src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_128_fma_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,twovfeps;
  88     real             *vftab;
  89     __m128d          dummy_mask,cutoff_mask;
  90     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  91     __m128d          one     = _mm_set1_pd(1.0);
  92     __m128d          two     = _mm_set1_pd(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_pd(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     nvdwtype         = fr->ntype;
 107     vdwparam         = fr->nbfp;
 108     vdwtype          = mdatoms->typeA;
 109
 110     vftab            = kernel_data->table_elec_vdw->data;
 111     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 112
 113     /* Setup water-specific parameters */
 114     inr              = nlist->iinr[0];
 115     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 116     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 117     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 118     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 144                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 145
 146         fix0             = _mm_setzero_pd();
 147         fiy0             = _mm_setzero_pd();
 148         fiz0             = _mm_setzero_pd();
 149         fix1             = _mm_setzero_pd();
 150         fiy1             = _mm_setzero_pd();
 151         fiz1             = _mm_setzero_pd();
 152         fix2             = _mm_setzero_pd();
 153         fiy2             = _mm_setzero_pd();
 154         fiz2             = _mm_setzero_pd();
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_pd();
 158         vvdwsum          = _mm_setzero_pd();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169
 170             /* load j atom coordinates */
 171             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_pd(ix0,jx0);
 176             dy00             = _mm_sub_pd(iy0,jy0);
 177             dz00             = _mm_sub_pd(iz0,jz0);
 178             dx10             = _mm_sub_pd(ix1,jx0);
 179             dy10             = _mm_sub_pd(iy1,jy0);
 180             dz10             = _mm_sub_pd(iz1,jz0);
 181             dx20             = _mm_sub_pd(ix2,jx0);
 182             dy20             = _mm_sub_pd(iy2,jy0);
 183             dz20             = _mm_sub_pd(iz2,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 187             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 188             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 189
 190             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 191             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 192             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 193
 194             /* Load parameters for j particles */
 195             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 196             vdwjidx0A        = 2*vdwtype[jnrA+0];
 197             vdwjidx0B        = 2*vdwtype[jnrB+0];
 198
 199             fjx0             = _mm_setzero_pd();
 200             fjy0             = _mm_setzero_pd();
 201             fjz0             = _mm_setzero_pd();
 202
 203             /**************************
 204              * CALCULATE INTERACTIONS *
 205              **************************/
 206
 207             r00              = _mm_mul_pd(rsq00,rinv00);
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq00             = _mm_mul_pd(iq0,jq0);
 211             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 212                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 213
 214             /* Calculate table index by multiplying r with table scale and truncate to integer */
 215             rt               = _mm_mul_pd(r00,vftabscale);
 216             vfitab           = _mm_cvttpd_epi32(rt);
 217 #ifdef __XOP__
 218             vfeps            = _mm_frcz_pd(rt);
 219 #else
 220             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 221 #endif
 222             twovfeps         = _mm_add_pd(vfeps,vfeps);
 223             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 224
 225             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 226             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 227             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 228             GMX_MM_TRANSPOSE2_PD(Y,F);
 229             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 230             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 231             GMX_MM_TRANSPOSE2_PD(G,H);
 232             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 233             VV               = _mm_macc_pd(vfeps,Fp,Y);
 234             velec            = _mm_mul_pd(qq00,VV);
 235             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 236             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 237
 238             /* CUBIC SPLINE TABLE DISPERSION */
 239             vfitab           = _mm_add_epi32(vfitab,ifour);
 240             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 241             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 242             GMX_MM_TRANSPOSE2_PD(Y,F);
 243             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 244             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 245             GMX_MM_TRANSPOSE2_PD(G,H);
 246             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 247             VV               = _mm_macc_pd(vfeps,Fp,Y);
 248             vvdw6            = _mm_mul_pd(c6_00,VV);
 249             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 250             fvdw6            = _mm_mul_pd(c6_00,FF);
 251
 252             /* CUBIC SPLINE TABLE REPULSION */
 253             vfitab           = _mm_add_epi32(vfitab,ifour);
 254             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 255             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 256             GMX_MM_TRANSPOSE2_PD(Y,F);
 257             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 258             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 259             GMX_MM_TRANSPOSE2_PD(G,H);
 260             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 261             VV               = _mm_macc_pd(vfeps,Fp,Y);
 262             vvdw12           = _mm_mul_pd(c12_00,VV);
 263             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 264             fvdw12           = _mm_mul_pd(c12_00,FF);
 265             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 266             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             velecsum         = _mm_add_pd(velecsum,velec);
 270             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 271
 272             fscal            = _mm_add_pd(felec,fvdw);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 276             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 277             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 278
 279             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 280             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 281             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             r10              = _mm_mul_pd(rsq10,rinv10);
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq10             = _mm_mul_pd(iq1,jq0);
 291
 292             /* Calculate table index by multiplying r with table scale and truncate to integer */
 293             rt               = _mm_mul_pd(r10,vftabscale);
 294             vfitab           = _mm_cvttpd_epi32(rt);
 295 #ifdef __XOP__
 296             vfeps            = _mm_frcz_pd(rt);
 297 #else
 298             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 299 #endif
 300             twovfeps         = _mm_add_pd(vfeps,vfeps);
 301             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 302
 303             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 304             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 305             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 306             GMX_MM_TRANSPOSE2_PD(Y,F);
 307             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 308             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 309             GMX_MM_TRANSPOSE2_PD(G,H);
 310             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 311             VV               = _mm_macc_pd(vfeps,Fp,Y);
 312             velec            = _mm_mul_pd(qq10,VV);
 313             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 314             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velecsum         = _mm_add_pd(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             /* Update vectorial force */
 322             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 323             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 324             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 325
 326             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 327             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 328             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             r20              = _mm_mul_pd(rsq20,rinv20);
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq20             = _mm_mul_pd(iq2,jq0);
 338
 339             /* Calculate table index by multiplying r with table scale and truncate to integer */
 340             rt               = _mm_mul_pd(r20,vftabscale);
 341             vfitab           = _mm_cvttpd_epi32(rt);
 342 #ifdef __XOP__
 343             vfeps            = _mm_frcz_pd(rt);
 344 #else
 345             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 346 #endif
 347             twovfeps         = _mm_add_pd(vfeps,vfeps);
 348             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 349
 350             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 351             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 352             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 353             GMX_MM_TRANSPOSE2_PD(Y,F);
 354             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 355             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 356             GMX_MM_TRANSPOSE2_PD(G,H);
 357             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 358             VV               = _mm_macc_pd(vfeps,Fp,Y);
 359             velec            = _mm_mul_pd(qq20,VV);
 360             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 361             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velecsum         = _mm_add_pd(velecsum,velec);
 365
 366             fscal            = felec;
 367
 368             /* Update vectorial force */
 369             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 370             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 371             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 372
 373             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 374             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 375             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 376
 377             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 378
 379             /* Inner loop uses 171 flops */
 380         }
 381
 382         if(jidx<j_index_end)
 383         {
 384
 385             jnrA             = jjnr[jidx];
 386             j_coord_offsetA  = DIM*jnrA;
 387
 388             /* load j atom coordinates */
 389             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 390                                               &jx0,&jy0,&jz0);
 391
 392             /* Calculate displacement vector */
 393             dx00             = _mm_sub_pd(ix0,jx0);
 394             dy00             = _mm_sub_pd(iy0,jy0);
 395             dz00             = _mm_sub_pd(iz0,jz0);
 396             dx10             = _mm_sub_pd(ix1,jx0);
 397             dy10             = _mm_sub_pd(iy1,jy0);
 398             dz10             = _mm_sub_pd(iz1,jz0);
 399             dx20             = _mm_sub_pd(ix2,jx0);
 400             dy20             = _mm_sub_pd(iy2,jy0);
 401             dz20             = _mm_sub_pd(iz2,jz0);
 402
 403             /* Calculate squared distance and things based on it */
 404             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 405             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 406             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 407
 408             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 409             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 410             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 411
 412             /* Load parameters for j particles */
 413             jq0              = _mm_load_sd(charge+jnrA+0);
 414             vdwjidx0A        = 2*vdwtype[jnrA+0];
 415
 416             fjx0             = _mm_setzero_pd();
 417             fjy0             = _mm_setzero_pd();
 418             fjz0             = _mm_setzero_pd();
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             r00              = _mm_mul_pd(rsq00,rinv00);
 425
 426             /* Compute parameters for interactions between i and j atoms */
 427             qq00             = _mm_mul_pd(iq0,jq0);
 428             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 429
 430             /* Calculate table index by multiplying r with table scale and truncate to integer */
 431             rt               = _mm_mul_pd(r00,vftabscale);
 432             vfitab           = _mm_cvttpd_epi32(rt);
 433 #ifdef __XOP__
 434             vfeps            = _mm_frcz_pd(rt);
 435 #else
 436             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 437 #endif
 438             twovfeps         = _mm_add_pd(vfeps,vfeps);
 439             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 440
 441             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 442             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 443             F                = _mm_setzero_pd();
 444             GMX_MM_TRANSPOSE2_PD(Y,F);
 445             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 446             H                = _mm_setzero_pd();
 447             GMX_MM_TRANSPOSE2_PD(G,H);
 448             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 449             VV               = _mm_macc_pd(vfeps,Fp,Y);
 450             velec            = _mm_mul_pd(qq00,VV);
 451             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 452             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 453
 454             /* CUBIC SPLINE TABLE DISPERSION */
 455             vfitab           = _mm_add_epi32(vfitab,ifour);
 456             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 457             F                = _mm_setzero_pd();
 458             GMX_MM_TRANSPOSE2_PD(Y,F);
 459             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 460             H                = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(G,H);
 462             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 463             VV               = _mm_macc_pd(vfeps,Fp,Y);
 464             vvdw6            = _mm_mul_pd(c6_00,VV);
 465             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 466             fvdw6            = _mm_mul_pd(c6_00,FF);
 467
 468             /* CUBIC SPLINE TABLE REPULSION */
 469             vfitab           = _mm_add_epi32(vfitab,ifour);
 470             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 471             F                = _mm_setzero_pd();
 472             GMX_MM_TRANSPOSE2_PD(Y,F);
 473             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 474             H                = _mm_setzero_pd();
 475             GMX_MM_TRANSPOSE2_PD(G,H);
 476             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 477             VV               = _mm_macc_pd(vfeps,Fp,Y);
 478             vvdw12           = _mm_mul_pd(c12_00,VV);
 479             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 480             fvdw12           = _mm_mul_pd(c12_00,FF);
 481             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 482             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 483
 484             /* Update potential sum for this i atom from the interaction with this j atom. */
 485             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 486             velecsum         = _mm_add_pd(velecsum,velec);
 487             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 488             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 489
 490             fscal            = _mm_add_pd(felec,fvdw);
 491
 492             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 493
 494             /* Update vectorial force */
 495             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 496             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 497             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 498
 499             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 500             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 501             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             r10              = _mm_mul_pd(rsq10,rinv10);
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq10             = _mm_mul_pd(iq1,jq0);
 511
 512             /* Calculate table index by multiplying r with table scale and truncate to integer */
 513             rt               = _mm_mul_pd(r10,vftabscale);
 514             vfitab           = _mm_cvttpd_epi32(rt);
 515 #ifdef __XOP__
 516             vfeps            = _mm_frcz_pd(rt);
 517 #else
 518             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 519 #endif
 520             twovfeps         = _mm_add_pd(vfeps,vfeps);
 521             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 522
 523             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 524             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 525             F                = _mm_setzero_pd();
 526             GMX_MM_TRANSPOSE2_PD(Y,F);
 527             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 528             H                = _mm_setzero_pd();
 529             GMX_MM_TRANSPOSE2_PD(G,H);
 530             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 531             VV               = _mm_macc_pd(vfeps,Fp,Y);
 532             velec            = _mm_mul_pd(qq10,VV);
 533             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 534             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 535
 536             /* Update potential sum for this i atom from the interaction with this j atom. */
 537             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 538             velecsum         = _mm_add_pd(velecsum,velec);
 539
 540             fscal            = felec;
 541
 542             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 543
 544             /* Update vectorial force */
 545             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 546             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 547             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 548
 549             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 550             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 551             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             r20              = _mm_mul_pd(rsq20,rinv20);
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq20             = _mm_mul_pd(iq2,jq0);
 561
 562             /* Calculate table index by multiplying r with table scale and truncate to integer */
 563             rt               = _mm_mul_pd(r20,vftabscale);
 564             vfitab           = _mm_cvttpd_epi32(rt);
 565 #ifdef __XOP__
 566             vfeps            = _mm_frcz_pd(rt);
 567 #else
 568             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 569 #endif
 570             twovfeps         = _mm_add_pd(vfeps,vfeps);
 571             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 572
 573             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 574             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 575             F                = _mm_setzero_pd();
 576             GMX_MM_TRANSPOSE2_PD(Y,F);
 577             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 578             H                = _mm_setzero_pd();
 579             GMX_MM_TRANSPOSE2_PD(G,H);
 580             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 581             VV               = _mm_macc_pd(vfeps,Fp,Y);
 582             velec            = _mm_mul_pd(qq20,VV);
 583             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 584             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 588             velecsum         = _mm_add_pd(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 593
 594             /* Update vectorial force */
 595             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 596             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 597             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 598
 599             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 600             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 601             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 602
 603             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 604
 605             /* Inner loop uses 171 flops */
 606         }
 607
 608         /* End of innermost loop */
 609
 610         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 611                                               f+i_coord_offset,fshift+i_shift_offset);
 612
 613         ggid                        = gid[iidx];
 614         /* Update potential energies */
 615         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 616         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 617
 618         /* Increment number of inner iterations */
 619         inneriter                  += j_index_end - j_index_start;
 620
 621         /* Outer loop uses 20 flops */
 622     }
 623
 624     /* Increment number of outer iterations */
 625     outeriter        += nri;
 626
 627     /* Update outer/inner flops */
 628
 629     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*171);
 630 }
 631 /*
 632  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 633  * Electrostatics interaction: CubicSplineTable
 634  * VdW interaction:            CubicSplineTable
 635  * Geometry:                   Water3-Particle
 636  * Calculate force/pot:        Force
 637  */
 638 void
 639 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 640                     (t_nblist * gmx_restrict                nlist,
 641                      rvec * gmx_restrict                    xx,
 642                      rvec * gmx_restrict                    ff,
 643                      t_forcerec * gmx_restrict              fr,
 644                      t_mdatoms * gmx_restrict               mdatoms,
 645                      nb_kernel_data_t * gmx_restrict        kernel_data,
 646                      t_nrnb * gmx_restrict                  nrnb)
 647 {
 648     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 649      * just 0 for non-waters.
 650      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 651      * jnr indices corresponding to data put in the four positions in the SIMD register.
 652      */
 653     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 654     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 655     int              jnrA,jnrB;
 656     int              j_coord_offsetA,j_coord_offsetB;
 657     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 658     real             rcutoff_scalar;
 659     real             *shiftvec,*fshift,*x,*f;
 660     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 661     int              vdwioffset0;
 662     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 663     int              vdwioffset1;
 664     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 665     int              vdwioffset2;
 666     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 667     int              vdwjidx0A,vdwjidx0B;
 668     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 669     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 670     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 671     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 672     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 673     real             *charge;
 674     int              nvdwtype;
 675     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 676     int              *vdwtype;
 677     real             *vdwparam;
 678     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 679     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 680     __m128i          vfitab;
 681     __m128i          ifour       = _mm_set1_epi32(4);
 682     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 683     real             *vftab;
 684     __m128d          dummy_mask,cutoff_mask;
 685     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 686     __m128d          one     = _mm_set1_pd(1.0);
 687     __m128d          two     = _mm_set1_pd(2.0);
 688     x                = xx[0];
 689     f                = ff[0];
 690
 691     nri              = nlist->nri;
 692     iinr             = nlist->iinr;
 693     jindex           = nlist->jindex;
 694     jjnr             = nlist->jjnr;
 695     shiftidx         = nlist->shift;
 696     gid              = nlist->gid;
 697     shiftvec         = fr->shift_vec[0];
 698     fshift           = fr->fshift[0];
 699     facel            = _mm_set1_pd(fr->epsfac);
 700     charge           = mdatoms->chargeA;
 701     nvdwtype         = fr->ntype;
 702     vdwparam         = fr->nbfp;
 703     vdwtype          = mdatoms->typeA;
 704
 705     vftab            = kernel_data->table_elec_vdw->data;
 706     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 707
 708     /* Setup water-specific parameters */
 709     inr              = nlist->iinr[0];
 710     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 711     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 712     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 713     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 714
 715     /* Avoid stupid compiler warnings */
 716     jnrA = jnrB = 0;
 717     j_coord_offsetA = 0;
 718     j_coord_offsetB = 0;
 719
 720     outeriter        = 0;
 721     inneriter        = 0;
 722
 723     /* Start outer loop over neighborlists */
 724     for(iidx=0; iidx<nri; iidx++)
 725     {
 726         /* Load shift vector for this list */
 727         i_shift_offset   = DIM*shiftidx[iidx];
 728
 729         /* Load limits for loop over neighbors */
 730         j_index_start    = jindex[iidx];
 731         j_index_end      = jindex[iidx+1];
 732
 733         /* Get outer coordinate index */
 734         inr              = iinr[iidx];
 735         i_coord_offset   = DIM*inr;
 736
 737         /* Load i particle coords and add shift vector */
 738         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 739                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 740
 741         fix0             = _mm_setzero_pd();
 742         fiy0             = _mm_setzero_pd();
 743         fiz0             = _mm_setzero_pd();
 744         fix1             = _mm_setzero_pd();
 745         fiy1             = _mm_setzero_pd();
 746         fiz1             = _mm_setzero_pd();
 747         fix2             = _mm_setzero_pd();
 748         fiy2             = _mm_setzero_pd();
 749         fiz2             = _mm_setzero_pd();
 750
 751         /* Start inner kernel loop */
 752         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 753         {
 754
 755             /* Get j neighbor index, and coordinate index */
 756             jnrA             = jjnr[jidx];
 757             jnrB             = jjnr[jidx+1];
 758             j_coord_offsetA  = DIM*jnrA;
 759             j_coord_offsetB  = DIM*jnrB;
 760
 761             /* load j atom coordinates */
 762             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 763                                               &jx0,&jy0,&jz0);
 764
 765             /* Calculate displacement vector */
 766             dx00             = _mm_sub_pd(ix0,jx0);
 767             dy00             = _mm_sub_pd(iy0,jy0);
 768             dz00             = _mm_sub_pd(iz0,jz0);
 769             dx10             = _mm_sub_pd(ix1,jx0);
 770             dy10             = _mm_sub_pd(iy1,jy0);
 771             dz10             = _mm_sub_pd(iz1,jz0);
 772             dx20             = _mm_sub_pd(ix2,jx0);
 773             dy20             = _mm_sub_pd(iy2,jy0);
 774             dz20             = _mm_sub_pd(iz2,jz0);
 775
 776             /* Calculate squared distance and things based on it */
 777             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 778             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 779             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 780
 781             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 782             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 783             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 784
 785             /* Load parameters for j particles */
 786             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 787             vdwjidx0A        = 2*vdwtype[jnrA+0];
 788             vdwjidx0B        = 2*vdwtype[jnrB+0];
 789
 790             fjx0             = _mm_setzero_pd();
 791             fjy0             = _mm_setzero_pd();
 792             fjz0             = _mm_setzero_pd();
 793
 794             /**************************
 795              * CALCULATE INTERACTIONS *
 796              **************************/
 797
 798             r00              = _mm_mul_pd(rsq00,rinv00);
 799
 800             /* Compute parameters for interactions between i and j atoms */
 801             qq00             = _mm_mul_pd(iq0,jq0);
 802             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 803                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 804
 805             /* Calculate table index by multiplying r with table scale and truncate to integer */
 806             rt               = _mm_mul_pd(r00,vftabscale);
 807             vfitab           = _mm_cvttpd_epi32(rt);
 808 #ifdef __XOP__
 809             vfeps            = _mm_frcz_pd(rt);
 810 #else
 811             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 812 #endif
 813             twovfeps         = _mm_add_pd(vfeps,vfeps);
 814             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 815
 816             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 817             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 818             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 819             GMX_MM_TRANSPOSE2_PD(Y,F);
 820             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 821             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 822             GMX_MM_TRANSPOSE2_PD(G,H);
 823             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 824             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 825             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 826
 827             /* CUBIC SPLINE TABLE DISPERSION */
 828             vfitab           = _mm_add_epi32(vfitab,ifour);
 829             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 830             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 831             GMX_MM_TRANSPOSE2_PD(Y,F);
 832             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 833             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 834             GMX_MM_TRANSPOSE2_PD(G,H);
 835             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 836             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 837             fvdw6            = _mm_mul_pd(c6_00,FF);
 838
 839             /* CUBIC SPLINE TABLE REPULSION */
 840             vfitab           = _mm_add_epi32(vfitab,ifour);
 841             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 842             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 843             GMX_MM_TRANSPOSE2_PD(Y,F);
 844             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 845             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 846             GMX_MM_TRANSPOSE2_PD(G,H);
 847             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 848             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 849             fvdw12           = _mm_mul_pd(c12_00,FF);
 850             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 851
 852             fscal            = _mm_add_pd(felec,fvdw);
 853
 854             /* Update vectorial force */
 855             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 856             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 857             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 858
 859             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 860             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 861             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             r10              = _mm_mul_pd(rsq10,rinv10);
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq10             = _mm_mul_pd(iq1,jq0);
 871
 872             /* Calculate table index by multiplying r with table scale and truncate to integer */
 873             rt               = _mm_mul_pd(r10,vftabscale);
 874             vfitab           = _mm_cvttpd_epi32(rt);
 875 #ifdef __XOP__
 876             vfeps            = _mm_frcz_pd(rt);
 877 #else
 878             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 879 #endif
 880             twovfeps         = _mm_add_pd(vfeps,vfeps);
 881             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 882
 883             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 884             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 885             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 886             GMX_MM_TRANSPOSE2_PD(Y,F);
 887             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 888             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 889             GMX_MM_TRANSPOSE2_PD(G,H);
 890             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 891             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 892             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 893
 894             fscal            = felec;
 895
 896             /* Update vectorial force */
 897             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 898             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 899             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 900
 901             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 902             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 903             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 904
 905             /**************************
 906              * CALCULATE INTERACTIONS *
 907              **************************/
 908
 909             r20              = _mm_mul_pd(rsq20,rinv20);
 910
 911             /* Compute parameters for interactions between i and j atoms */
 912             qq20             = _mm_mul_pd(iq2,jq0);
 913
 914             /* Calculate table index by multiplying r with table scale and truncate to integer */
 915             rt               = _mm_mul_pd(r20,vftabscale);
 916             vfitab           = _mm_cvttpd_epi32(rt);
 917 #ifdef __XOP__
 918             vfeps            = _mm_frcz_pd(rt);
 919 #else
 920             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 921 #endif
 922             twovfeps         = _mm_add_pd(vfeps,vfeps);
 923             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 924
 925             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 926             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 927             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 928             GMX_MM_TRANSPOSE2_PD(Y,F);
 929             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 930             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 931             GMX_MM_TRANSPOSE2_PD(G,H);
 932             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 933             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 934             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 935
 936             fscal            = felec;
 937
 938             /* Update vectorial force */
 939             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 940             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 941             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 942
 943             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 944             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 945             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 946
 947             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 948
 949             /* Inner loop uses 151 flops */
 950         }
 951
 952         if(jidx<j_index_end)
 953         {
 954
 955             jnrA             = jjnr[jidx];
 956             j_coord_offsetA  = DIM*jnrA;
 957
 958             /* load j atom coordinates */
 959             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 960                                               &jx0,&jy0,&jz0);
 961
 962             /* Calculate displacement vector */
 963             dx00             = _mm_sub_pd(ix0,jx0);
 964             dy00             = _mm_sub_pd(iy0,jy0);
 965             dz00             = _mm_sub_pd(iz0,jz0);
 966             dx10             = _mm_sub_pd(ix1,jx0);
 967             dy10             = _mm_sub_pd(iy1,jy0);
 968             dz10             = _mm_sub_pd(iz1,jz0);
 969             dx20             = _mm_sub_pd(ix2,jx0);
 970             dy20             = _mm_sub_pd(iy2,jy0);
 971             dz20             = _mm_sub_pd(iz2,jz0);
 972
 973             /* Calculate squared distance and things based on it */
 974             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 975             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 976             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 977
 978             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 979             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 980             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 981
 982             /* Load parameters for j particles */
 983             jq0              = _mm_load_sd(charge+jnrA+0);
 984             vdwjidx0A        = 2*vdwtype[jnrA+0];
 985
 986             fjx0             = _mm_setzero_pd();
 987             fjy0             = _mm_setzero_pd();
 988             fjz0             = _mm_setzero_pd();
 989
 990             /**************************
 991              * CALCULATE INTERACTIONS *
 992              **************************/
 993
 994             r00              = _mm_mul_pd(rsq00,rinv00);
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             qq00             = _mm_mul_pd(iq0,jq0);
 998             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 999
1000             /* Calculate table index by multiplying r with table scale and truncate to integer */
1001             rt               = _mm_mul_pd(r00,vftabscale);
1002             vfitab           = _mm_cvttpd_epi32(rt);
1003 #ifdef __XOP__
1004             vfeps            = _mm_frcz_pd(rt);
1005 #else
1006             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1007 #endif
1008             twovfeps         = _mm_add_pd(vfeps,vfeps);
1009             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1010
1011             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1012             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1013             F                = _mm_setzero_pd();
1014             GMX_MM_TRANSPOSE2_PD(Y,F);
1015             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1016             H                = _mm_setzero_pd();
1017             GMX_MM_TRANSPOSE2_PD(G,H);
1018             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1019             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1020             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1021
1022             /* CUBIC SPLINE TABLE DISPERSION */
1023             vfitab           = _mm_add_epi32(vfitab,ifour);
1024             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1025             F                = _mm_setzero_pd();
1026             GMX_MM_TRANSPOSE2_PD(Y,F);
1027             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1028             H                = _mm_setzero_pd();
1029             GMX_MM_TRANSPOSE2_PD(G,H);
1030             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1031             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1032             fvdw6            = _mm_mul_pd(c6_00,FF);
1033
1034             /* CUBIC SPLINE TABLE REPULSION */
1035             vfitab           = _mm_add_epi32(vfitab,ifour);
1036             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1037             F                = _mm_setzero_pd();
1038             GMX_MM_TRANSPOSE2_PD(Y,F);
1039             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1040             H                = _mm_setzero_pd();
1041             GMX_MM_TRANSPOSE2_PD(G,H);
1042             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1043             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1044             fvdw12           = _mm_mul_pd(c12_00,FF);
1045             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1046
1047             fscal            = _mm_add_pd(felec,fvdw);
1048
1049             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1050
1051             /* Update vectorial force */
1052             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1053             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1054             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1055
1056             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1057             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1058             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1059
1060             /**************************
1061              * CALCULATE INTERACTIONS *
1062              **************************/
1063
1064             r10              = _mm_mul_pd(rsq10,rinv10);
1065
1066             /* Compute parameters for interactions between i and j atoms */
1067             qq10             = _mm_mul_pd(iq1,jq0);
1068
1069             /* Calculate table index by multiplying r with table scale and truncate to integer */
1070             rt               = _mm_mul_pd(r10,vftabscale);
1071             vfitab           = _mm_cvttpd_epi32(rt);
1072 #ifdef __XOP__
1073             vfeps            = _mm_frcz_pd(rt);
1074 #else
1075             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1076 #endif
1077             twovfeps         = _mm_add_pd(vfeps,vfeps);
1078             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1079
1080             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1081             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1082             F                = _mm_setzero_pd();
1083             GMX_MM_TRANSPOSE2_PD(Y,F);
1084             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1085             H                = _mm_setzero_pd();
1086             GMX_MM_TRANSPOSE2_PD(G,H);
1087             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1088             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1089             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1090
1091             fscal            = felec;
1092
1093             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1094
1095             /* Update vectorial force */
1096             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1097             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1098             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1099
1100             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1101             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1102             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1103
1104             /**************************
1105              * CALCULATE INTERACTIONS *
1106              **************************/
1107
1108             r20              = _mm_mul_pd(rsq20,rinv20);
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             qq20             = _mm_mul_pd(iq2,jq0);
1112
1113             /* Calculate table index by multiplying r with table scale and truncate to integer */
1114             rt               = _mm_mul_pd(r20,vftabscale);
1115             vfitab           = _mm_cvttpd_epi32(rt);
1116 #ifdef __XOP__
1117             vfeps            = _mm_frcz_pd(rt);
1118 #else
1119             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1120 #endif
1121             twovfeps         = _mm_add_pd(vfeps,vfeps);
1122             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1123
1124             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1125             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1126             F                = _mm_setzero_pd();
1127             GMX_MM_TRANSPOSE2_PD(Y,F);
1128             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1129             H                = _mm_setzero_pd();
1130             GMX_MM_TRANSPOSE2_PD(G,H);
1131             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
1132             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
1133             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1138
1139             /* Update vectorial force */
1140             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1141             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1142             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1143
1144             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1145             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1146             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1147
1148             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1149
1150             /* Inner loop uses 151 flops */
1151         }
1152
1153         /* End of innermost loop */
1154
1155         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1156                                               f+i_coord_offset,fshift+i_shift_offset);
1157
1158         /* Increment number of inner iterations */
1159         inneriter                  += j_index_end - j_index_start;
1160
1161         /* Outer loop uses 18 flops */
1162     }
1163
1164     /* Increment number of outer iterations */
1165     outeriter        += nri;
1166
1167     /* Update outer/inner flops */
1168
1169     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*151);
1170 }