src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSw_VdwNone_GeomP1P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              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              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  72     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  73     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  74     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  75     real             *charge;
  76     __m128i          ewitab;
  77     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  78     real             *ewtab;
  79     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  80     real             rswitch_scalar,d_scalar;
  81     __m128           dummy_mask,cutoff_mask;
  82     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  83     __m128           one     = _mm_set1_ps(1.0);
  84     __m128           two     = _mm_set1_ps(2.0);
  85     x                = xx[0];
  86     f                = ff[0];
  87
  88     nri              = nlist->nri;
  89     iinr             = nlist->iinr;
  90     jindex           = nlist->jindex;
  91     jjnr             = nlist->jjnr;
  92     shiftidx         = nlist->shift;
  93     gid              = nlist->gid;
  94     shiftvec         = fr->shift_vec[0];
  95     fshift           = fr->fshift[0];
  96     facel            = _mm_set1_ps(fr->epsfac);
  97     charge           = mdatoms->chargeA;
  98
  99     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 100     ewtab            = fr->ic->tabq_coul_FDV0;
 101     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 102     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 103
 104     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 105     rcutoff_scalar   = fr->rcoulomb;
 106     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 107     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 108
 109     rswitch_scalar   = fr->rcoulomb_switch;
 110     rswitch          = _mm_set1_ps(rswitch_scalar);
 111     /* Setup switch parameters */
 112     d_scalar         = rcutoff_scalar-rswitch_scalar;
 113     d                = _mm_set1_ps(d_scalar);
 114     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 115     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 116     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 117     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 118     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 119     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = jnrC = jnrD = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125     j_coord_offsetC = 0;
 126     j_coord_offsetD = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     for(iidx=0;iidx<4*DIM;iidx++)
 132     {
 133         scratch[iidx] = 0.0;
 134     }
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 152
 153         fix0             = _mm_setzero_ps();
 154         fiy0             = _mm_setzero_ps();
 155         fiz0             = _mm_setzero_ps();
 156
 157         /* Load parameters for i particles */
 158         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 159
 160         /* Reset potential sums */
 161         velecsum         = _mm_setzero_ps();
 162
 163         /* Start inner kernel loop */
 164         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 165         {
 166
 167             /* Get j neighbor index, and coordinate index */
 168             jnrA             = jjnr[jidx];
 169             jnrB             = jjnr[jidx+1];
 170             jnrC             = jjnr[jidx+2];
 171             jnrD             = jjnr[jidx+3];
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               x+j_coord_offsetC,x+j_coord_offsetD,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_ps(ix0,jx0);
 184             dy00             = _mm_sub_ps(iy0,jy0);
 185             dz00             = _mm_sub_ps(iz0,jz0);
 186
 187             /* Calculate squared distance and things based on it */
 188             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 189
 190             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 191
 192             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 196                                                               charge+jnrC+0,charge+jnrD+0);
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_mm_any_lt(rsq00,rcutoff2))
 203             {
 204
 205             r00              = _mm_mul_ps(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _mm_mul_ps(iq0,jq0);
 209
 210             /* EWALD ELECTROSTATICS */
 211
 212             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 213             ewrt             = _mm_mul_ps(r00,ewtabscale);
 214             ewitab           = _mm_cvttps_epi32(ewrt);
 215             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 216             ewitab           = _mm_slli_epi32(ewitab,2);
 217             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 218             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 219             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 220             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 221             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 222             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 223             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 224             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 225             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 226
 227             d                = _mm_sub_ps(r00,rswitch);
 228             d                = _mm_max_ps(d,_mm_setzero_ps());
 229             d2               = _mm_mul_ps(d,d);
 230             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 231
 232             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 233
 234             /* Evaluate switch function */
 235             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 236             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 237             velec            = _mm_mul_ps(velec,sw);
 238             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 239
 240             /* Update potential sum for this i atom from the interaction with this j atom. */
 241             velec            = _mm_and_ps(velec,cutoff_mask);
 242             velecsum         = _mm_add_ps(velecsum,velec);
 243
 244             fscal            = felec;
 245
 246             fscal            = _mm_and_ps(fscal,cutoff_mask);
 247
 248             /* Calculate temporary vectorial force */
 249             tx               = _mm_mul_ps(fscal,dx00);
 250             ty               = _mm_mul_ps(fscal,dy00);
 251             tz               = _mm_mul_ps(fscal,dz00);
 252
 253             /* Update vectorial force */
 254             fix0             = _mm_add_ps(fix0,tx);
 255             fiy0             = _mm_add_ps(fiy0,ty);
 256             fiz0             = _mm_add_ps(fiz0,tz);
 257
 258             fjptrA             = f+j_coord_offsetA;
 259             fjptrB             = f+j_coord_offsetB;
 260             fjptrC             = f+j_coord_offsetC;
 261             fjptrD             = f+j_coord_offsetD;
 262             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 263
 264             }
 265
 266             /* Inner loop uses 65 flops */
 267         }
 268
 269         if(jidx<j_index_end)
 270         {
 271
 272             /* Get j neighbor index, and coordinate index */
 273             jnrlistA         = jjnr[jidx];
 274             jnrlistB         = jjnr[jidx+1];
 275             jnrlistC         = jjnr[jidx+2];
 276             jnrlistD         = jjnr[jidx+3];
 277             /* Sign of each element will be negative for non-real atoms.
 278              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 279              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 280              */
 281             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 282             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 283             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 284             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 285             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 286             j_coord_offsetA  = DIM*jnrA;
 287             j_coord_offsetB  = DIM*jnrB;
 288             j_coord_offsetC  = DIM*jnrC;
 289             j_coord_offsetD  = DIM*jnrD;
 290
 291             /* load j atom coordinates */
 292             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 293                                               x+j_coord_offsetC,x+j_coord_offsetD,
 294                                               &jx0,&jy0,&jz0);
 295
 296             /* Calculate displacement vector */
 297             dx00             = _mm_sub_ps(ix0,jx0);
 298             dy00             = _mm_sub_ps(iy0,jy0);
 299             dz00             = _mm_sub_ps(iz0,jz0);
 300
 301             /* Calculate squared distance and things based on it */
 302             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 303
 304             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 305
 306             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 307
 308             /* Load parameters for j particles */
 309             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 310                                                               charge+jnrC+0,charge+jnrD+0);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             if (gmx_mm_any_lt(rsq00,rcutoff2))
 317             {
 318
 319             r00              = _mm_mul_ps(rsq00,rinv00);
 320             r00              = _mm_andnot_ps(dummy_mask,r00);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq00             = _mm_mul_ps(iq0,jq0);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 328             ewrt             = _mm_mul_ps(r00,ewtabscale);
 329             ewitab           = _mm_cvttps_epi32(ewrt);
 330             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 334             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 335             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 336             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 337             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 338             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 339             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 340             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 341
 342             d                = _mm_sub_ps(r00,rswitch);
 343             d                = _mm_max_ps(d,_mm_setzero_ps());
 344             d2               = _mm_mul_ps(d,d);
 345             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 346
 347             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 348
 349             /* Evaluate switch function */
 350             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 351             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 352             velec            = _mm_mul_ps(velec,sw);
 353             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velec            = _mm_and_ps(velec,cutoff_mask);
 357             velec            = _mm_andnot_ps(dummy_mask,velec);
 358             velecsum         = _mm_add_ps(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             fscal            = _mm_and_ps(fscal,cutoff_mask);
 363
 364             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 365
 366             /* Calculate temporary vectorial force */
 367             tx               = _mm_mul_ps(fscal,dx00);
 368             ty               = _mm_mul_ps(fscal,dy00);
 369             tz               = _mm_mul_ps(fscal,dz00);
 370
 371             /* Update vectorial force */
 372             fix0             = _mm_add_ps(fix0,tx);
 373             fiy0             = _mm_add_ps(fiy0,ty);
 374             fiz0             = _mm_add_ps(fiz0,tz);
 375
 376             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 377             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 378             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 379             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 380             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 381
 382             }
 383
 384             /* Inner loop uses 66 flops */
 385         }
 386
 387         /* End of innermost loop */
 388
 389         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 390                                               f+i_coord_offset,fshift+i_shift_offset);
 391
 392         ggid                        = gid[iidx];
 393         /* Update potential energies */
 394         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 395
 396         /* Increment number of inner iterations */
 397         inneriter                  += j_index_end - j_index_start;
 398
 399         /* Outer loop uses 8 flops */
 400     }
 401
 402     /* Increment number of outer iterations */
 403     outeriter        += nri;
 404
 405     /* Update outer/inner flops */
 406
 407     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*66);
 408 }
 409 /*
 410  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sse2_single
 411  * Electrostatics interaction: Ewald
 412  * VdW interaction:            None
 413  * Geometry:                   Particle-Particle
 414  * Calculate force/pot:        Force
 415  */
 416 void
 417 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sse2_single
 418                     (t_nblist * gmx_restrict                nlist,
 419                      rvec * gmx_restrict                    xx,
 420                      rvec * gmx_restrict                    ff,
 421                      t_forcerec * gmx_restrict              fr,
 422                      t_mdatoms * gmx_restrict               mdatoms,
 423                      nb_kernel_data_t * gmx_restrict        kernel_data,
 424                      t_nrnb * gmx_restrict                  nrnb)
 425 {
 426     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 427      * just 0 for non-waters.
 428      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 429      * jnr indices corresponding to data put in the four positions in the SIMD register.
 430      */
 431     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 432     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 433     int              jnrA,jnrB,jnrC,jnrD;
 434     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 435     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 436     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 437     real             rcutoff_scalar;
 438     real             *shiftvec,*fshift,*x,*f;
 439     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 440     real             scratch[4*DIM];
 441     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 442     int              vdwioffset0;
 443     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 444     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 445     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 446     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 447     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 448     real             *charge;
 449     __m128i          ewitab;
 450     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 451     real             *ewtab;
 452     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 453     real             rswitch_scalar,d_scalar;
 454     __m128           dummy_mask,cutoff_mask;
 455     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 456     __m128           one     = _mm_set1_ps(1.0);
 457     __m128           two     = _mm_set1_ps(2.0);
 458     x                = xx[0];
 459     f                = ff[0];
 460
 461     nri              = nlist->nri;
 462     iinr             = nlist->iinr;
 463     jindex           = nlist->jindex;
 464     jjnr             = nlist->jjnr;
 465     shiftidx         = nlist->shift;
 466     gid              = nlist->gid;
 467     shiftvec         = fr->shift_vec[0];
 468     fshift           = fr->fshift[0];
 469     facel            = _mm_set1_ps(fr->epsfac);
 470     charge           = mdatoms->chargeA;
 471
 472     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 473     ewtab            = fr->ic->tabq_coul_FDV0;
 474     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 475     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 476
 477     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 478     rcutoff_scalar   = fr->rcoulomb;
 479     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 480     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 481
 482     rswitch_scalar   = fr->rcoulomb_switch;
 483     rswitch          = _mm_set1_ps(rswitch_scalar);
 484     /* Setup switch parameters */
 485     d_scalar         = rcutoff_scalar-rswitch_scalar;
 486     d                = _mm_set1_ps(d_scalar);
 487     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 488     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 489     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 490     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 491     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 492     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 493
 494     /* Avoid stupid compiler warnings */
 495     jnrA = jnrB = jnrC = jnrD = 0;
 496     j_coord_offsetA = 0;
 497     j_coord_offsetB = 0;
 498     j_coord_offsetC = 0;
 499     j_coord_offsetD = 0;
 500
 501     outeriter        = 0;
 502     inneriter        = 0;
 503
 504     for(iidx=0;iidx<4*DIM;iidx++)
 505     {
 506         scratch[iidx] = 0.0;
 507     }
 508
 509     /* Start outer loop over neighborlists */
 510     for(iidx=0; iidx<nri; iidx++)
 511     {
 512         /* Load shift vector for this list */
 513         i_shift_offset   = DIM*shiftidx[iidx];
 514
 515         /* Load limits for loop over neighbors */
 516         j_index_start    = jindex[iidx];
 517         j_index_end      = jindex[iidx+1];
 518
 519         /* Get outer coordinate index */
 520         inr              = iinr[iidx];
 521         i_coord_offset   = DIM*inr;
 522
 523         /* Load i particle coords and add shift vector */
 524         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 525
 526         fix0             = _mm_setzero_ps();
 527         fiy0             = _mm_setzero_ps();
 528         fiz0             = _mm_setzero_ps();
 529
 530         /* Load parameters for i particles */
 531         iq0              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+0));
 532
 533         /* Start inner kernel loop */
 534         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 535         {
 536
 537             /* Get j neighbor index, and coordinate index */
 538             jnrA             = jjnr[jidx];
 539             jnrB             = jjnr[jidx+1];
 540             jnrC             = jjnr[jidx+2];
 541             jnrD             = jjnr[jidx+3];
 542             j_coord_offsetA  = DIM*jnrA;
 543             j_coord_offsetB  = DIM*jnrB;
 544             j_coord_offsetC  = DIM*jnrC;
 545             j_coord_offsetD  = DIM*jnrD;
 546
 547             /* load j atom coordinates */
 548             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 549                                               x+j_coord_offsetC,x+j_coord_offsetD,
 550                                               &jx0,&jy0,&jz0);
 551
 552             /* Calculate displacement vector */
 553             dx00             = _mm_sub_ps(ix0,jx0);
 554             dy00             = _mm_sub_ps(iy0,jy0);
 555             dz00             = _mm_sub_ps(iz0,jz0);
 556
 557             /* Calculate squared distance and things based on it */
 558             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 559
 560             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 561
 562             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 563
 564             /* Load parameters for j particles */
 565             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 566                                                               charge+jnrC+0,charge+jnrD+0);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm_any_lt(rsq00,rcutoff2))
 573             {
 574
 575             r00              = _mm_mul_ps(rsq00,rinv00);
 576
 577             /* Compute parameters for interactions between i and j atoms */
 578             qq00             = _mm_mul_ps(iq0,jq0);
 579
 580             /* EWALD ELECTROSTATICS */
 581
 582             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 583             ewrt             = _mm_mul_ps(r00,ewtabscale);
 584             ewitab           = _mm_cvttps_epi32(ewrt);
 585             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 586             ewitab           = _mm_slli_epi32(ewitab,2);
 587             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 588             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 589             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 590             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 591             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 592             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 593             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 594             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 595             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 596
 597             d                = _mm_sub_ps(r00,rswitch);
 598             d                = _mm_max_ps(d,_mm_setzero_ps());
 599             d2               = _mm_mul_ps(d,d);
 600             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 601
 602             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 603
 604             /* Evaluate switch function */
 605             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 606             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 607             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_and_ps(fscal,cutoff_mask);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_ps(fscal,dx00);
 615             ty               = _mm_mul_ps(fscal,dy00);
 616             tz               = _mm_mul_ps(fscal,dz00);
 617
 618             /* Update vectorial force */
 619             fix0             = _mm_add_ps(fix0,tx);
 620             fiy0             = _mm_add_ps(fiy0,ty);
 621             fiz0             = _mm_add_ps(fiz0,tz);
 622
 623             fjptrA             = f+j_coord_offsetA;
 624             fjptrB             = f+j_coord_offsetB;
 625             fjptrC             = f+j_coord_offsetC;
 626             fjptrD             = f+j_coord_offsetD;
 627             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 628
 629             }
 630
 631             /* Inner loop uses 62 flops */
 632         }
 633
 634         if(jidx<j_index_end)
 635         {
 636
 637             /* Get j neighbor index, and coordinate index */
 638             jnrlistA         = jjnr[jidx];
 639             jnrlistB         = jjnr[jidx+1];
 640             jnrlistC         = jjnr[jidx+2];
 641             jnrlistD         = jjnr[jidx+3];
 642             /* Sign of each element will be negative for non-real atoms.
 643              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 644              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 645              */
 646             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 647             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 648             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 649             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 650             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 651             j_coord_offsetA  = DIM*jnrA;
 652             j_coord_offsetB  = DIM*jnrB;
 653             j_coord_offsetC  = DIM*jnrC;
 654             j_coord_offsetD  = DIM*jnrD;
 655
 656             /* load j atom coordinates */
 657             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 658                                               x+j_coord_offsetC,x+j_coord_offsetD,
 659                                               &jx0,&jy0,&jz0);
 660
 661             /* Calculate displacement vector */
 662             dx00             = _mm_sub_ps(ix0,jx0);
 663             dy00             = _mm_sub_ps(iy0,jy0);
 664             dz00             = _mm_sub_ps(iz0,jz0);
 665
 666             /* Calculate squared distance and things based on it */
 667             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 668
 669             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 670
 671             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 672
 673             /* Load parameters for j particles */
 674             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 675                                                               charge+jnrC+0,charge+jnrD+0);
 676
 677             /**************************
 678              * CALCULATE INTERACTIONS *
 679              **************************/
 680
 681             if (gmx_mm_any_lt(rsq00,rcutoff2))
 682             {
 683
 684             r00              = _mm_mul_ps(rsq00,rinv00);
 685             r00              = _mm_andnot_ps(dummy_mask,r00);
 686
 687             /* Compute parameters for interactions between i and j atoms */
 688             qq00             = _mm_mul_ps(iq0,jq0);
 689
 690             /* EWALD ELECTROSTATICS */
 691
 692             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 693             ewrt             = _mm_mul_ps(r00,ewtabscale);
 694             ewitab           = _mm_cvttps_epi32(ewrt);
 695             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 696             ewitab           = _mm_slli_epi32(ewitab,2);
 697             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 698             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 699             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 700             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 701             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 702             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 703             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 704             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 705             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 706
 707             d                = _mm_sub_ps(r00,rswitch);
 708             d                = _mm_max_ps(d,_mm_setzero_ps());
 709             d2               = _mm_mul_ps(d,d);
 710             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 711
 712             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 713
 714             /* Evaluate switch function */
 715             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 716             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 717             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 718
 719             fscal            = felec;
 720
 721             fscal            = _mm_and_ps(fscal,cutoff_mask);
 722
 723             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 724
 725             /* Calculate temporary vectorial force */
 726             tx               = _mm_mul_ps(fscal,dx00);
 727             ty               = _mm_mul_ps(fscal,dy00);
 728             tz               = _mm_mul_ps(fscal,dz00);
 729
 730             /* Update vectorial force */
 731             fix0             = _mm_add_ps(fix0,tx);
 732             fiy0             = _mm_add_ps(fiy0,ty);
 733             fiz0             = _mm_add_ps(fiz0,tz);
 734
 735             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 736             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 737             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 738             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 739             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 740
 741             }
 742
 743             /* Inner loop uses 63 flops */
 744         }
 745
 746         /* End of innermost loop */
 747
 748         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 749                                               f+i_coord_offset,fshift+i_shift_offset);
 750
 751         /* Increment number of inner iterations */
 752         inneriter                  += j_index_end - j_index_start;
 753
 754         /* Outer loop uses 7 flops */
 755     }
 756
 757     /* Increment number of outer iterations */
 758     outeriter        += nri;
 759
 760     /* Update outer/inner flops */
 761
 762     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*63);
 763 }