src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_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_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_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              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  78     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  79     __m128i          vfitab;
  80     __m128i          ifour       = _mm_set1_epi32(4);
  81     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  82     real             *vftab;
  83     __m128d          dummy_mask,cutoff_mask;
  84     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  85     __m128d          one     = _mm_set1_pd(1.0);
  86     __m128d          two     = _mm_set1_pd(2.0);
  87     x                = xx[0];
  88     f                = ff[0];
  89
  90     nri              = nlist->nri;
  91     iinr             = nlist->iinr;
  92     jindex           = nlist->jindex;
  93     jjnr             = nlist->jjnr;
  94     shiftidx         = nlist->shift;
  95     gid              = nlist->gid;
  96     shiftvec         = fr->shift_vec[0];
  97     fshift           = fr->fshift[0];
  98     facel            = _mm_set1_pd(fr->epsfac);
  99     charge           = mdatoms->chargeA;
 100     krf              = _mm_set1_pd(fr->ic->k_rf);
 101     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 102     crf              = _mm_set1_pd(fr->ic->c_rf);
 103     nvdwtype         = fr->ntype;
 104     vdwparam         = fr->nbfp;
 105     vdwtype          = mdatoms->typeA;
 106
 107     vftab            = kernel_data->table_vdw->data;
 108     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 109
 110     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 111     rcutoff_scalar   = fr->rcoulomb;
 112     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 113     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     /* Start outer loop over neighborlists */
 124     for(iidx=0; iidx<nri; iidx++)
 125     {
 126         /* Load shift vector for this list */
 127         i_shift_offset   = DIM*shiftidx[iidx];
 128
 129         /* Load limits for loop over neighbors */
 130         j_index_start    = jindex[iidx];
 131         j_index_end      = jindex[iidx+1];
 132
 133         /* Get outer coordinate index */
 134         inr              = iinr[iidx];
 135         i_coord_offset   = DIM*inr;
 136
 137         /* Load i particle coords and add shift vector */
 138         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 139
 140         fix0             = _mm_setzero_pd();
 141         fiy0             = _mm_setzero_pd();
 142         fiz0             = _mm_setzero_pd();
 143
 144         /* Load parameters for i particles */
 145         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 146         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148         /* Reset potential sums */
 149         velecsum         = _mm_setzero_pd();
 150         vvdwsum          = _mm_setzero_pd();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161
 162             /* load j atom coordinates */
 163             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 164                                               &jx0,&jy0,&jz0);
 165
 166             /* Calculate displacement vector */
 167             dx00             = _mm_sub_pd(ix0,jx0);
 168             dy00             = _mm_sub_pd(iy0,jy0);
 169             dz00             = _mm_sub_pd(iz0,jz0);
 170
 171             /* Calculate squared distance and things based on it */
 172             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 173
 174             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 175
 176             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 177
 178             /* Load parameters for j particles */
 179             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 180             vdwjidx0A        = 2*vdwtype[jnrA+0];
 181             vdwjidx0B        = 2*vdwtype[jnrB+0];
 182
 183             /**************************
 184              * CALCULATE INTERACTIONS *
 185              **************************/
 186
 187             if (gmx_mm_any_lt(rsq00,rcutoff2))
 188             {
 189
 190             r00              = _mm_mul_pd(rsq00,rinv00);
 191
 192             /* Compute parameters for interactions between i and j atoms */
 193             qq00             = _mm_mul_pd(iq0,jq0);
 194             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 195                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 196
 197             /* Calculate table index by multiplying r with table scale and truncate to integer */
 198             rt               = _mm_mul_pd(r00,vftabscale);
 199             vfitab           = _mm_cvttpd_epi32(rt);
 200 #ifdef __XOP__
 201             vfeps            = _mm_frcz_pd(rt);
 202 #else
 203             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 204 #endif
 205             twovfeps         = _mm_add_pd(vfeps,vfeps);
 206             vfitab           = _mm_slli_epi32(vfitab,3);
 207
 208             /* REACTION-FIELD ELECTROSTATICS */
 209             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 210             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 211
 212             /* CUBIC SPLINE TABLE DISPERSION */
 213             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 214             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 215             GMX_MM_TRANSPOSE2_PD(Y,F);
 216             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 217             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 218             GMX_MM_TRANSPOSE2_PD(G,H);
 219             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 220             VV               = _mm_macc_pd(vfeps,Fp,Y);
 221             vvdw6            = _mm_mul_pd(c6_00,VV);
 222             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 223             fvdw6            = _mm_mul_pd(c6_00,FF);
 224
 225             /* CUBIC SPLINE TABLE REPULSION */
 226             vfitab           = _mm_add_epi32(vfitab,ifour);
 227             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 228             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 229             GMX_MM_TRANSPOSE2_PD(Y,F);
 230             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 231             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 232             GMX_MM_TRANSPOSE2_PD(G,H);
 233             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 234             VV               = _mm_macc_pd(vfeps,Fp,Y);
 235             vvdw12           = _mm_mul_pd(c12_00,VV);
 236             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 237             fvdw12           = _mm_mul_pd(c12_00,FF);
 238             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 239             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 240
 241             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velec            = _mm_and_pd(velec,cutoff_mask);
 245             velecsum         = _mm_add_pd(velecsum,velec);
 246             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 247             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 248
 249             fscal            = _mm_add_pd(felec,fvdw);
 250
 251             fscal            = _mm_and_pd(fscal,cutoff_mask);
 252
 253             /* Update vectorial force */
 254             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 255             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 256             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 257
 258             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 259                                                    _mm_mul_pd(dx00,fscal),
 260                                                    _mm_mul_pd(dy00,fscal),
 261                                                    _mm_mul_pd(dz00,fscal));
 262
 263             }
 264
 265             /* Inner loop uses 75 flops */
 266         }
 267
 268         if(jidx<j_index_end)
 269         {
 270
 271             jnrA             = jjnr[jidx];
 272             j_coord_offsetA  = DIM*jnrA;
 273
 274             /* load j atom coordinates */
 275             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 276                                               &jx0,&jy0,&jz0);
 277
 278             /* Calculate displacement vector */
 279             dx00             = _mm_sub_pd(ix0,jx0);
 280             dy00             = _mm_sub_pd(iy0,jy0);
 281             dz00             = _mm_sub_pd(iz0,jz0);
 282
 283             /* Calculate squared distance and things based on it */
 284             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 285
 286             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 287
 288             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 289
 290             /* Load parameters for j particles */
 291             jq0              = _mm_load_sd(charge+jnrA+0);
 292             vdwjidx0A        = 2*vdwtype[jnrA+0];
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq00,rcutoff2))
 299             {
 300
 301             r00              = _mm_mul_pd(rsq00,rinv00);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq00             = _mm_mul_pd(iq0,jq0);
 305             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 306
 307             /* Calculate table index by multiplying r with table scale and truncate to integer */
 308             rt               = _mm_mul_pd(r00,vftabscale);
 309             vfitab           = _mm_cvttpd_epi32(rt);
 310 #ifdef __XOP__
 311             vfeps            = _mm_frcz_pd(rt);
 312 #else
 313             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 314 #endif
 315             twovfeps         = _mm_add_pd(vfeps,vfeps);
 316             vfitab           = _mm_slli_epi32(vfitab,3);
 317
 318             /* REACTION-FIELD ELECTROSTATICS */
 319             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 320             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 321
 322             /* CUBIC SPLINE TABLE DISPERSION */
 323             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 324             F                = _mm_setzero_pd();
 325             GMX_MM_TRANSPOSE2_PD(Y,F);
 326             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 327             H                = _mm_setzero_pd();
 328             GMX_MM_TRANSPOSE2_PD(G,H);
 329             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 330             VV               = _mm_macc_pd(vfeps,Fp,Y);
 331             vvdw6            = _mm_mul_pd(c6_00,VV);
 332             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 333             fvdw6            = _mm_mul_pd(c6_00,FF);
 334
 335             /* CUBIC SPLINE TABLE REPULSION */
 336             vfitab           = _mm_add_epi32(vfitab,ifour);
 337             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 338             F                = _mm_setzero_pd();
 339             GMX_MM_TRANSPOSE2_PD(Y,F);
 340             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 341             H                = _mm_setzero_pd();
 342             GMX_MM_TRANSPOSE2_PD(G,H);
 343             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 344             VV               = _mm_macc_pd(vfeps,Fp,Y);
 345             vvdw12           = _mm_mul_pd(c12_00,VV);
 346             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 347             fvdw12           = _mm_mul_pd(c12_00,FF);
 348             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 349             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 350
 351             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velec            = _mm_and_pd(velec,cutoff_mask);
 355             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 356             velecsum         = _mm_add_pd(velecsum,velec);
 357             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 358             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 359             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 360
 361             fscal            = _mm_add_pd(felec,fvdw);
 362
 363             fscal            = _mm_and_pd(fscal,cutoff_mask);
 364
 365             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 366
 367             /* Update vectorial force */
 368             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 369             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 370             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 371
 372             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 373                                                    _mm_mul_pd(dx00,fscal),
 374                                                    _mm_mul_pd(dy00,fscal),
 375                                                    _mm_mul_pd(dz00,fscal));
 376
 377             }
 378
 379             /* Inner loop uses 75 flops */
 380         }
 381
 382         /* End of innermost loop */
 383
 384         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 385                                               f+i_coord_offset,fshift+i_shift_offset);
 386
 387         ggid                        = gid[iidx];
 388         /* Update potential energies */
 389         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 390         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 391
 392         /* Increment number of inner iterations */
 393         inneriter                  += j_index_end - j_index_start;
 394
 395         /* Outer loop uses 9 flops */
 396     }
 397
 398     /* Increment number of outer iterations */
 399     outeriter        += nri;
 400
 401     /* Update outer/inner flops */
 402
 403     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*75);
 404 }
 405 /*
 406  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 407  * Electrostatics interaction: ReactionField
 408  * VdW interaction:            CubicSplineTable
 409  * Geometry:                   Particle-Particle
 410  * Calculate force/pot:        Force
 411  */
 412 void
 413 nb_kernel_ElecRFCut_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 414                     (t_nblist * gmx_restrict                nlist,
 415                      rvec * gmx_restrict                    xx,
 416                      rvec * gmx_restrict                    ff,
 417                      t_forcerec * gmx_restrict              fr,
 418                      t_mdatoms * gmx_restrict               mdatoms,
 419                      nb_kernel_data_t * gmx_restrict        kernel_data,
 420                      t_nrnb * gmx_restrict                  nrnb)
 421 {
 422     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 423      * just 0 for non-waters.
 424      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 425      * jnr indices corresponding to data put in the four positions in the SIMD register.
 426      */
 427     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 428     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 429     int              jnrA,jnrB;
 430     int              j_coord_offsetA,j_coord_offsetB;
 431     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 432     real             rcutoff_scalar;
 433     real             *shiftvec,*fshift,*x,*f;
 434     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 435     int              vdwioffset0;
 436     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 437     int              vdwjidx0A,vdwjidx0B;
 438     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 439     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 440     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 441     real             *charge;
 442     int              nvdwtype;
 443     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 444     int              *vdwtype;
 445     real             *vdwparam;
 446     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 447     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 448     __m128i          vfitab;
 449     __m128i          ifour       = _mm_set1_epi32(4);
 450     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 451     real             *vftab;
 452     __m128d          dummy_mask,cutoff_mask;
 453     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 454     __m128d          one     = _mm_set1_pd(1.0);
 455     __m128d          two     = _mm_set1_pd(2.0);
 456     x                = xx[0];
 457     f                = ff[0];
 458
 459     nri              = nlist->nri;
 460     iinr             = nlist->iinr;
 461     jindex           = nlist->jindex;
 462     jjnr             = nlist->jjnr;
 463     shiftidx         = nlist->shift;
 464     gid              = nlist->gid;
 465     shiftvec         = fr->shift_vec[0];
 466     fshift           = fr->fshift[0];
 467     facel            = _mm_set1_pd(fr->epsfac);
 468     charge           = mdatoms->chargeA;
 469     krf              = _mm_set1_pd(fr->ic->k_rf);
 470     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 471     crf              = _mm_set1_pd(fr->ic->c_rf);
 472     nvdwtype         = fr->ntype;
 473     vdwparam         = fr->nbfp;
 474     vdwtype          = mdatoms->typeA;
 475
 476     vftab            = kernel_data->table_vdw->data;
 477     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 478
 479     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 480     rcutoff_scalar   = fr->rcoulomb;
 481     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 482     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 483
 484     /* Avoid stupid compiler warnings */
 485     jnrA = jnrB = 0;
 486     j_coord_offsetA = 0;
 487     j_coord_offsetB = 0;
 488
 489     outeriter        = 0;
 490     inneriter        = 0;
 491
 492     /* Start outer loop over neighborlists */
 493     for(iidx=0; iidx<nri; iidx++)
 494     {
 495         /* Load shift vector for this list */
 496         i_shift_offset   = DIM*shiftidx[iidx];
 497
 498         /* Load limits for loop over neighbors */
 499         j_index_start    = jindex[iidx];
 500         j_index_end      = jindex[iidx+1];
 501
 502         /* Get outer coordinate index */
 503         inr              = iinr[iidx];
 504         i_coord_offset   = DIM*inr;
 505
 506         /* Load i particle coords and add shift vector */
 507         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 508
 509         fix0             = _mm_setzero_pd();
 510         fiy0             = _mm_setzero_pd();
 511         fiz0             = _mm_setzero_pd();
 512
 513         /* Load parameters for i particles */
 514         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 515         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 516
 517         /* Start inner kernel loop */
 518         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 519         {
 520
 521             /* Get j neighbor index, and coordinate index */
 522             jnrA             = jjnr[jidx];
 523             jnrB             = jjnr[jidx+1];
 524             j_coord_offsetA  = DIM*jnrA;
 525             j_coord_offsetB  = DIM*jnrB;
 526
 527             /* load j atom coordinates */
 528             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 529                                               &jx0,&jy0,&jz0);
 530
 531             /* Calculate displacement vector */
 532             dx00             = _mm_sub_pd(ix0,jx0);
 533             dy00             = _mm_sub_pd(iy0,jy0);
 534             dz00             = _mm_sub_pd(iz0,jz0);
 535
 536             /* Calculate squared distance and things based on it */
 537             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 538
 539             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 540
 541             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 542
 543             /* Load parameters for j particles */
 544             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 545             vdwjidx0A        = 2*vdwtype[jnrA+0];
 546             vdwjidx0B        = 2*vdwtype[jnrB+0];
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             if (gmx_mm_any_lt(rsq00,rcutoff2))
 553             {
 554
 555             r00              = _mm_mul_pd(rsq00,rinv00);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq00             = _mm_mul_pd(iq0,jq0);
 559             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 560                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 561
 562             /* Calculate table index by multiplying r with table scale and truncate to integer */
 563             rt               = _mm_mul_pd(r00,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(vfitab,3);
 572
 573             /* REACTION-FIELD ELECTROSTATICS */
 574             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 575
 576             /* CUBIC SPLINE TABLE DISPERSION */
 577             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 578             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 579             GMX_MM_TRANSPOSE2_PD(Y,F);
 580             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 581             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 582             GMX_MM_TRANSPOSE2_PD(G,H);
 583             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 584             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 585             fvdw6            = _mm_mul_pd(c6_00,FF);
 586
 587             /* CUBIC SPLINE TABLE REPULSION */
 588             vfitab           = _mm_add_epi32(vfitab,ifour);
 589             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 590             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 591             GMX_MM_TRANSPOSE2_PD(Y,F);
 592             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 593             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 594             GMX_MM_TRANSPOSE2_PD(G,H);
 595             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 596             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 597             fvdw12           = _mm_mul_pd(c12_00,FF);
 598             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 599
 600             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 601
 602             fscal            = _mm_add_pd(felec,fvdw);
 603
 604             fscal            = _mm_and_pd(fscal,cutoff_mask);
 605
 606             /* Update vectorial force */
 607             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 608             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 609             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 610
 611             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 612                                                    _mm_mul_pd(dx00,fscal),
 613                                                    _mm_mul_pd(dy00,fscal),
 614                                                    _mm_mul_pd(dz00,fscal));
 615
 616             }
 617
 618             /* Inner loop uses 60 flops */
 619         }
 620
 621         if(jidx<j_index_end)
 622         {
 623
 624             jnrA             = jjnr[jidx];
 625             j_coord_offsetA  = DIM*jnrA;
 626
 627             /* load j atom coordinates */
 628             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 629                                               &jx0,&jy0,&jz0);
 630
 631             /* Calculate displacement vector */
 632             dx00             = _mm_sub_pd(ix0,jx0);
 633             dy00             = _mm_sub_pd(iy0,jy0);
 634             dz00             = _mm_sub_pd(iz0,jz0);
 635
 636             /* Calculate squared distance and things based on it */
 637             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 638
 639             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 640
 641             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 642
 643             /* Load parameters for j particles */
 644             jq0              = _mm_load_sd(charge+jnrA+0);
 645             vdwjidx0A        = 2*vdwtype[jnrA+0];
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             if (gmx_mm_any_lt(rsq00,rcutoff2))
 652             {
 653
 654             r00              = _mm_mul_pd(rsq00,rinv00);
 655
 656             /* Compute parameters for interactions between i and j atoms */
 657             qq00             = _mm_mul_pd(iq0,jq0);
 658             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 659
 660             /* Calculate table index by multiplying r with table scale and truncate to integer */
 661             rt               = _mm_mul_pd(r00,vftabscale);
 662             vfitab           = _mm_cvttpd_epi32(rt);
 663 #ifdef __XOP__
 664             vfeps            = _mm_frcz_pd(rt);
 665 #else
 666             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 667 #endif
 668             twovfeps         = _mm_add_pd(vfeps,vfeps);
 669             vfitab           = _mm_slli_epi32(vfitab,3);
 670
 671             /* REACTION-FIELD ELECTROSTATICS */
 672             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 673
 674             /* CUBIC SPLINE TABLE DISPERSION */
 675             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 676             F                = _mm_setzero_pd();
 677             GMX_MM_TRANSPOSE2_PD(Y,F);
 678             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 679             H                = _mm_setzero_pd();
 680             GMX_MM_TRANSPOSE2_PD(G,H);
 681             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 682             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 683             fvdw6            = _mm_mul_pd(c6_00,FF);
 684
 685             /* CUBIC SPLINE TABLE REPULSION */
 686             vfitab           = _mm_add_epi32(vfitab,ifour);
 687             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 688             F                = _mm_setzero_pd();
 689             GMX_MM_TRANSPOSE2_PD(Y,F);
 690             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 691             H                = _mm_setzero_pd();
 692             GMX_MM_TRANSPOSE2_PD(G,H);
 693             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 694             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 695             fvdw12           = _mm_mul_pd(c12_00,FF);
 696             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 697
 698             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 699
 700             fscal            = _mm_add_pd(felec,fvdw);
 701
 702             fscal            = _mm_and_pd(fscal,cutoff_mask);
 703
 704             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 705
 706             /* Update vectorial force */
 707             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 708             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 709             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 710
 711             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 712                                                    _mm_mul_pd(dx00,fscal),
 713                                                    _mm_mul_pd(dy00,fscal),
 714                                                    _mm_mul_pd(dz00,fscal));
 715
 716             }
 717
 718             /* Inner loop uses 60 flops */
 719         }
 720
 721         /* End of innermost loop */
 722
 723         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 724                                               f+i_coord_offset,fshift+i_shift_offset);
 725
 726         /* Increment number of inner iterations */
 727         inneriter                  += j_index_end - j_index_start;
 728
 729         /* Outer loop uses 7 flops */
 730     }
 731
 732     /* Increment number of outer iterations */
 733     outeriter        += nri;
 734
 735     /* Update outer/inner flops */
 736
 737     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 738 }