src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJ_GeomW4P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          ewitab;
  92     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  93     real             *ewtab;
  94     __m128           dummy_mask,cutoff_mask;
  95     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  96     __m128           one     = _mm_set1_ps(1.0);
  97     __m128           two     = _mm_set1_ps(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm_set1_ps(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 116     ewtab            = fr->ic->tabq_coul_FDV0;
 117     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 118     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 123     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 124     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 125     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 159
 160         fix0             = _mm_setzero_ps();
 161         fiy0             = _mm_setzero_ps();
 162         fiz0             = _mm_setzero_ps();
 163         fix1             = _mm_setzero_ps();
 164         fiy1             = _mm_setzero_ps();
 165         fiz1             = _mm_setzero_ps();
 166         fix2             = _mm_setzero_ps();
 167         fiy2             = _mm_setzero_ps();
 168         fiz2             = _mm_setzero_ps();
 169         fix3             = _mm_setzero_ps();
 170         fiy3             = _mm_setzero_ps();
 171         fiz3             = _mm_setzero_ps();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm_setzero_ps();
 175         vvdwsum          = _mm_setzero_ps();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             jnrC             = jjnr[jidx+2];
 185             jnrD             = jjnr[jidx+3];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               x+j_coord_offsetC,x+j_coord_offsetD,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_ps(ix0,jx0);
 198             dy00             = _mm_sub_ps(iy0,jy0);
 199             dz00             = _mm_sub_ps(iz0,jz0);
 200             dx10             = _mm_sub_ps(ix1,jx0);
 201             dy10             = _mm_sub_ps(iy1,jy0);
 202             dz10             = _mm_sub_ps(iz1,jz0);
 203             dx20             = _mm_sub_ps(ix2,jx0);
 204             dy20             = _mm_sub_ps(iy2,jy0);
 205             dz20             = _mm_sub_ps(iz2,jz0);
 206             dx30             = _mm_sub_ps(ix3,jx0);
 207             dy30             = _mm_sub_ps(iy3,jy0);
 208             dz30             = _mm_sub_ps(iz3,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 212             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 213             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 214             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 215
 216             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 217             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 218             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 219
 220             rinvsq00         = gmx_mm_inv_ps(rsq00);
 221             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 223             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             fjx0             = _mm_setzero_ps();
 234             fjy0             = _mm_setzero_ps();
 235             fjz0             = _mm_setzero_ps();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 243                                          vdwparam+vdwioffset0+vdwjidx0B,
 244                                          vdwparam+vdwioffset0+vdwjidx0C,
 245                                          vdwparam+vdwioffset0+vdwjidx0D,
 246                                          &c6_00,&c12_00);
 247
 248             /* LENNARD-JONES DISPERSION/REPULSION */
 249
 250             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 251             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 252             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 253             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 254             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 255
 256             /* Update potential sum for this i atom from the interaction with this j atom. */
 257             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 258
 259             fscal            = fvdw;
 260
 261             /* Calculate temporary vectorial force */
 262             tx               = _mm_mul_ps(fscal,dx00);
 263             ty               = _mm_mul_ps(fscal,dy00);
 264             tz               = _mm_mul_ps(fscal,dz00);
 265
 266             /* Update vectorial force */
 267             fix0             = _mm_add_ps(fix0,tx);
 268             fiy0             = _mm_add_ps(fiy0,ty);
 269             fiz0             = _mm_add_ps(fiz0,tz);
 270
 271             fjx0             = _mm_add_ps(fjx0,tx);
 272             fjy0             = _mm_add_ps(fjy0,ty);
 273             fjz0             = _mm_add_ps(fjz0,tz);
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             r10              = _mm_mul_ps(rsq10,rinv10);
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             qq10             = _mm_mul_ps(iq1,jq0);
 283
 284             /* EWALD ELECTROSTATICS */
 285
 286             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 287             ewrt             = _mm_mul_ps(r10,ewtabscale);
 288             ewitab           = _mm_cvttps_epi32(ewrt);
 289             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 290             ewitab           = _mm_slli_epi32(ewitab,2);
 291             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 292             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 293             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 294             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 295             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 296             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 297             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 298             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 299             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velecsum         = _mm_add_ps(velecsum,velec);
 303
 304             fscal            = felec;
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_ps(fscal,dx10);
 308             ty               = _mm_mul_ps(fscal,dy10);
 309             tz               = _mm_mul_ps(fscal,dz10);
 310
 311             /* Update vectorial force */
 312             fix1             = _mm_add_ps(fix1,tx);
 313             fiy1             = _mm_add_ps(fiy1,ty);
 314             fiz1             = _mm_add_ps(fiz1,tz);
 315
 316             fjx0             = _mm_add_ps(fjx0,tx);
 317             fjy0             = _mm_add_ps(fjy0,ty);
 318             fjz0             = _mm_add_ps(fjz0,tz);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r20              = _mm_mul_ps(rsq20,rinv20);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq20             = _mm_mul_ps(iq2,jq0);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _mm_mul_ps(r20,ewtabscale);
 333             ewitab           = _mm_cvttps_epi32(ewrt);
 334             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 335             ewitab           = _mm_slli_epi32(ewitab,2);
 336             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 337             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 338             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 339             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 340             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 341             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 342             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 343             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 344             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_ps(fscal,dx20);
 353             ty               = _mm_mul_ps(fscal,dy20);
 354             tz               = _mm_mul_ps(fscal,dz20);
 355
 356             /* Update vectorial force */
 357             fix2             = _mm_add_ps(fix2,tx);
 358             fiy2             = _mm_add_ps(fiy2,ty);
 359             fiz2             = _mm_add_ps(fiz2,tz);
 360
 361             fjx0             = _mm_add_ps(fjx0,tx);
 362             fjy0             = _mm_add_ps(fjy0,ty);
 363             fjz0             = _mm_add_ps(fjz0,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             r30              = _mm_mul_ps(rsq30,rinv30);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq30             = _mm_mul_ps(iq3,jq0);
 373
 374             /* EWALD ELECTROSTATICS */
 375
 376             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 377             ewrt             = _mm_mul_ps(r30,ewtabscale);
 378             ewitab           = _mm_cvttps_epi32(ewrt);
 379             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 380             ewitab           = _mm_slli_epi32(ewitab,2);
 381             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 382             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 383             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 384             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 385             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 386             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 387             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 388             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 389             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 390
 391             /* Update potential sum for this i atom from the interaction with this j atom. */
 392             velecsum         = _mm_add_ps(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             /* Calculate temporary vectorial force */
 397             tx               = _mm_mul_ps(fscal,dx30);
 398             ty               = _mm_mul_ps(fscal,dy30);
 399             tz               = _mm_mul_ps(fscal,dz30);
 400
 401             /* Update vectorial force */
 402             fix3             = _mm_add_ps(fix3,tx);
 403             fiy3             = _mm_add_ps(fiy3,ty);
 404             fiz3             = _mm_add_ps(fiz3,tz);
 405
 406             fjx0             = _mm_add_ps(fjx0,tx);
 407             fjy0             = _mm_add_ps(fjy0,ty);
 408             fjz0             = _mm_add_ps(fjz0,tz);
 409
 410             fjptrA             = f+j_coord_offsetA;
 411             fjptrB             = f+j_coord_offsetB;
 412             fjptrC             = f+j_coord_offsetC;
 413             fjptrD             = f+j_coord_offsetD;
 414
 415             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 416
 417             /* Inner loop uses 155 flops */
 418         }
 419
 420         if(jidx<j_index_end)
 421         {
 422
 423             /* Get j neighbor index, and coordinate index */
 424             jnrlistA         = jjnr[jidx];
 425             jnrlistB         = jjnr[jidx+1];
 426             jnrlistC         = jjnr[jidx+2];
 427             jnrlistD         = jjnr[jidx+3];
 428             /* Sign of each element will be negative for non-real atoms.
 429              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 430              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 431              */
 432             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 433             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 434             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 435             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 436             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 437             j_coord_offsetA  = DIM*jnrA;
 438             j_coord_offsetB  = DIM*jnrB;
 439             j_coord_offsetC  = DIM*jnrC;
 440             j_coord_offsetD  = DIM*jnrD;
 441
 442             /* load j atom coordinates */
 443             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                               x+j_coord_offsetC,x+j_coord_offsetD,
 445                                               &jx0,&jy0,&jz0);
 446
 447             /* Calculate displacement vector */
 448             dx00             = _mm_sub_ps(ix0,jx0);
 449             dy00             = _mm_sub_ps(iy0,jy0);
 450             dz00             = _mm_sub_ps(iz0,jz0);
 451             dx10             = _mm_sub_ps(ix1,jx0);
 452             dy10             = _mm_sub_ps(iy1,jy0);
 453             dz10             = _mm_sub_ps(iz1,jz0);
 454             dx20             = _mm_sub_ps(ix2,jx0);
 455             dy20             = _mm_sub_ps(iy2,jy0);
 456             dz20             = _mm_sub_ps(iz2,jz0);
 457             dx30             = _mm_sub_ps(ix3,jx0);
 458             dy30             = _mm_sub_ps(iy3,jy0);
 459             dz30             = _mm_sub_ps(iz3,jz0);
 460
 461             /* Calculate squared distance and things based on it */
 462             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 463             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 464             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 465             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 466
 467             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 468             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 469             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 470
 471             rinvsq00         = gmx_mm_inv_ps(rsq00);
 472             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 473             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 474             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 475
 476             /* Load parameters for j particles */
 477             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 478                                                               charge+jnrC+0,charge+jnrD+0);
 479             vdwjidx0A        = 2*vdwtype[jnrA+0];
 480             vdwjidx0B        = 2*vdwtype[jnrB+0];
 481             vdwjidx0C        = 2*vdwtype[jnrC+0];
 482             vdwjidx0D        = 2*vdwtype[jnrD+0];
 483
 484             fjx0             = _mm_setzero_ps();
 485             fjy0             = _mm_setzero_ps();
 486             fjz0             = _mm_setzero_ps();
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             /* Compute parameters for interactions between i and j atoms */
 493             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 494                                          vdwparam+vdwioffset0+vdwjidx0B,
 495                                          vdwparam+vdwioffset0+vdwjidx0C,
 496                                          vdwparam+vdwioffset0+vdwjidx0D,
 497                                          &c6_00,&c12_00);
 498
 499             /* LENNARD-JONES DISPERSION/REPULSION */
 500
 501             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 502             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 503             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 504             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 505             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 506
 507             /* Update potential sum for this i atom from the interaction with this j atom. */
 508             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 509             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 510
 511             fscal            = fvdw;
 512
 513             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 514
 515             /* Calculate temporary vectorial force */
 516             tx               = _mm_mul_ps(fscal,dx00);
 517             ty               = _mm_mul_ps(fscal,dy00);
 518             tz               = _mm_mul_ps(fscal,dz00);
 519
 520             /* Update vectorial force */
 521             fix0             = _mm_add_ps(fix0,tx);
 522             fiy0             = _mm_add_ps(fiy0,ty);
 523             fiz0             = _mm_add_ps(fiz0,tz);
 524
 525             fjx0             = _mm_add_ps(fjx0,tx);
 526             fjy0             = _mm_add_ps(fjy0,ty);
 527             fjz0             = _mm_add_ps(fjz0,tz);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r10              = _mm_mul_ps(rsq10,rinv10);
 534             r10              = _mm_andnot_ps(dummy_mask,r10);
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq10             = _mm_mul_ps(iq1,jq0);
 538
 539             /* EWALD ELECTROSTATICS */
 540
 541             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 542             ewrt             = _mm_mul_ps(r10,ewtabscale);
 543             ewitab           = _mm_cvttps_epi32(ewrt);
 544             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 545             ewitab           = _mm_slli_epi32(ewitab,2);
 546             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 547             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 548             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 549             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 550             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 551             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 552             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 553             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 554             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_andnot_ps(dummy_mask,velec);
 558             velecsum         = _mm_add_ps(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_ps(fscal,dx10);
 566             ty               = _mm_mul_ps(fscal,dy10);
 567             tz               = _mm_mul_ps(fscal,dz10);
 568
 569             /* Update vectorial force */
 570             fix1             = _mm_add_ps(fix1,tx);
 571             fiy1             = _mm_add_ps(fiy1,ty);
 572             fiz1             = _mm_add_ps(fiz1,tz);
 573
 574             fjx0             = _mm_add_ps(fjx0,tx);
 575             fjy0             = _mm_add_ps(fjy0,ty);
 576             fjz0             = _mm_add_ps(fjz0,tz);
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             r20              = _mm_mul_ps(rsq20,rinv20);
 583             r20              = _mm_andnot_ps(dummy_mask,r20);
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             qq20             = _mm_mul_ps(iq2,jq0);
 587
 588             /* EWALD ELECTROSTATICS */
 589
 590             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 591             ewrt             = _mm_mul_ps(r20,ewtabscale);
 592             ewitab           = _mm_cvttps_epi32(ewrt);
 593             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 594             ewitab           = _mm_slli_epi32(ewitab,2);
 595             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 596             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 597             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 598             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 599             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 600             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 601             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 602             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 603             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_andnot_ps(dummy_mask,velec);
 607             velecsum         = _mm_add_ps(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_ps(fscal,dx20);
 615             ty               = _mm_mul_ps(fscal,dy20);
 616             tz               = _mm_mul_ps(fscal,dz20);
 617
 618             /* Update vectorial force */
 619             fix2             = _mm_add_ps(fix2,tx);
 620             fiy2             = _mm_add_ps(fiy2,ty);
 621             fiz2             = _mm_add_ps(fiz2,tz);
 622
 623             fjx0             = _mm_add_ps(fjx0,tx);
 624             fjy0             = _mm_add_ps(fjy0,ty);
 625             fjz0             = _mm_add_ps(fjz0,tz);
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             r30              = _mm_mul_ps(rsq30,rinv30);
 632             r30              = _mm_andnot_ps(dummy_mask,r30);
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             qq30             = _mm_mul_ps(iq3,jq0);
 636
 637             /* EWALD ELECTROSTATICS */
 638
 639             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 640             ewrt             = _mm_mul_ps(r30,ewtabscale);
 641             ewitab           = _mm_cvttps_epi32(ewrt);
 642             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 643             ewitab           = _mm_slli_epi32(ewitab,2);
 644             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 645             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 646             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 647             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 648             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 649             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 650             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 651             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 652             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 653
 654             /* Update potential sum for this i atom from the interaction with this j atom. */
 655             velec            = _mm_andnot_ps(dummy_mask,velec);
 656             velecsum         = _mm_add_ps(velecsum,velec);
 657
 658             fscal            = felec;
 659
 660             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 661
 662             /* Calculate temporary vectorial force */
 663             tx               = _mm_mul_ps(fscal,dx30);
 664             ty               = _mm_mul_ps(fscal,dy30);
 665             tz               = _mm_mul_ps(fscal,dz30);
 666
 667             /* Update vectorial force */
 668             fix3             = _mm_add_ps(fix3,tx);
 669             fiy3             = _mm_add_ps(fiy3,ty);
 670             fiz3             = _mm_add_ps(fiz3,tz);
 671
 672             fjx0             = _mm_add_ps(fjx0,tx);
 673             fjy0             = _mm_add_ps(fjy0,ty);
 674             fjz0             = _mm_add_ps(fjz0,tz);
 675
 676             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 677             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 678             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 679             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 680
 681             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 682
 683             /* Inner loop uses 158 flops */
 684         }
 685
 686         /* End of innermost loop */
 687
 688         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 689                                               f+i_coord_offset,fshift+i_shift_offset);
 690
 691         ggid                        = gid[iidx];
 692         /* Update potential energies */
 693         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 694         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 695
 696         /* Increment number of inner iterations */
 697         inneriter                  += j_index_end - j_index_start;
 698
 699         /* Outer loop uses 26 flops */
 700     }
 701
 702     /* Increment number of outer iterations */
 703     outeriter        += nri;
 704
 705     /* Update outer/inner flops */
 706
 707     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*158);
 708 }
 709 /*
 710  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_single
 711  * Electrostatics interaction: Ewald
 712  * VdW interaction:            LennardJones
 713  * Geometry:                   Water4-Particle
 714  * Calculate force/pot:        Force
 715  */
 716 void
 717 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_single
 718                     (t_nblist * gmx_restrict                nlist,
 719                      rvec * gmx_restrict                    xx,
 720                      rvec * gmx_restrict                    ff,
 721                      t_forcerec * gmx_restrict              fr,
 722                      t_mdatoms * gmx_restrict               mdatoms,
 723                      nb_kernel_data_t * gmx_restrict        kernel_data,
 724                      t_nrnb * gmx_restrict                  nrnb)
 725 {
 726     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 727      * just 0 for non-waters.
 728      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 729      * jnr indices corresponding to data put in the four positions in the SIMD register.
 730      */
 731     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 732     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 733     int              jnrA,jnrB,jnrC,jnrD;
 734     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 735     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 736     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 737     real             rcutoff_scalar;
 738     real             *shiftvec,*fshift,*x,*f;
 739     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 740     real             scratch[4*DIM];
 741     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 742     int              vdwioffset0;
 743     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 744     int              vdwioffset1;
 745     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 746     int              vdwioffset2;
 747     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 748     int              vdwioffset3;
 749     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 750     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 751     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 752     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 753     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 754     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 755     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 756     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 757     real             *charge;
 758     int              nvdwtype;
 759     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 760     int              *vdwtype;
 761     real             *vdwparam;
 762     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 763     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 764     __m128i          ewitab;
 765     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 766     real             *ewtab;
 767     __m128           dummy_mask,cutoff_mask;
 768     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 769     __m128           one     = _mm_set1_ps(1.0);
 770     __m128           two     = _mm_set1_ps(2.0);
 771     x                = xx[0];
 772     f                = ff[0];
 773
 774     nri              = nlist->nri;
 775     iinr             = nlist->iinr;
 776     jindex           = nlist->jindex;
 777     jjnr             = nlist->jjnr;
 778     shiftidx         = nlist->shift;
 779     gid              = nlist->gid;
 780     shiftvec         = fr->shift_vec[0];
 781     fshift           = fr->fshift[0];
 782     facel            = _mm_set1_ps(fr->epsfac);
 783     charge           = mdatoms->chargeA;
 784     nvdwtype         = fr->ntype;
 785     vdwparam         = fr->nbfp;
 786     vdwtype          = mdatoms->typeA;
 787
 788     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 789     ewtab            = fr->ic->tabq_coul_F;
 790     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 791     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 792
 793     /* Setup water-specific parameters */
 794     inr              = nlist->iinr[0];
 795     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 796     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 797     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 798     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 799
 800     /* Avoid stupid compiler warnings */
 801     jnrA = jnrB = jnrC = jnrD = 0;
 802     j_coord_offsetA = 0;
 803     j_coord_offsetB = 0;
 804     j_coord_offsetC = 0;
 805     j_coord_offsetD = 0;
 806
 807     outeriter        = 0;
 808     inneriter        = 0;
 809
 810     for(iidx=0;iidx<4*DIM;iidx++)
 811     {
 812         scratch[iidx] = 0.0;
 813     }
 814
 815     /* Start outer loop over neighborlists */
 816     for(iidx=0; iidx<nri; iidx++)
 817     {
 818         /* Load shift vector for this list */
 819         i_shift_offset   = DIM*shiftidx[iidx];
 820
 821         /* Load limits for loop over neighbors */
 822         j_index_start    = jindex[iidx];
 823         j_index_end      = jindex[iidx+1];
 824
 825         /* Get outer coordinate index */
 826         inr              = iinr[iidx];
 827         i_coord_offset   = DIM*inr;
 828
 829         /* Load i particle coords and add shift vector */
 830         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 831                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 832
 833         fix0             = _mm_setzero_ps();
 834         fiy0             = _mm_setzero_ps();
 835         fiz0             = _mm_setzero_ps();
 836         fix1             = _mm_setzero_ps();
 837         fiy1             = _mm_setzero_ps();
 838         fiz1             = _mm_setzero_ps();
 839         fix2             = _mm_setzero_ps();
 840         fiy2             = _mm_setzero_ps();
 841         fiz2             = _mm_setzero_ps();
 842         fix3             = _mm_setzero_ps();
 843         fiy3             = _mm_setzero_ps();
 844         fiz3             = _mm_setzero_ps();
 845
 846         /* Start inner kernel loop */
 847         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 848         {
 849
 850             /* Get j neighbor index, and coordinate index */
 851             jnrA             = jjnr[jidx];
 852             jnrB             = jjnr[jidx+1];
 853             jnrC             = jjnr[jidx+2];
 854             jnrD             = jjnr[jidx+3];
 855             j_coord_offsetA  = DIM*jnrA;
 856             j_coord_offsetB  = DIM*jnrB;
 857             j_coord_offsetC  = DIM*jnrC;
 858             j_coord_offsetD  = DIM*jnrD;
 859
 860             /* load j atom coordinates */
 861             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 862                                               x+j_coord_offsetC,x+j_coord_offsetD,
 863                                               &jx0,&jy0,&jz0);
 864
 865             /* Calculate displacement vector */
 866             dx00             = _mm_sub_ps(ix0,jx0);
 867             dy00             = _mm_sub_ps(iy0,jy0);
 868             dz00             = _mm_sub_ps(iz0,jz0);
 869             dx10             = _mm_sub_ps(ix1,jx0);
 870             dy10             = _mm_sub_ps(iy1,jy0);
 871             dz10             = _mm_sub_ps(iz1,jz0);
 872             dx20             = _mm_sub_ps(ix2,jx0);
 873             dy20             = _mm_sub_ps(iy2,jy0);
 874             dz20             = _mm_sub_ps(iz2,jz0);
 875             dx30             = _mm_sub_ps(ix3,jx0);
 876             dy30             = _mm_sub_ps(iy3,jy0);
 877             dz30             = _mm_sub_ps(iz3,jz0);
 878
 879             /* Calculate squared distance and things based on it */
 880             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 881             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 882             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 883             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 884
 885             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 886             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 887             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 888
 889             rinvsq00         = gmx_mm_inv_ps(rsq00);
 890             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 891             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 892             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 893
 894             /* Load parameters for j particles */
 895             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 896                                                               charge+jnrC+0,charge+jnrD+0);
 897             vdwjidx0A        = 2*vdwtype[jnrA+0];
 898             vdwjidx0B        = 2*vdwtype[jnrB+0];
 899             vdwjidx0C        = 2*vdwtype[jnrC+0];
 900             vdwjidx0D        = 2*vdwtype[jnrD+0];
 901
 902             fjx0             = _mm_setzero_ps();
 903             fjy0             = _mm_setzero_ps();
 904             fjz0             = _mm_setzero_ps();
 905
 906             /**************************
 907              * CALCULATE INTERACTIONS *
 908              **************************/
 909
 910             /* Compute parameters for interactions between i and j atoms */
 911             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 912                                          vdwparam+vdwioffset0+vdwjidx0B,
 913                                          vdwparam+vdwioffset0+vdwjidx0C,
 914                                          vdwparam+vdwioffset0+vdwjidx0D,
 915                                          &c6_00,&c12_00);
 916
 917             /* LENNARD-JONES DISPERSION/REPULSION */
 918
 919             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 920             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 921
 922             fscal            = fvdw;
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm_mul_ps(fscal,dx00);
 926             ty               = _mm_mul_ps(fscal,dy00);
 927             tz               = _mm_mul_ps(fscal,dz00);
 928
 929             /* Update vectorial force */
 930             fix0             = _mm_add_ps(fix0,tx);
 931             fiy0             = _mm_add_ps(fiy0,ty);
 932             fiz0             = _mm_add_ps(fiz0,tz);
 933
 934             fjx0             = _mm_add_ps(fjx0,tx);
 935             fjy0             = _mm_add_ps(fjy0,ty);
 936             fjz0             = _mm_add_ps(fjz0,tz);
 937
 938             /**************************
 939              * CALCULATE INTERACTIONS *
 940              **************************/
 941
 942             r10              = _mm_mul_ps(rsq10,rinv10);
 943
 944             /* Compute parameters for interactions between i and j atoms */
 945             qq10             = _mm_mul_ps(iq1,jq0);
 946
 947             /* EWALD ELECTROSTATICS */
 948
 949             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 950             ewrt             = _mm_mul_ps(r10,ewtabscale);
 951             ewitab           = _mm_cvttps_epi32(ewrt);
 952             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 953             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 954                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 955                                          &ewtabF,&ewtabFn);
 956             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 957             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 958
 959             fscal            = felec;
 960
 961             /* Calculate temporary vectorial force */
 962             tx               = _mm_mul_ps(fscal,dx10);
 963             ty               = _mm_mul_ps(fscal,dy10);
 964             tz               = _mm_mul_ps(fscal,dz10);
 965
 966             /* Update vectorial force */
 967             fix1             = _mm_add_ps(fix1,tx);
 968             fiy1             = _mm_add_ps(fiy1,ty);
 969             fiz1             = _mm_add_ps(fiz1,tz);
 970
 971             fjx0             = _mm_add_ps(fjx0,tx);
 972             fjy0             = _mm_add_ps(fjy0,ty);
 973             fjz0             = _mm_add_ps(fjz0,tz);
 974
 975             /**************************
 976              * CALCULATE INTERACTIONS *
 977              **************************/
 978
 979             r20              = _mm_mul_ps(rsq20,rinv20);
 980
 981             /* Compute parameters for interactions between i and j atoms */
 982             qq20             = _mm_mul_ps(iq2,jq0);
 983
 984             /* EWALD ELECTROSTATICS */
 985
 986             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 987             ewrt             = _mm_mul_ps(r20,ewtabscale);
 988             ewitab           = _mm_cvttps_epi32(ewrt);
 989             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 990             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 991                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 992                                          &ewtabF,&ewtabFn);
 993             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 994             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 995
 996             fscal            = felec;
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm_mul_ps(fscal,dx20);
1000             ty               = _mm_mul_ps(fscal,dy20);
1001             tz               = _mm_mul_ps(fscal,dz20);
1002
1003             /* Update vectorial force */
1004             fix2             = _mm_add_ps(fix2,tx);
1005             fiy2             = _mm_add_ps(fiy2,ty);
1006             fiz2             = _mm_add_ps(fiz2,tz);
1007
1008             fjx0             = _mm_add_ps(fjx0,tx);
1009             fjy0             = _mm_add_ps(fjy0,ty);
1010             fjz0             = _mm_add_ps(fjz0,tz);
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             r30              = _mm_mul_ps(rsq30,rinv30);
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             qq30             = _mm_mul_ps(iq3,jq0);
1020
1021             /* EWALD ELECTROSTATICS */
1022
1023             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1024             ewrt             = _mm_mul_ps(r30,ewtabscale);
1025             ewitab           = _mm_cvttps_epi32(ewrt);
1026             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1027             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1028                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1029                                          &ewtabF,&ewtabFn);
1030             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1031             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1032
1033             fscal            = felec;
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm_mul_ps(fscal,dx30);
1037             ty               = _mm_mul_ps(fscal,dy30);
1038             tz               = _mm_mul_ps(fscal,dz30);
1039
1040             /* Update vectorial force */
1041             fix3             = _mm_add_ps(fix3,tx);
1042             fiy3             = _mm_add_ps(fiy3,ty);
1043             fiz3             = _mm_add_ps(fiz3,tz);
1044
1045             fjx0             = _mm_add_ps(fjx0,tx);
1046             fjy0             = _mm_add_ps(fjy0,ty);
1047             fjz0             = _mm_add_ps(fjz0,tz);
1048
1049             fjptrA             = f+j_coord_offsetA;
1050             fjptrB             = f+j_coord_offsetB;
1051             fjptrC             = f+j_coord_offsetC;
1052             fjptrD             = f+j_coord_offsetD;
1053
1054             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1055
1056             /* Inner loop uses 135 flops */
1057         }
1058
1059         if(jidx<j_index_end)
1060         {
1061
1062             /* Get j neighbor index, and coordinate index */
1063             jnrlistA         = jjnr[jidx];
1064             jnrlistB         = jjnr[jidx+1];
1065             jnrlistC         = jjnr[jidx+2];
1066             jnrlistD         = jjnr[jidx+3];
1067             /* Sign of each element will be negative for non-real atoms.
1068              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1069              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1070              */
1071             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1072             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1073             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1074             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1075             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1076             j_coord_offsetA  = DIM*jnrA;
1077             j_coord_offsetB  = DIM*jnrB;
1078             j_coord_offsetC  = DIM*jnrC;
1079             j_coord_offsetD  = DIM*jnrD;
1080
1081             /* load j atom coordinates */
1082             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1083                                               x+j_coord_offsetC,x+j_coord_offsetD,
1084                                               &jx0,&jy0,&jz0);
1085
1086             /* Calculate displacement vector */
1087             dx00             = _mm_sub_ps(ix0,jx0);
1088             dy00             = _mm_sub_ps(iy0,jy0);
1089             dz00             = _mm_sub_ps(iz0,jz0);
1090             dx10             = _mm_sub_ps(ix1,jx0);
1091             dy10             = _mm_sub_ps(iy1,jy0);
1092             dz10             = _mm_sub_ps(iz1,jz0);
1093             dx20             = _mm_sub_ps(ix2,jx0);
1094             dy20             = _mm_sub_ps(iy2,jy0);
1095             dz20             = _mm_sub_ps(iz2,jz0);
1096             dx30             = _mm_sub_ps(ix3,jx0);
1097             dy30             = _mm_sub_ps(iy3,jy0);
1098             dz30             = _mm_sub_ps(iz3,jz0);
1099
1100             /* Calculate squared distance and things based on it */
1101             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1102             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1103             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1104             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1105
1106             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1107             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1108             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1109
1110             rinvsq00         = gmx_mm_inv_ps(rsq00);
1111             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1112             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1113             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1114
1115             /* Load parameters for j particles */
1116             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1117                                                               charge+jnrC+0,charge+jnrD+0);
1118             vdwjidx0A        = 2*vdwtype[jnrA+0];
1119             vdwjidx0B        = 2*vdwtype[jnrB+0];
1120             vdwjidx0C        = 2*vdwtype[jnrC+0];
1121             vdwjidx0D        = 2*vdwtype[jnrD+0];
1122
1123             fjx0             = _mm_setzero_ps();
1124             fjy0             = _mm_setzero_ps();
1125             fjz0             = _mm_setzero_ps();
1126
1127             /**************************
1128              * CALCULATE INTERACTIONS *
1129              **************************/
1130
1131             /* Compute parameters for interactions between i and j atoms */
1132             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1133                                          vdwparam+vdwioffset0+vdwjidx0B,
1134                                          vdwparam+vdwioffset0+vdwjidx0C,
1135                                          vdwparam+vdwioffset0+vdwjidx0D,
1136                                          &c6_00,&c12_00);
1137
1138             /* LENNARD-JONES DISPERSION/REPULSION */
1139
1140             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1141             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1142
1143             fscal            = fvdw;
1144
1145             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1146
1147             /* Calculate temporary vectorial force */
1148             tx               = _mm_mul_ps(fscal,dx00);
1149             ty               = _mm_mul_ps(fscal,dy00);
1150             tz               = _mm_mul_ps(fscal,dz00);
1151
1152             /* Update vectorial force */
1153             fix0             = _mm_add_ps(fix0,tx);
1154             fiy0             = _mm_add_ps(fiy0,ty);
1155             fiz0             = _mm_add_ps(fiz0,tz);
1156
1157             fjx0             = _mm_add_ps(fjx0,tx);
1158             fjy0             = _mm_add_ps(fjy0,ty);
1159             fjz0             = _mm_add_ps(fjz0,tz);
1160
1161             /**************************
1162              * CALCULATE INTERACTIONS *
1163              **************************/
1164
1165             r10              = _mm_mul_ps(rsq10,rinv10);
1166             r10              = _mm_andnot_ps(dummy_mask,r10);
1167
1168             /* Compute parameters for interactions between i and j atoms */
1169             qq10             = _mm_mul_ps(iq1,jq0);
1170
1171             /* EWALD ELECTROSTATICS */
1172
1173             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1174             ewrt             = _mm_mul_ps(r10,ewtabscale);
1175             ewitab           = _mm_cvttps_epi32(ewrt);
1176             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1177             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1178                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1179                                          &ewtabF,&ewtabFn);
1180             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1181             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1182
1183             fscal            = felec;
1184
1185             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1186
1187             /* Calculate temporary vectorial force */
1188             tx               = _mm_mul_ps(fscal,dx10);
1189             ty               = _mm_mul_ps(fscal,dy10);
1190             tz               = _mm_mul_ps(fscal,dz10);
1191
1192             /* Update vectorial force */
1193             fix1             = _mm_add_ps(fix1,tx);
1194             fiy1             = _mm_add_ps(fiy1,ty);
1195             fiz1             = _mm_add_ps(fiz1,tz);
1196
1197             fjx0             = _mm_add_ps(fjx0,tx);
1198             fjy0             = _mm_add_ps(fjy0,ty);
1199             fjz0             = _mm_add_ps(fjz0,tz);
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             r20              = _mm_mul_ps(rsq20,rinv20);
1206             r20              = _mm_andnot_ps(dummy_mask,r20);
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             qq20             = _mm_mul_ps(iq2,jq0);
1210
1211             /* EWALD ELECTROSTATICS */
1212
1213             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1214             ewrt             = _mm_mul_ps(r20,ewtabscale);
1215             ewitab           = _mm_cvttps_epi32(ewrt);
1216             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1217             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1218                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1219                                          &ewtabF,&ewtabFn);
1220             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1221             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1222
1223             fscal            = felec;
1224
1225             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1226
1227             /* Calculate temporary vectorial force */
1228             tx               = _mm_mul_ps(fscal,dx20);
1229             ty               = _mm_mul_ps(fscal,dy20);
1230             tz               = _mm_mul_ps(fscal,dz20);
1231
1232             /* Update vectorial force */
1233             fix2             = _mm_add_ps(fix2,tx);
1234             fiy2             = _mm_add_ps(fiy2,ty);
1235             fiz2             = _mm_add_ps(fiz2,tz);
1236
1237             fjx0             = _mm_add_ps(fjx0,tx);
1238             fjy0             = _mm_add_ps(fjy0,ty);
1239             fjz0             = _mm_add_ps(fjz0,tz);
1240
1241             /**************************
1242              * CALCULATE INTERACTIONS *
1243              **************************/
1244
1245             r30              = _mm_mul_ps(rsq30,rinv30);
1246             r30              = _mm_andnot_ps(dummy_mask,r30);
1247
1248             /* Compute parameters for interactions between i and j atoms */
1249             qq30             = _mm_mul_ps(iq3,jq0);
1250
1251             /* EWALD ELECTROSTATICS */
1252
1253             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1254             ewrt             = _mm_mul_ps(r30,ewtabscale);
1255             ewitab           = _mm_cvttps_epi32(ewrt);
1256             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1257             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1258                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1259                                          &ewtabF,&ewtabFn);
1260             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1261             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1262
1263             fscal            = felec;
1264
1265             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1266
1267             /* Calculate temporary vectorial force */
1268             tx               = _mm_mul_ps(fscal,dx30);
1269             ty               = _mm_mul_ps(fscal,dy30);
1270             tz               = _mm_mul_ps(fscal,dz30);
1271
1272             /* Update vectorial force */
1273             fix3             = _mm_add_ps(fix3,tx);
1274             fiy3             = _mm_add_ps(fiy3,ty);
1275             fiz3             = _mm_add_ps(fiz3,tz);
1276
1277             fjx0             = _mm_add_ps(fjx0,tx);
1278             fjy0             = _mm_add_ps(fjy0,ty);
1279             fjz0             = _mm_add_ps(fjz0,tz);
1280
1281             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1282             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1283             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1284             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1285
1286             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1287
1288             /* Inner loop uses 138 flops */
1289         }
1290
1291         /* End of innermost loop */
1292
1293         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1294                                               f+i_coord_offset,fshift+i_shift_offset);
1295
1296         /* Increment number of inner iterations */
1297         inneriter                  += j_index_end - j_index_start;
1298
1299         /* Outer loop uses 24 flops */
1300     }
1301
1302     /* Increment number of outer iterations */
1303     outeriter        += nri;
1304
1305     /* Update outer/inner flops */
1306
1307     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*138);
1308 }