src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_sse2_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_double 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_double.h"
  34 #include "kernelutil_x86_sse2_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse2_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse2_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  88     real             *vftab;
  89     __m128i          ewitab;
  90     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     real             *ewtab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 115
 116     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 124     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 125     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 126     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 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_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 152                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 153
 154         fix0             = _mm_setzero_pd();
 155         fiy0             = _mm_setzero_pd();
 156         fiz0             = _mm_setzero_pd();
 157         fix1             = _mm_setzero_pd();
 158         fiy1             = _mm_setzero_pd();
 159         fiz1             = _mm_setzero_pd();
 160         fix2             = _mm_setzero_pd();
 161         fiy2             = _mm_setzero_pd();
 162         fiz2             = _mm_setzero_pd();
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_pd();
 166         vvdwsum          = _mm_setzero_pd();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_pd(ix0,jx0);
 184             dy00             = _mm_sub_pd(iy0,jy0);
 185             dz00             = _mm_sub_pd(iz0,jz0);
 186             dx10             = _mm_sub_pd(ix1,jx0);
 187             dy10             = _mm_sub_pd(iy1,jy0);
 188             dz10             = _mm_sub_pd(iz1,jz0);
 189             dx20             = _mm_sub_pd(ix2,jx0);
 190             dy20             = _mm_sub_pd(iy2,jy0);
 191             dz20             = _mm_sub_pd(iz2,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 195             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 196             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 197
 198             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 199             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 200             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 201
 202             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 203             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 204             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 208             vdwjidx0A        = 2*vdwtype[jnrA+0];
 209             vdwjidx0B        = 2*vdwtype[jnrB+0];
 210
 211             fjx0             = _mm_setzero_pd();
 212             fjy0             = _mm_setzero_pd();
 213             fjz0             = _mm_setzero_pd();
 214
 215             /**************************
 216              * CALCULATE INTERACTIONS *
 217              **************************/
 218
 219             r00              = _mm_mul_pd(rsq00,rinv00);
 220
 221             /* Compute parameters for interactions between i and j atoms */
 222             qq00             = _mm_mul_pd(iq0,jq0);
 223             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 224                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 225
 226             /* Calculate table index by multiplying r with table scale and truncate to integer */
 227             rt               = _mm_mul_pd(r00,vftabscale);
 228             vfitab           = _mm_cvttpd_epi32(rt);
 229             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 230             vfitab           = _mm_slli_epi32(vfitab,3);
 231
 232             /* EWALD ELECTROSTATICS */
 233
 234             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 235             ewrt             = _mm_mul_pd(r00,ewtabscale);
 236             ewitab           = _mm_cvttpd_epi32(ewrt);
 237             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 238             ewitab           = _mm_slli_epi32(ewitab,2);
 239             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 240             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 241             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 242             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 243             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 244             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 245             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 246             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 247             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 248             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 249
 250             /* CUBIC SPLINE TABLE DISPERSION */
 251             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 252             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 253             GMX_MM_TRANSPOSE2_PD(Y,F);
 254             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 255             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 256             GMX_MM_TRANSPOSE2_PD(G,H);
 257             Heps             = _mm_mul_pd(vfeps,H);
 258             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 259             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 260             vvdw6            = _mm_mul_pd(c6_00,VV);
 261             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 262             fvdw6            = _mm_mul_pd(c6_00,FF);
 263
 264             /* CUBIC SPLINE TABLE REPULSION */
 265             vfitab           = _mm_add_epi32(vfitab,ifour);
 266             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 267             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 268             GMX_MM_TRANSPOSE2_PD(Y,F);
 269             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 270             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 271             GMX_MM_TRANSPOSE2_PD(G,H);
 272             Heps             = _mm_mul_pd(vfeps,H);
 273             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 274             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 275             vvdw12           = _mm_mul_pd(c12_00,VV);
 276             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 277             fvdw12           = _mm_mul_pd(c12_00,FF);
 278             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 279             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_pd(velecsum,velec);
 283             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 284
 285             fscal            = _mm_add_pd(felec,fvdw);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_pd(fscal,dx00);
 289             ty               = _mm_mul_pd(fscal,dy00);
 290             tz               = _mm_mul_pd(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_add_pd(fix0,tx);
 294             fiy0             = _mm_add_pd(fiy0,ty);
 295             fiz0             = _mm_add_pd(fiz0,tz);
 296
 297             fjx0             = _mm_add_pd(fjx0,tx);
 298             fjy0             = _mm_add_pd(fjy0,ty);
 299             fjz0             = _mm_add_pd(fjz0,tz);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             r10              = _mm_mul_pd(rsq10,rinv10);
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm_mul_pd(iq1,jq0);
 309
 310             /* EWALD ELECTROSTATICS */
 311
 312             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 313             ewrt             = _mm_mul_pd(r10,ewtabscale);
 314             ewitab           = _mm_cvttpd_epi32(ewrt);
 315             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 316             ewitab           = _mm_slli_epi32(ewitab,2);
 317             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 318             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 319             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 320             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 321             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 322             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 323             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 324             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 325             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 326             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_pd(fscal,dx10);
 335             ty               = _mm_mul_pd(fscal,dy10);
 336             tz               = _mm_mul_pd(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm_add_pd(fix1,tx);
 340             fiy1             = _mm_add_pd(fiy1,ty);
 341             fiz1             = _mm_add_pd(fiz1,tz);
 342
 343             fjx0             = _mm_add_pd(fjx0,tx);
 344             fjy0             = _mm_add_pd(fjy0,ty);
 345             fjz0             = _mm_add_pd(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r20              = _mm_mul_pd(rsq20,rinv20);
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_pd(iq2,jq0);
 355
 356             /* EWALD ELECTROSTATICS */
 357
 358             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 359             ewrt             = _mm_mul_pd(r20,ewtabscale);
 360             ewitab           = _mm_cvttpd_epi32(ewrt);
 361             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 362             ewitab           = _mm_slli_epi32(ewitab,2);
 363             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 364             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 365             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 366             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 367             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 368             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 369             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 370             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 371             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 372             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velecsum         = _mm_add_pd(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm_mul_pd(fscal,dx20);
 381             ty               = _mm_mul_pd(fscal,dy20);
 382             tz               = _mm_mul_pd(fscal,dz20);
 383
 384             /* Update vectorial force */
 385             fix2             = _mm_add_pd(fix2,tx);
 386             fiy2             = _mm_add_pd(fiy2,ty);
 387             fiz2             = _mm_add_pd(fiz2,tz);
 388
 389             fjx0             = _mm_add_pd(fjx0,tx);
 390             fjy0             = _mm_add_pd(fjy0,ty);
 391             fjz0             = _mm_add_pd(fjz0,tz);
 392
 393             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 394
 395             /* Inner loop uses 160 flops */
 396         }
 397
 398         if(jidx<j_index_end)
 399         {
 400
 401             jnrA             = jjnr[jidx];
 402             j_coord_offsetA  = DIM*jnrA;
 403
 404             /* load j atom coordinates */
 405             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 406                                               &jx0,&jy0,&jz0);
 407
 408             /* Calculate displacement vector */
 409             dx00             = _mm_sub_pd(ix0,jx0);
 410             dy00             = _mm_sub_pd(iy0,jy0);
 411             dz00             = _mm_sub_pd(iz0,jz0);
 412             dx10             = _mm_sub_pd(ix1,jx0);
 413             dy10             = _mm_sub_pd(iy1,jy0);
 414             dz10             = _mm_sub_pd(iz1,jz0);
 415             dx20             = _mm_sub_pd(ix2,jx0);
 416             dy20             = _mm_sub_pd(iy2,jy0);
 417             dz20             = _mm_sub_pd(iz2,jz0);
 418
 419             /* Calculate squared distance and things based on it */
 420             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 421             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 422             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 423
 424             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 425             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 426             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 427
 428             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 429             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 430             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 431
 432             /* Load parameters for j particles */
 433             jq0              = _mm_load_sd(charge+jnrA+0);
 434             vdwjidx0A        = 2*vdwtype[jnrA+0];
 435
 436             fjx0             = _mm_setzero_pd();
 437             fjy0             = _mm_setzero_pd();
 438             fjz0             = _mm_setzero_pd();
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             r00              = _mm_mul_pd(rsq00,rinv00);
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             qq00             = _mm_mul_pd(iq0,jq0);
 448             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 449
 450             /* Calculate table index by multiplying r with table scale and truncate to integer */
 451             rt               = _mm_mul_pd(r00,vftabscale);
 452             vfitab           = _mm_cvttpd_epi32(rt);
 453             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 454             vfitab           = _mm_slli_epi32(vfitab,3);
 455
 456             /* EWALD ELECTROSTATICS */
 457
 458             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 459             ewrt             = _mm_mul_pd(r00,ewtabscale);
 460             ewitab           = _mm_cvttpd_epi32(ewrt);
 461             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 462             ewitab           = _mm_slli_epi32(ewitab,2);
 463             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 464             ewtabD           = _mm_setzero_pd();
 465             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 466             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 467             ewtabFn          = _mm_setzero_pd();
 468             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 469             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 470             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 471             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 472             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 473
 474             /* CUBIC SPLINE TABLE DISPERSION */
 475             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 476             F                = _mm_setzero_pd();
 477             GMX_MM_TRANSPOSE2_PD(Y,F);
 478             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 479             H                = _mm_setzero_pd();
 480             GMX_MM_TRANSPOSE2_PD(G,H);
 481             Heps             = _mm_mul_pd(vfeps,H);
 482             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 483             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 484             vvdw6            = _mm_mul_pd(c6_00,VV);
 485             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 486             fvdw6            = _mm_mul_pd(c6_00,FF);
 487
 488             /* CUBIC SPLINE TABLE REPULSION */
 489             vfitab           = _mm_add_epi32(vfitab,ifour);
 490             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 491             F                = _mm_setzero_pd();
 492             GMX_MM_TRANSPOSE2_PD(Y,F);
 493             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 494             H                = _mm_setzero_pd();
 495             GMX_MM_TRANSPOSE2_PD(G,H);
 496             Heps             = _mm_mul_pd(vfeps,H);
 497             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 498             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 499             vvdw12           = _mm_mul_pd(c12_00,VV);
 500             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 501             fvdw12           = _mm_mul_pd(c12_00,FF);
 502             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 503             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 507             velecsum         = _mm_add_pd(velecsum,velec);
 508             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 509             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 510
 511             fscal            = _mm_add_pd(felec,fvdw);
 512
 513             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 514
 515             /* Calculate temporary vectorial force */
 516             tx               = _mm_mul_pd(fscal,dx00);
 517             ty               = _mm_mul_pd(fscal,dy00);
 518             tz               = _mm_mul_pd(fscal,dz00);
 519
 520             /* Update vectorial force */
 521             fix0             = _mm_add_pd(fix0,tx);
 522             fiy0             = _mm_add_pd(fiy0,ty);
 523             fiz0             = _mm_add_pd(fiz0,tz);
 524
 525             fjx0             = _mm_add_pd(fjx0,tx);
 526             fjy0             = _mm_add_pd(fjy0,ty);
 527             fjz0             = _mm_add_pd(fjz0,tz);
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             r10              = _mm_mul_pd(rsq10,rinv10);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq10             = _mm_mul_pd(iq1,jq0);
 537
 538             /* EWALD ELECTROSTATICS */
 539
 540             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 541             ewrt             = _mm_mul_pd(r10,ewtabscale);
 542             ewitab           = _mm_cvttpd_epi32(ewrt);
 543             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 544             ewitab           = _mm_slli_epi32(ewitab,2);
 545             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 546             ewtabD           = _mm_setzero_pd();
 547             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 548             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 549             ewtabFn          = _mm_setzero_pd();
 550             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 551             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 552             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 553             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 554             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 558             velecsum         = _mm_add_pd(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_pd(fscal,dx10);
 566             ty               = _mm_mul_pd(fscal,dy10);
 567             tz               = _mm_mul_pd(fscal,dz10);
 568
 569             /* Update vectorial force */
 570             fix1             = _mm_add_pd(fix1,tx);
 571             fiy1             = _mm_add_pd(fiy1,ty);
 572             fiz1             = _mm_add_pd(fiz1,tz);
 573
 574             fjx0             = _mm_add_pd(fjx0,tx);
 575             fjy0             = _mm_add_pd(fjy0,ty);
 576             fjz0             = _mm_add_pd(fjz0,tz);
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             r20              = _mm_mul_pd(rsq20,rinv20);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq20             = _mm_mul_pd(iq2,jq0);
 586
 587             /* EWALD ELECTROSTATICS */
 588
 589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 590             ewrt             = _mm_mul_pd(r20,ewtabscale);
 591             ewitab           = _mm_cvttpd_epi32(ewrt);
 592             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 593             ewitab           = _mm_slli_epi32(ewitab,2);
 594             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 595             ewtabD           = _mm_setzero_pd();
 596             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 597             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 598             ewtabFn          = _mm_setzero_pd();
 599             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 600             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 601             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 602             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 603             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 607             velecsum         = _mm_add_pd(velecsum,velec);
 608
 609             fscal            = felec;
 610
 611             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_pd(fscal,dx20);
 615             ty               = _mm_mul_pd(fscal,dy20);
 616             tz               = _mm_mul_pd(fscal,dz20);
 617
 618             /* Update vectorial force */
 619             fix2             = _mm_add_pd(fix2,tx);
 620             fiy2             = _mm_add_pd(fiy2,ty);
 621             fiz2             = _mm_add_pd(fiz2,tz);
 622
 623             fjx0             = _mm_add_pd(fjx0,tx);
 624             fjy0             = _mm_add_pd(fjy0,ty);
 625             fjz0             = _mm_add_pd(fjz0,tz);
 626
 627             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 628
 629             /* Inner loop uses 160 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 635                                               f+i_coord_offset,fshift+i_shift_offset);
 636
 637         ggid                        = gid[iidx];
 638         /* Update potential energies */
 639         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 640         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 641
 642         /* Increment number of inner iterations */
 643         inneriter                  += j_index_end - j_index_start;
 644
 645         /* Outer loop uses 20 flops */
 646     }
 647
 648     /* Increment number of outer iterations */
 649     outeriter        += nri;
 650
 651     /* Update outer/inner flops */
 652
 653     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 654 }
 655 /*
 656  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_double
 657  * Electrostatics interaction: Ewald
 658  * VdW interaction:            CubicSplineTable
 659  * Geometry:                   Water3-Particle
 660  * Calculate force/pot:        Force
 661  */
 662 void
 663 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse2_double
 664                     (t_nblist * gmx_restrict                nlist,
 665                      rvec * gmx_restrict                    xx,
 666                      rvec * gmx_restrict                    ff,
 667                      t_forcerec * gmx_restrict              fr,
 668                      t_mdatoms * gmx_restrict               mdatoms,
 669                      nb_kernel_data_t * gmx_restrict        kernel_data,
 670                      t_nrnb * gmx_restrict                  nrnb)
 671 {
 672     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 673      * just 0 for non-waters.
 674      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 675      * jnr indices corresponding to data put in the four positions in the SIMD register.
 676      */
 677     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 678     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 679     int              jnrA,jnrB;
 680     int              j_coord_offsetA,j_coord_offsetB;
 681     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 682     real             rcutoff_scalar;
 683     real             *shiftvec,*fshift,*x,*f;
 684     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 685     int              vdwioffset0;
 686     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 687     int              vdwioffset1;
 688     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 689     int              vdwioffset2;
 690     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 691     int              vdwjidx0A,vdwjidx0B;
 692     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 693     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 694     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 695     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 696     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 697     real             *charge;
 698     int              nvdwtype;
 699     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 700     int              *vdwtype;
 701     real             *vdwparam;
 702     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 703     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 704     __m128i          vfitab;
 705     __m128i          ifour       = _mm_set1_epi32(4);
 706     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 707     real             *vftab;
 708     __m128i          ewitab;
 709     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 710     real             *ewtab;
 711     __m128d          dummy_mask,cutoff_mask;
 712     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 713     __m128d          one     = _mm_set1_pd(1.0);
 714     __m128d          two     = _mm_set1_pd(2.0);
 715     x                = xx[0];
 716     f                = ff[0];
 717
 718     nri              = nlist->nri;
 719     iinr             = nlist->iinr;
 720     jindex           = nlist->jindex;
 721     jjnr             = nlist->jjnr;
 722     shiftidx         = nlist->shift;
 723     gid              = nlist->gid;
 724     shiftvec         = fr->shift_vec[0];
 725     fshift           = fr->fshift[0];
 726     facel            = _mm_set1_pd(fr->epsfac);
 727     charge           = mdatoms->chargeA;
 728     nvdwtype         = fr->ntype;
 729     vdwparam         = fr->nbfp;
 730     vdwtype          = mdatoms->typeA;
 731
 732     vftab            = kernel_data->table_vdw->data;
 733     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 734
 735     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 736     ewtab            = fr->ic->tabq_coul_F;
 737     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 738     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 739
 740     /* Setup water-specific parameters */
 741     inr              = nlist->iinr[0];
 742     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 743     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 744     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 745     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 746
 747     /* Avoid stupid compiler warnings */
 748     jnrA = jnrB = 0;
 749     j_coord_offsetA = 0;
 750     j_coord_offsetB = 0;
 751
 752     outeriter        = 0;
 753     inneriter        = 0;
 754
 755     /* Start outer loop over neighborlists */
 756     for(iidx=0; iidx<nri; iidx++)
 757     {
 758         /* Load shift vector for this list */
 759         i_shift_offset   = DIM*shiftidx[iidx];
 760
 761         /* Load limits for loop over neighbors */
 762         j_index_start    = jindex[iidx];
 763         j_index_end      = jindex[iidx+1];
 764
 765         /* Get outer coordinate index */
 766         inr              = iinr[iidx];
 767         i_coord_offset   = DIM*inr;
 768
 769         /* Load i particle coords and add shift vector */
 770         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 771                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 772
 773         fix0             = _mm_setzero_pd();
 774         fiy0             = _mm_setzero_pd();
 775         fiz0             = _mm_setzero_pd();
 776         fix1             = _mm_setzero_pd();
 777         fiy1             = _mm_setzero_pd();
 778         fiz1             = _mm_setzero_pd();
 779         fix2             = _mm_setzero_pd();
 780         fiy2             = _mm_setzero_pd();
 781         fiz2             = _mm_setzero_pd();
 782
 783         /* Start inner kernel loop */
 784         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 785         {
 786
 787             /* Get j neighbor index, and coordinate index */
 788             jnrA             = jjnr[jidx];
 789             jnrB             = jjnr[jidx+1];
 790             j_coord_offsetA  = DIM*jnrA;
 791             j_coord_offsetB  = DIM*jnrB;
 792
 793             /* load j atom coordinates */
 794             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 795                                               &jx0,&jy0,&jz0);
 796
 797             /* Calculate displacement vector */
 798             dx00             = _mm_sub_pd(ix0,jx0);
 799             dy00             = _mm_sub_pd(iy0,jy0);
 800             dz00             = _mm_sub_pd(iz0,jz0);
 801             dx10             = _mm_sub_pd(ix1,jx0);
 802             dy10             = _mm_sub_pd(iy1,jy0);
 803             dz10             = _mm_sub_pd(iz1,jz0);
 804             dx20             = _mm_sub_pd(ix2,jx0);
 805             dy20             = _mm_sub_pd(iy2,jy0);
 806             dz20             = _mm_sub_pd(iz2,jz0);
 807
 808             /* Calculate squared distance and things based on it */
 809             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 810             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 811             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 812
 813             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 814             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 815             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 816
 817             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 818             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 819             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 820
 821             /* Load parameters for j particles */
 822             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 823             vdwjidx0A        = 2*vdwtype[jnrA+0];
 824             vdwjidx0B        = 2*vdwtype[jnrB+0];
 825
 826             fjx0             = _mm_setzero_pd();
 827             fjy0             = _mm_setzero_pd();
 828             fjz0             = _mm_setzero_pd();
 829
 830             /**************************
 831              * CALCULATE INTERACTIONS *
 832              **************************/
 833
 834             r00              = _mm_mul_pd(rsq00,rinv00);
 835
 836             /* Compute parameters for interactions between i and j atoms */
 837             qq00             = _mm_mul_pd(iq0,jq0);
 838             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 839                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 840
 841             /* Calculate table index by multiplying r with table scale and truncate to integer */
 842             rt               = _mm_mul_pd(r00,vftabscale);
 843             vfitab           = _mm_cvttpd_epi32(rt);
 844             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 845             vfitab           = _mm_slli_epi32(vfitab,3);
 846
 847             /* EWALD ELECTROSTATICS */
 848
 849             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 850             ewrt             = _mm_mul_pd(r00,ewtabscale);
 851             ewitab           = _mm_cvttpd_epi32(ewrt);
 852             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 853             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 854                                          &ewtabF,&ewtabFn);
 855             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 856             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 857
 858             /* CUBIC SPLINE TABLE DISPERSION */
 859             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 860             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 861             GMX_MM_TRANSPOSE2_PD(Y,F);
 862             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 863             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 864             GMX_MM_TRANSPOSE2_PD(G,H);
 865             Heps             = _mm_mul_pd(vfeps,H);
 866             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 867             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 868             fvdw6            = _mm_mul_pd(c6_00,FF);
 869
 870             /* CUBIC SPLINE TABLE REPULSION */
 871             vfitab           = _mm_add_epi32(vfitab,ifour);
 872             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 873             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 874             GMX_MM_TRANSPOSE2_PD(Y,F);
 875             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 876             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 877             GMX_MM_TRANSPOSE2_PD(G,H);
 878             Heps             = _mm_mul_pd(vfeps,H);
 879             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 880             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 881             fvdw12           = _mm_mul_pd(c12_00,FF);
 882             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 883
 884             fscal            = _mm_add_pd(felec,fvdw);
 885
 886             /* Calculate temporary vectorial force */
 887             tx               = _mm_mul_pd(fscal,dx00);
 888             ty               = _mm_mul_pd(fscal,dy00);
 889             tz               = _mm_mul_pd(fscal,dz00);
 890
 891             /* Update vectorial force */
 892             fix0             = _mm_add_pd(fix0,tx);
 893             fiy0             = _mm_add_pd(fiy0,ty);
 894             fiz0             = _mm_add_pd(fiz0,tz);
 895
 896             fjx0             = _mm_add_pd(fjx0,tx);
 897             fjy0             = _mm_add_pd(fjy0,ty);
 898             fjz0             = _mm_add_pd(fjz0,tz);
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             r10              = _mm_mul_pd(rsq10,rinv10);
 905
 906             /* Compute parameters for interactions between i and j atoms */
 907             qq10             = _mm_mul_pd(iq1,jq0);
 908
 909             /* EWALD ELECTROSTATICS */
 910
 911             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 912             ewrt             = _mm_mul_pd(r10,ewtabscale);
 913             ewitab           = _mm_cvttpd_epi32(ewrt);
 914             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 915             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 916                                          &ewtabF,&ewtabFn);
 917             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 918             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 919
 920             fscal            = felec;
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm_mul_pd(fscal,dx10);
 924             ty               = _mm_mul_pd(fscal,dy10);
 925             tz               = _mm_mul_pd(fscal,dz10);
 926
 927             /* Update vectorial force */
 928             fix1             = _mm_add_pd(fix1,tx);
 929             fiy1             = _mm_add_pd(fiy1,ty);
 930             fiz1             = _mm_add_pd(fiz1,tz);
 931
 932             fjx0             = _mm_add_pd(fjx0,tx);
 933             fjy0             = _mm_add_pd(fjy0,ty);
 934             fjz0             = _mm_add_pd(fjz0,tz);
 935
 936             /**************************
 937              * CALCULATE INTERACTIONS *
 938              **************************/
 939
 940             r20              = _mm_mul_pd(rsq20,rinv20);
 941
 942             /* Compute parameters for interactions between i and j atoms */
 943             qq20             = _mm_mul_pd(iq2,jq0);
 944
 945             /* EWALD ELECTROSTATICS */
 946
 947             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 948             ewrt             = _mm_mul_pd(r20,ewtabscale);
 949             ewitab           = _mm_cvttpd_epi32(ewrt);
 950             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 951             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 952                                          &ewtabF,&ewtabFn);
 953             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 954             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 955
 956             fscal            = felec;
 957
 958             /* Calculate temporary vectorial force */
 959             tx               = _mm_mul_pd(fscal,dx20);
 960             ty               = _mm_mul_pd(fscal,dy20);
 961             tz               = _mm_mul_pd(fscal,dz20);
 962
 963             /* Update vectorial force */
 964             fix2             = _mm_add_pd(fix2,tx);
 965             fiy2             = _mm_add_pd(fiy2,ty);
 966             fiz2             = _mm_add_pd(fiz2,tz);
 967
 968             fjx0             = _mm_add_pd(fjx0,tx);
 969             fjy0             = _mm_add_pd(fjy0,ty);
 970             fjz0             = _mm_add_pd(fjz0,tz);
 971
 972             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 973
 974             /* Inner loop uses 137 flops */
 975         }
 976
 977         if(jidx<j_index_end)
 978         {
 979
 980             jnrA             = jjnr[jidx];
 981             j_coord_offsetA  = DIM*jnrA;
 982
 983             /* load j atom coordinates */
 984             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 985                                               &jx0,&jy0,&jz0);
 986
 987             /* Calculate displacement vector */
 988             dx00             = _mm_sub_pd(ix0,jx0);
 989             dy00             = _mm_sub_pd(iy0,jy0);
 990             dz00             = _mm_sub_pd(iz0,jz0);
 991             dx10             = _mm_sub_pd(ix1,jx0);
 992             dy10             = _mm_sub_pd(iy1,jy0);
 993             dz10             = _mm_sub_pd(iz1,jz0);
 994             dx20             = _mm_sub_pd(ix2,jx0);
 995             dy20             = _mm_sub_pd(iy2,jy0);
 996             dz20             = _mm_sub_pd(iz2,jz0);
 997
 998             /* Calculate squared distance and things based on it */
 999             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1000             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1001             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1002
1003             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1004             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1005             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1006
1007             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1008             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1009             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1010
1011             /* Load parameters for j particles */
1012             jq0              = _mm_load_sd(charge+jnrA+0);
1013             vdwjidx0A        = 2*vdwtype[jnrA+0];
1014
1015             fjx0             = _mm_setzero_pd();
1016             fjy0             = _mm_setzero_pd();
1017             fjz0             = _mm_setzero_pd();
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             r00              = _mm_mul_pd(rsq00,rinv00);
1024
1025             /* Compute parameters for interactions between i and j atoms */
1026             qq00             = _mm_mul_pd(iq0,jq0);
1027             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1028
1029             /* Calculate table index by multiplying r with table scale and truncate to integer */
1030             rt               = _mm_mul_pd(r00,vftabscale);
1031             vfitab           = _mm_cvttpd_epi32(rt);
1032             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1033             vfitab           = _mm_slli_epi32(vfitab,3);
1034
1035             /* EWALD ELECTROSTATICS */
1036
1037             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1038             ewrt             = _mm_mul_pd(r00,ewtabscale);
1039             ewitab           = _mm_cvttpd_epi32(ewrt);
1040             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1041             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1042             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1043             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1044
1045             /* CUBIC SPLINE TABLE DISPERSION */
1046             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1047             F                = _mm_setzero_pd();
1048             GMX_MM_TRANSPOSE2_PD(Y,F);
1049             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1050             H                = _mm_setzero_pd();
1051             GMX_MM_TRANSPOSE2_PD(G,H);
1052             Heps             = _mm_mul_pd(vfeps,H);
1053             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1054             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1055             fvdw6            = _mm_mul_pd(c6_00,FF);
1056
1057             /* CUBIC SPLINE TABLE REPULSION */
1058             vfitab           = _mm_add_epi32(vfitab,ifour);
1059             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1060             F                = _mm_setzero_pd();
1061             GMX_MM_TRANSPOSE2_PD(Y,F);
1062             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1063             H                = _mm_setzero_pd();
1064             GMX_MM_TRANSPOSE2_PD(G,H);
1065             Heps             = _mm_mul_pd(vfeps,H);
1066             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1067             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1068             fvdw12           = _mm_mul_pd(c12_00,FF);
1069             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1070
1071             fscal            = _mm_add_pd(felec,fvdw);
1072
1073             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1074
1075             /* Calculate temporary vectorial force */
1076             tx               = _mm_mul_pd(fscal,dx00);
1077             ty               = _mm_mul_pd(fscal,dy00);
1078             tz               = _mm_mul_pd(fscal,dz00);
1079
1080             /* Update vectorial force */
1081             fix0             = _mm_add_pd(fix0,tx);
1082             fiy0             = _mm_add_pd(fiy0,ty);
1083             fiz0             = _mm_add_pd(fiz0,tz);
1084
1085             fjx0             = _mm_add_pd(fjx0,tx);
1086             fjy0             = _mm_add_pd(fjy0,ty);
1087             fjz0             = _mm_add_pd(fjz0,tz);
1088
1089             /**************************
1090              * CALCULATE INTERACTIONS *
1091              **************************/
1092
1093             r10              = _mm_mul_pd(rsq10,rinv10);
1094
1095             /* Compute parameters for interactions between i and j atoms */
1096             qq10             = _mm_mul_pd(iq1,jq0);
1097
1098             /* EWALD ELECTROSTATICS */
1099
1100             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1101             ewrt             = _mm_mul_pd(r10,ewtabscale);
1102             ewitab           = _mm_cvttpd_epi32(ewrt);
1103             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1104             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1105             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1106             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1107
1108             fscal            = felec;
1109
1110             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm_mul_pd(fscal,dx10);
1114             ty               = _mm_mul_pd(fscal,dy10);
1115             tz               = _mm_mul_pd(fscal,dz10);
1116
1117             /* Update vectorial force */
1118             fix1             = _mm_add_pd(fix1,tx);
1119             fiy1             = _mm_add_pd(fiy1,ty);
1120             fiz1             = _mm_add_pd(fiz1,tz);
1121
1122             fjx0             = _mm_add_pd(fjx0,tx);
1123             fjy0             = _mm_add_pd(fjy0,ty);
1124             fjz0             = _mm_add_pd(fjz0,tz);
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             r20              = _mm_mul_pd(rsq20,rinv20);
1131
1132             /* Compute parameters for interactions between i and j atoms */
1133             qq20             = _mm_mul_pd(iq2,jq0);
1134
1135             /* EWALD ELECTROSTATICS */
1136
1137             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1138             ewrt             = _mm_mul_pd(r20,ewtabscale);
1139             ewitab           = _mm_cvttpd_epi32(ewrt);
1140             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1141             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1142             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1143             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1144
1145             fscal            = felec;
1146
1147             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1148
1149             /* Calculate temporary vectorial force */
1150             tx               = _mm_mul_pd(fscal,dx20);
1151             ty               = _mm_mul_pd(fscal,dy20);
1152             tz               = _mm_mul_pd(fscal,dz20);
1153
1154             /* Update vectorial force */
1155             fix2             = _mm_add_pd(fix2,tx);
1156             fiy2             = _mm_add_pd(fiy2,ty);
1157             fiz2             = _mm_add_pd(fiz2,tz);
1158
1159             fjx0             = _mm_add_pd(fjx0,tx);
1160             fjy0             = _mm_add_pd(fjy0,ty);
1161             fjz0             = _mm_add_pd(fjz0,tz);
1162
1163             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1164
1165             /* Inner loop uses 137 flops */
1166         }
1167
1168         /* End of innermost loop */
1169
1170         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1171                                               f+i_coord_offset,fshift+i_shift_offset);
1172
1173         /* Increment number of inner iterations */
1174         inneriter                  += j_index_end - j_index_start;
1175
1176         /* Outer loop uses 18 flops */
1177     }
1178
1179     /* Increment number of outer iterations */
1180     outeriter        += nri;
1181
1182     /* Update outer/inner flops */
1183
1184     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1185 }