrt);
1083             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1084             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1085                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1086                                          &ewtabF,&ewtabFn);
1087             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1088             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1089
1090             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm_and_ps(fscal,cutoff_mask);
1095
1096             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1097
1098             /* Calculate temporary vectorial force */
1099             tx               = _mm_mul_ps(fscal,dx20);
1100             ty               = _mm_mul_ps(fscal,dy20);
1101             tz               = _mm_mul_ps(fscal,dz20);
1102
1103             /* Update vectorial force */
1104             fix2             = _mm_add_ps(fix2,tx);
1105             fiy2             = _mm_add_ps(fiy2,ty);
1106             fiz2             = _mm_add_ps(fiz2,tz);
1107
1108             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1109                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1110                                                    tx,ty,tz);
1111
1112             }
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             if (gmx_mm_any_lt(rsq30,rcutoff2))
1119             {
1120
1121             r30              = _mm_mul_ps(rsq30,rinv30);
1122             r30              = _mm_andnot_ps(dummy_mask,r30);
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             qq30             = _mm_mul_ps(iq3,jq0);
1126
1127             /* EWALD ELECTROSTATICS */
1128
1129             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1130             ewrt             = _mm_mul_ps(r30,ewtabscale);
1131             ewitab           = _mm_cvttps_epi32(ewrt);
1132             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1133             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1134                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1135                                          &ewtabF,&ewtabFn);
1136             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1137             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1138
1139             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1140
1141             fscal            = felec;
1142
1143             fscal            = _mm_and_ps(fscal,cutoff_mask);
1144
1145             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1146
1147             /* Calculate temporary vectorial force */
1148             tx               = _mm_mul_ps(fscal,dx30);
1149             ty               = _mm_mul_ps(fscal,dy30);
1150             tz               = _mm_mul_ps(fscal,dz30);
1151
1152             /* Update vectorial force */
1153             fix3             = _mm_add_ps(fix3,tx);
1154             fiy3             = _mm_add_ps(fiy3,ty);
1155             fiz3             = _mm_add_ps(fiz3,tz);
1156
1157             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1158                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1159                                                    tx,ty,tz);
1160
1161             }
1162
1163             /* Inner loop uses 120 flops */
1164         }
1165
1166         /* End of innermost loop */
1167
1168         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1169                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1170
1171         /* Increment number of inner iterations */
1172         inneriter                  += j_index_end - j_index_start;
1173
1174         /* Outer loop uses 27 flops */
1175     }
1176
1177     /* Increment number of outer iterations */
1178     outeriter        += nri;
1179
1180     /* Update outer/inner flops */
1181
1182     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*27 + inneriter*120);
1183 }