src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_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_ElecRFCut_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_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              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  91     real             *vftab;
  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     krf              = _mm_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_pd(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 122     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 123     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 155                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 156
 157         fix0             = _mm_setzero_pd();
 158         fiy0             = _mm_setzero_pd();
 159         fiz0             = _mm_setzero_pd();
 160         fix1             = _mm_setzero_pd();
 161         fiy1             = _mm_setzero_pd();
 162         fiz1             = _mm_setzero_pd();
 163         fix2             = _mm_setzero_pd();
 164         fiy2             = _mm_setzero_pd();
 165         fiz2             = _mm_setzero_pd();
 166         fix3             = _mm_setzero_pd();
 167         fiy3             = _mm_setzero_pd();
 168         fiz3             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172         vvdwsum          = _mm_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_pd(ix0,jx0);
 190             dy00             = _mm_sub_pd(iy0,jy0);
 191             dz00             = _mm_sub_pd(iz0,jz0);
 192             dx10             = _mm_sub_pd(ix1,jx0);
 193             dy10             = _mm_sub_pd(iy1,jy0);
 194             dz10             = _mm_sub_pd(iz1,jz0);
 195             dx20             = _mm_sub_pd(ix2,jx0);
 196             dy20             = _mm_sub_pd(iy2,jy0);
 197             dz20             = _mm_sub_pd(iz2,jz0);
 198             dx30             = _mm_sub_pd(ix3,jx0);
 199             dy30             = _mm_sub_pd(iy3,jy0);
 200             dz30             = _mm_sub_pd(iz3,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 206             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 207
 208             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 209             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 210             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 211             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 212
 213             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 215             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 219             vdwjidx0A        = 2*vdwtype[jnrA+0];
 220             vdwjidx0B        = 2*vdwtype[jnrB+0];
 221
 222             fjx0             = _mm_setzero_pd();
 223             fjy0             = _mm_setzero_pd();
 224             fjz0             = _mm_setzero_pd();
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             r00              = _mm_mul_pd(rsq00,rinv00);
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 235
 236             /* Calculate table index by multiplying r with table scale and truncate to integer */
 237             rt               = _mm_mul_pd(r00,vftabscale);
 238             vfitab           = _mm_cvttpd_epi32(rt);
 239 #ifdef __XOP__
 240             vfeps            = _mm_frcz_pd(rt);
 241 #else
 242             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 243 #endif
 244             twovfeps         = _mm_add_pd(vfeps,vfeps);
 245             vfitab           = _mm_slli_epi32(vfitab,3);
 246
 247             /* CUBIC SPLINE TABLE DISPERSION */
 248             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 249             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 250             GMX_MM_TRANSPOSE2_PD(Y,F);
 251             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 252             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 253             GMX_MM_TRANSPOSE2_PD(G,H);
 254             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 255             VV               = _mm_macc_pd(vfeps,Fp,Y);
 256             vvdw6            = _mm_mul_pd(c6_00,VV);
 257             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 258             fvdw6            = _mm_mul_pd(c6_00,FF);
 259
 260             /* CUBIC SPLINE TABLE REPULSION */
 261             vfitab           = _mm_add_epi32(vfitab,ifour);
 262             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 263             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 264             GMX_MM_TRANSPOSE2_PD(Y,F);
 265             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 266             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 267             GMX_MM_TRANSPOSE2_PD(G,H);
 268             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 269             VV               = _mm_macc_pd(vfeps,Fp,Y);
 270             vvdw12           = _mm_mul_pd(c12_00,VV);
 271             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 272             fvdw12           = _mm_mul_pd(c12_00,FF);
 273             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 274             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 278
 279             fscal            = fvdw;
 280
 281             /* Update vectorial force */
 282             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 283             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 284             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 285
 286             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 287             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 288             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             if (gmx_mm_any_lt(rsq10,rcutoff2))
 295             {
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq10             = _mm_mul_pd(iq1,jq0);
 299
 300             /* REACTION-FIELD ELECTROSTATICS */
 301             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 302             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 303
 304             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velec            = _mm_and_pd(velec,cutoff_mask);
 308             velecsum         = _mm_add_pd(velecsum,velec);
 309
 310             fscal            = felec;
 311
 312             fscal            = _mm_and_pd(fscal,cutoff_mask);
 313
 314             /* Update vectorial force */
 315             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 316             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 317             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 318
 319             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 320             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 321             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 322
 323             }
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             if (gmx_mm_any_lt(rsq20,rcutoff2))
 330             {
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq20             = _mm_mul_pd(iq2,jq0);
 334
 335             /* REACTION-FIELD ELECTROSTATICS */
 336             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 337             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 338
 339             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _mm_and_pd(velec,cutoff_mask);
 343             velecsum         = _mm_add_pd(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             fscal            = _mm_and_pd(fscal,cutoff_mask);
 348
 349             /* Update vectorial force */
 350             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 351             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 352             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 353
 354             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 355             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 356             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 357
 358             }
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             if (gmx_mm_any_lt(rsq30,rcutoff2))
 365             {
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq30             = _mm_mul_pd(iq3,jq0);
 369
 370             /* REACTION-FIELD ELECTROSTATICS */
 371             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 372             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 373
 374             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velec            = _mm_and_pd(velec,cutoff_mask);
 378             velecsum         = _mm_add_pd(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             fscal            = _mm_and_pd(fscal,cutoff_mask);
 383
 384             /* Update vectorial force */
 385             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 386             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 387             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 388
 389             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 390             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 391             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 392
 393             }
 394
 395             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 396
 397             /* Inner loop uses 179 flops */
 398         }
 399
 400         if(jidx<j_index_end)
 401         {
 402
 403             jnrA             = jjnr[jidx];
 404             j_coord_offsetA  = DIM*jnrA;
 405
 406             /* load j atom coordinates */
 407             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 408                                               &jx0,&jy0,&jz0);
 409
 410             /* Calculate displacement vector */
 411             dx00             = _mm_sub_pd(ix0,jx0);
 412             dy00             = _mm_sub_pd(iy0,jy0);
 413             dz00             = _mm_sub_pd(iz0,jz0);
 414             dx10             = _mm_sub_pd(ix1,jx0);
 415             dy10             = _mm_sub_pd(iy1,jy0);
 416             dz10             = _mm_sub_pd(iz1,jz0);
 417             dx20             = _mm_sub_pd(ix2,jx0);
 418             dy20             = _mm_sub_pd(iy2,jy0);
 419             dz20             = _mm_sub_pd(iz2,jz0);
 420             dx30             = _mm_sub_pd(ix3,jx0);
 421             dy30             = _mm_sub_pd(iy3,jy0);
 422             dz30             = _mm_sub_pd(iz3,jz0);
 423
 424             /* Calculate squared distance and things based on it */
 425             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 426             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 427             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 428             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 429
 430             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 431             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 432             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 433             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 434
 435             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 436             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 437             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 438
 439             /* Load parameters for j particles */
 440             jq0              = _mm_load_sd(charge+jnrA+0);
 441             vdwjidx0A        = 2*vdwtype[jnrA+0];
 442
 443             fjx0             = _mm_setzero_pd();
 444             fjy0             = _mm_setzero_pd();
 445             fjz0             = _mm_setzero_pd();
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             r00              = _mm_mul_pd(rsq00,rinv00);
 452
 453             /* Compute parameters for interactions between i and j atoms */
 454             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 455
 456             /* Calculate table index by multiplying r with table scale and truncate to integer */
 457             rt               = _mm_mul_pd(r00,vftabscale);
 458             vfitab           = _mm_cvttpd_epi32(rt);
 459 #ifdef __XOP__
 460             vfeps            = _mm_frcz_pd(rt);
 461 #else
 462             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 463 #endif
 464             twovfeps         = _mm_add_pd(vfeps,vfeps);
 465             vfitab           = _mm_slli_epi32(vfitab,3);
 466
 467             /* CUBIC SPLINE TABLE DISPERSION */
 468             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 469             F                = _mm_setzero_pd();
 470             GMX_MM_TRANSPOSE2_PD(Y,F);
 471             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 472             H                = _mm_setzero_pd();
 473             GMX_MM_TRANSPOSE2_PD(G,H);
 474             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 475             VV               = _mm_macc_pd(vfeps,Fp,Y);
 476             vvdw6            = _mm_mul_pd(c6_00,VV);
 477             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 478             fvdw6            = _mm_mul_pd(c6_00,FF);
 479
 480             /* CUBIC SPLINE TABLE REPULSION */
 481             vfitab           = _mm_add_epi32(vfitab,ifour);
 482             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 483             F                = _mm_setzero_pd();
 484             GMX_MM_TRANSPOSE2_PD(Y,F);
 485             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 486             H                = _mm_setzero_pd();
 487             GMX_MM_TRANSPOSE2_PD(G,H);
 488             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 489             VV               = _mm_macc_pd(vfeps,Fp,Y);
 490             vvdw12           = _mm_mul_pd(c12_00,VV);
 491             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 492             fvdw12           = _mm_mul_pd(c12_00,FF);
 493             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 494             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 495
 496             /* Update potential sum for this i atom from the interaction with this j atom. */
 497             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 498             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 499
 500             fscal            = fvdw;
 501
 502             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 503
 504             /* Update vectorial force */
 505             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 506             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 507             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 508
 509             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 510             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 511             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             if (gmx_mm_any_lt(rsq10,rcutoff2))
 518             {
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq10             = _mm_mul_pd(iq1,jq0);
 522
 523             /* REACTION-FIELD ELECTROSTATICS */
 524             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 525             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 526
 527             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm_and_pd(velec,cutoff_mask);
 531             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 532             velecsum         = _mm_add_pd(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm_and_pd(fscal,cutoff_mask);
 537
 538             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 539
 540             /* Update vectorial force */
 541             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 542             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 543             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 544
 545             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 546             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 547             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 548
 549             }
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             if (gmx_mm_any_lt(rsq20,rcutoff2))
 556             {
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq20             = _mm_mul_pd(iq2,jq0);
 560
 561             /* REACTION-FIELD ELECTROSTATICS */
 562             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 563             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 564
 565             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _mm_and_pd(velec,cutoff_mask);
 569             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 570             velecsum         = _mm_add_pd(velecsum,velec);
 571
 572             fscal            = felec;
 573
 574             fscal            = _mm_and_pd(fscal,cutoff_mask);
 575
 576             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 577
 578             /* Update vectorial force */
 579             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 580             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 581             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 582
 583             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 584             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 585             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 586
 587             }
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             if (gmx_mm_any_lt(rsq30,rcutoff2))
 594             {
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq30             = _mm_mul_pd(iq3,jq0);
 598
 599             /* REACTION-FIELD ELECTROSTATICS */
 600             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 601             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 602
 603             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 604
 605             /* Update potential sum for this i atom from the interaction with this j atom. */
 606             velec            = _mm_and_pd(velec,cutoff_mask);
 607             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 608             velecsum         = _mm_add_pd(velecsum,velec);
 609
 610             fscal            = felec;
 611
 612             fscal            = _mm_and_pd(fscal,cutoff_mask);
 613
 614             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 615
 616             /* Update vectorial force */
 617             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 618             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 619             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 620
 621             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 622             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 623             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 624
 625             }
 626
 627             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 628
 629             /* Inner loop uses 179 flops */
 630         }
 631
 632         /* End of innermost loop */
 633
 634         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 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 26 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_W4_VF,outeriter*26 + inneriter*179);
 654 }
 655 /*
 656  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 657  * Electrostatics interaction: ReactionField
 658  * VdW interaction:            CubicSplineTable
 659  * Geometry:                   Water4-Particle
 660  * Calculate force/pot:        Force
 661  */
 662 void
 663 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_avx_128_fma_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              vdwioffset3;
 692     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 693     int              vdwjidx0A,vdwjidx0B;
 694     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 695     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 696     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 697     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 698     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 699     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 700     real             *charge;
 701     int              nvdwtype;
 702     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 703     int              *vdwtype;
 704     real             *vdwparam;
 705     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 706     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 707     __m128i          vfitab;
 708     __m128i          ifour       = _mm_set1_epi32(4);
 709     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 710     real             *vftab;
 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     krf              = _mm_set1_pd(fr->ic->k_rf);
 729     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 730     crf              = _mm_set1_pd(fr->ic->c_rf);
 731     nvdwtype         = fr->ntype;
 732     vdwparam         = fr->nbfp;
 733     vdwtype          = mdatoms->typeA;
 734
 735     vftab            = kernel_data->table_vdw->data;
 736     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 737
 738     /* Setup water-specific parameters */
 739     inr              = nlist->iinr[0];
 740     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 741     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 742     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 743     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 744
 745     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 746     rcutoff_scalar   = fr->rcoulomb;
 747     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 748     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 749
 750     /* Avoid stupid compiler warnings */
 751     jnrA = jnrB = 0;
 752     j_coord_offsetA = 0;
 753     j_coord_offsetB = 0;
 754
 755     outeriter        = 0;
 756     inneriter        = 0;
 757
 758     /* Start outer loop over neighborlists */
 759     for(iidx=0; iidx<nri; iidx++)
 760     {
 761         /* Load shift vector for this list */
 762         i_shift_offset   = DIM*shiftidx[iidx];
 763
 764         /* Load limits for loop over neighbors */
 765         j_index_start    = jindex[iidx];
 766         j_index_end      = jindex[iidx+1];
 767
 768         /* Get outer coordinate index */
 769         inr              = iinr[iidx];
 770         i_coord_offset   = DIM*inr;
 771
 772         /* Load i particle coords and add shift vector */
 773         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 774                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 775
 776         fix0             = _mm_setzero_pd();
 777         fiy0             = _mm_setzero_pd();
 778         fiz0             = _mm_setzero_pd();
 779         fix1             = _mm_setzero_pd();
 780         fiy1             = _mm_setzero_pd();
 781         fiz1             = _mm_setzero_pd();
 782         fix2             = _mm_setzero_pd();
 783         fiy2             = _mm_setzero_pd();
 784         fiz2             = _mm_setzero_pd();
 785         fix3             = _mm_setzero_pd();
 786         fiy3             = _mm_setzero_pd();
 787         fiz3             = _mm_setzero_pd();
 788
 789         /* Start inner kernel loop */
 790         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 791         {
 792
 793             /* Get j neighbor index, and coordinate index */
 794             jnrA             = jjnr[jidx];
 795             jnrB             = jjnr[jidx+1];
 796             j_coord_offsetA  = DIM*jnrA;
 797             j_coord_offsetB  = DIM*jnrB;
 798
 799             /* load j atom coordinates */
 800             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 801                                               &jx0,&jy0,&jz0);
 802
 803             /* Calculate displacement vector */
 804             dx00             = _mm_sub_pd(ix0,jx0);
 805             dy00             = _mm_sub_pd(iy0,jy0);
 806             dz00             = _mm_sub_pd(iz0,jz0);
 807             dx10             = _mm_sub_pd(ix1,jx0);
 808             dy10             = _mm_sub_pd(iy1,jy0);
 809             dz10             = _mm_sub_pd(iz1,jz0);
 810             dx20             = _mm_sub_pd(ix2,jx0);
 811             dy20             = _mm_sub_pd(iy2,jy0);
 812             dz20             = _mm_sub_pd(iz2,jz0);
 813             dx30             = _mm_sub_pd(ix3,jx0);
 814             dy30             = _mm_sub_pd(iy3,jy0);
 815             dz30             = _mm_sub_pd(iz3,jz0);
 816
 817             /* Calculate squared distance and things based on it */
 818             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 819             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 820             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 821             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 822
 823             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 824             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 825             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 826             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 827
 828             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 829             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 830             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 831
 832             /* Load parameters for j particles */
 833             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 834             vdwjidx0A        = 2*vdwtype[jnrA+0];
 835             vdwjidx0B        = 2*vdwtype[jnrB+0];
 836
 837             fjx0             = _mm_setzero_pd();
 838             fjy0             = _mm_setzero_pd();
 839             fjz0             = _mm_setzero_pd();
 840
 841             /**************************
 842              * CALCULATE INTERACTIONS *
 843              **************************/
 844
 845             r00              = _mm_mul_pd(rsq00,rinv00);
 846
 847             /* Compute parameters for interactions between i and j atoms */
 848             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 849                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 850
 851             /* Calculate table index by multiplying r with table scale and truncate to integer */
 852             rt               = _mm_mul_pd(r00,vftabscale);
 853             vfitab           = _mm_cvttpd_epi32(rt);
 854 #ifdef __XOP__
 855             vfeps            = _mm_frcz_pd(rt);
 856 #else
 857             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 858 #endif
 859             twovfeps         = _mm_add_pd(vfeps,vfeps);
 860             vfitab           = _mm_slli_epi32(vfitab,3);
 861
 862             /* CUBIC SPLINE TABLE DISPERSION */
 863             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 864             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 865             GMX_MM_TRANSPOSE2_PD(Y,F);
 866             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 867             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 868             GMX_MM_TRANSPOSE2_PD(G,H);
 869             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 870             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 871             fvdw6            = _mm_mul_pd(c6_00,FF);
 872
 873             /* CUBIC SPLINE TABLE REPULSION */
 874             vfitab           = _mm_add_epi32(vfitab,ifour);
 875             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 876             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 877             GMX_MM_TRANSPOSE2_PD(Y,F);
 878             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 879             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 880             GMX_MM_TRANSPOSE2_PD(G,H);
 881             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 882             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 883             fvdw12           = _mm_mul_pd(c12_00,FF);
 884             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 885
 886             fscal            = fvdw;
 887
 888             /* Update vectorial force */
 889             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 890             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 891             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 892
 893             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 894             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 895             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             if (gmx_mm_any_lt(rsq10,rcutoff2))
 902             {
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             qq10             = _mm_mul_pd(iq1,jq0);
 906
 907             /* REACTION-FIELD ELECTROSTATICS */
 908             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 909
 910             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 911
 912             fscal            = felec;
 913
 914             fscal            = _mm_and_pd(fscal,cutoff_mask);
 915
 916             /* Update vectorial force */
 917             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 918             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 919             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 920
 921             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 922             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 923             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 924
 925             }
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             if (gmx_mm_any_lt(rsq20,rcutoff2))
 932             {
 933
 934             /* Compute parameters for interactions between i and j atoms */
 935             qq20             = _mm_mul_pd(iq2,jq0);
 936
 937             /* REACTION-FIELD ELECTROSTATICS */
 938             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 939
 940             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 941
 942             fscal            = felec;
 943
 944             fscal            = _mm_and_pd(fscal,cutoff_mask);
 945
 946             /* Update vectorial force */
 947             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 948             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 949             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 950
 951             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 952             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 953             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 954
 955             }
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             if (gmx_mm_any_lt(rsq30,rcutoff2))
 962             {
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq30             = _mm_mul_pd(iq3,jq0);
 966
 967             /* REACTION-FIELD ELECTROSTATICS */
 968             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 969
 970             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 971
 972             fscal            = felec;
 973
 974             fscal            = _mm_and_pd(fscal,cutoff_mask);
 975
 976             /* Update vectorial force */
 977             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 978             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 979             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 980
 981             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 982             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 983             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 984
 985             }
 986
 987             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 988
 989             /* Inner loop uses 153 flops */
 990         }
 991
 992         if(jidx<j_index_end)
 993         {
 994
 995             jnrA             = jjnr[jidx];
 996             j_coord_offsetA  = DIM*jnrA;
 997
 998             /* load j atom coordinates */
 999             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1000                                               &jx0,&jy0,&jz0);
1001
1002             /* Calculate displacement vector */
1003             dx00             = _mm_sub_pd(ix0,jx0);
1004             dy00             = _mm_sub_pd(iy0,jy0);
1005             dz00             = _mm_sub_pd(iz0,jz0);
1006             dx10             = _mm_sub_pd(ix1,jx0);
1007             dy10             = _mm_sub_pd(iy1,jy0);
1008             dz10             = _mm_sub_pd(iz1,jz0);
1009             dx20             = _mm_sub_pd(ix2,jx0);
1010             dy20             = _mm_sub_pd(iy2,jy0);
1011             dz20             = _mm_sub_pd(iz2,jz0);
1012             dx30             = _mm_sub_pd(ix3,jx0);
1013             dy30             = _mm_sub_pd(iy3,jy0);
1014             dz30             = _mm_sub_pd(iz3,jz0);
1015
1016             /* Calculate squared distance and things based on it */
1017             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1018             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1019             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1020             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1021
1022             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1023             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1024             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1025             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1026
1027             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1028             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1029             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1030
1031             /* Load parameters for j particles */
1032             jq0              = _mm_load_sd(charge+jnrA+0);
1033             vdwjidx0A        = 2*vdwtype[jnrA+0];
1034
1035             fjx0             = _mm_setzero_pd();
1036             fjy0             = _mm_setzero_pd();
1037             fjz0             = _mm_setzero_pd();
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             r00              = _mm_mul_pd(rsq00,rinv00);
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1047
1048             /* Calculate table index by multiplying r with table scale and truncate to integer */
1049             rt               = _mm_mul_pd(r00,vftabscale);
1050             vfitab           = _mm_cvttpd_epi32(rt);
1051 #ifdef __XOP__
1052             vfeps            = _mm_frcz_pd(rt);
1053 #else
1054             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1055 #endif
1056             twovfeps         = _mm_add_pd(vfeps,vfeps);
1057             vfitab           = _mm_slli_epi32(vfitab,3);
1058
1059             /* CUBIC SPLINE TABLE DISPERSION */
1060             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1061             F                = _mm_setzero_pd();
1062             GMX_MM_TRANSPOSE2_PD(Y,F);
1063             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1064             H                = _mm_setzero_pd();
1065             GMX_MM_TRANSPOSE2_PD(G,H);
1066             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1067             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1068             fvdw6            = _mm_mul_pd(c6_00,FF);
1069
1070             /* CUBIC SPLINE TABLE REPULSION */
1071             vfitab           = _mm_add_epi32(vfitab,ifour);
1072             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1073             F                = _mm_setzero_pd();
1074             GMX_MM_TRANSPOSE2_PD(Y,F);
1075             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1076             H                = _mm_setzero_pd();
1077             GMX_MM_TRANSPOSE2_PD(G,H);
1078             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1079             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1080             fvdw12           = _mm_mul_pd(c12_00,FF);
1081             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1082
1083             fscal            = fvdw;
1084
1085             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1086
1087             /* Update vectorial force */
1088             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1089             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1090             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1091
1092             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1093             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1094             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             if (gmx_mm_any_lt(rsq10,rcutoff2))
1101             {
1102
1103             /* Compute parameters for interactions between i and j atoms */
1104             qq10             = _mm_mul_pd(iq1,jq0);
1105
1106             /* REACTION-FIELD ELECTROSTATICS */
1107             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1108
1109             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1110
1111             fscal            = felec;
1112
1113             fscal            = _mm_and_pd(fscal,cutoff_mask);
1114
1115             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1116
1117             /* Update vectorial force */
1118             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1119             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1120             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1121
1122             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1123             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1124             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1125
1126             }
1127
1128             /**************************
1129              * CALCULATE INTERACTIONS *
1130              **************************/
1131
1132             if (gmx_mm_any_lt(rsq20,rcutoff2))
1133             {
1134
1135             /* Compute parameters for interactions between i and j atoms */
1136             qq20             = _mm_mul_pd(iq2,jq0);
1137
1138             /* REACTION-FIELD ELECTROSTATICS */
1139             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1140
1141             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1142
1143             fscal            = felec;
1144
1145             fscal            = _mm_and_pd(fscal,cutoff_mask);
1146
1147             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1148
1149             /* Update vectorial force */
1150             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1151             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1152             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1153
1154             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1155             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1156             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1157
1158             }
1159
1160             /**************************
1161              * CALCULATE INTERACTIONS *
1162              **************************/
1163
1164             if (gmx_mm_any_lt(rsq30,rcutoff2))
1165             {
1166
1167             /* Compute parameters for interactions between i and j atoms */
1168             qq30             = _mm_mul_pd(iq3,jq0);
1169
1170             /* REACTION-FIELD ELECTROSTATICS */
1171             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
1172
1173             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1174
1175             fscal            = felec;
1176
1177             fscal            = _mm_and_pd(fscal,cutoff_mask);
1178
1179             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1180
1181             /* Update vectorial force */
1182             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1183             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1184             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1185
1186             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1187             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1188             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1189
1190             }
1191
1192             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1193
1194             /* Inner loop uses 153 flops */
1195         }
1196
1197         /* End of innermost loop */
1198
1199         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1200                                               f+i_coord_offset,fshift+i_shift_offset);
1201
1202         /* Increment number of inner iterations */
1203         inneriter                  += j_index_end - j_index_start;
1204
1205         /* Outer loop uses 24 flops */
1206     }
1207
1208     /* Increment number of outer iterations */
1209     outeriter        += nri;
1210
1211     /* Update outer/inner flops */
1212
1213     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*153);
1214 }