src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_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_ElecCSTab_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_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     nvdwtype         = fr->ntype;
 101     vdwparam         = fr->nbfp;
 102     vdwtype          = mdatoms->typeA;
 103
 104     vftab            = kernel_data->table_elec_vdw->data;
 105     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 106
 107     /* Avoid stupid compiler warnings */
 108     jnrA = jnrB = 0;
 109     j_coord_offsetA = 0;
 110     j_coord_offsetB = 0;
 111
 112     outeriter        = 0;
 113     inneriter        = 0;
 114
 115     /* Start outer loop over neighborlists */
 116     for(iidx=0; iidx<nri; iidx++)
 117     {
 118         /* Load shift vector for this list */
 119         i_shift_offset   = DIM*shiftidx[iidx];
 120
 121         /* Load limits for loop over neighbors */
 122         j_index_start    = jindex[iidx];
 123         j_index_end      = jindex[iidx+1];
 124
 125         /* Get outer coordinate index */
 126         inr              = iinr[iidx];
 127         i_coord_offset   = DIM*inr;
 128
 129         /* Load i particle coords and add shift vector */
 130         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 131
 132         fix0             = _mm_setzero_pd();
 133         fiy0             = _mm_setzero_pd();
 134         fiz0             = _mm_setzero_pd();
 135
 136         /* Load parameters for i particles */
 137         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 138         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140         /* Reset potential sums */
 141         velecsum         = _mm_setzero_pd();
 142         vvdwsum          = _mm_setzero_pd();
 143
 144         /* Start inner kernel loop */
 145         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 146         {
 147
 148             /* Get j neighbor index, and coordinate index */
 149             jnrA             = jjnr[jidx];
 150             jnrB             = jjnr[jidx+1];
 151             j_coord_offsetA  = DIM*jnrA;
 152             j_coord_offsetB  = DIM*jnrB;
 153
 154             /* load j atom coordinates */
 155             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 156                                               &jx0,&jy0,&jz0);
 157
 158             /* Calculate displacement vector */
 159             dx00             = _mm_sub_pd(ix0,jx0);
 160             dy00             = _mm_sub_pd(iy0,jy0);
 161             dz00             = _mm_sub_pd(iz0,jz0);
 162
 163             /* Calculate squared distance and things based on it */
 164             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 165
 166             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 167
 168             /* Load parameters for j particles */
 169             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 170             vdwjidx0A        = 2*vdwtype[jnrA+0];
 171             vdwjidx0B        = 2*vdwtype[jnrB+0];
 172
 173             /**************************
 174              * CALCULATE INTERACTIONS *
 175              **************************/
 176
 177             r00              = _mm_mul_pd(rsq00,rinv00);
 178
 179             /* Compute parameters for interactions between i and j atoms */
 180             qq00             = _mm_mul_pd(iq0,jq0);
 181             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 182                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 183
 184             /* Calculate table index by multiplying r with table scale and truncate to integer */
 185             rt               = _mm_mul_pd(r00,vftabscale);
 186             vfitab           = _mm_cvttpd_epi32(rt);
 187 #ifdef __XOP__
 188             vfeps            = _mm_frcz_pd(rt);
 189 #else
 190             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 191 #endif
 192             twovfeps         = _mm_add_pd(vfeps,vfeps);
 193             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 194
 195             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 196             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 197             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 198             GMX_MM_TRANSPOSE2_PD(Y,F);
 199             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 200             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 201             GMX_MM_TRANSPOSE2_PD(G,H);
 202             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 203             VV               = _mm_macc_pd(vfeps,Fp,Y);
 204             velec            = _mm_mul_pd(qq00,VV);
 205             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 206             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 207
 208             /* CUBIC SPLINE TABLE DISPERSION */
 209             vfitab           = _mm_add_epi32(vfitab,ifour);
 210             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 211             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 212             GMX_MM_TRANSPOSE2_PD(Y,F);
 213             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 214             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 215             GMX_MM_TRANSPOSE2_PD(G,H);
 216             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 217             VV               = _mm_macc_pd(vfeps,Fp,Y);
 218             vvdw6            = _mm_mul_pd(c6_00,VV);
 219             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 220             fvdw6            = _mm_mul_pd(c6_00,FF);
 221
 222             /* CUBIC SPLINE TABLE REPULSION */
 223             vfitab           = _mm_add_epi32(vfitab,ifour);
 224             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 225             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 226             GMX_MM_TRANSPOSE2_PD(Y,F);
 227             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 228             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 229             GMX_MM_TRANSPOSE2_PD(G,H);
 230             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 231             VV               = _mm_macc_pd(vfeps,Fp,Y);
 232             vvdw12           = _mm_mul_pd(c12_00,VV);
 233             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 234             fvdw12           = _mm_mul_pd(c12_00,FF);
 235             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 236             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 237
 238             /* Update potential sum for this i atom from the interaction with this j atom. */
 239             velecsum         = _mm_add_pd(velecsum,velec);
 240             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 241
 242             fscal            = _mm_add_pd(felec,fvdw);
 243
 244             /* Update vectorial force */
 245             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 246             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 247             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 248
 249             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 250                                                    _mm_mul_pd(dx00,fscal),
 251                                                    _mm_mul_pd(dy00,fscal),
 252                                                    _mm_mul_pd(dz00,fscal));
 253
 254             /* Inner loop uses 76 flops */
 255         }
 256
 257         if(jidx<j_index_end)
 258         {
 259
 260             jnrA             = jjnr[jidx];
 261             j_coord_offsetA  = DIM*jnrA;
 262
 263             /* load j atom coordinates */
 264             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 265                                               &jx0,&jy0,&jz0);
 266
 267             /* Calculate displacement vector */
 268             dx00             = _mm_sub_pd(ix0,jx0);
 269             dy00             = _mm_sub_pd(iy0,jy0);
 270             dz00             = _mm_sub_pd(iz0,jz0);
 271
 272             /* Calculate squared distance and things based on it */
 273             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 274
 275             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 276
 277             /* Load parameters for j particles */
 278             jq0              = _mm_load_sd(charge+jnrA+0);
 279             vdwjidx0A        = 2*vdwtype[jnrA+0];
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             r00              = _mm_mul_pd(rsq00,rinv00);
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq00             = _mm_mul_pd(iq0,jq0);
 289             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 290
 291             /* Calculate table index by multiplying r with table scale and truncate to integer */
 292             rt               = _mm_mul_pd(r00,vftabscale);
 293             vfitab           = _mm_cvttpd_epi32(rt);
 294 #ifdef __XOP__
 295             vfeps            = _mm_frcz_pd(rt);
 296 #else
 297             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 298 #endif
 299             twovfeps         = _mm_add_pd(vfeps,vfeps);
 300             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 301
 302             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 303             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 304             F                = _mm_setzero_pd();
 305             GMX_MM_TRANSPOSE2_PD(Y,F);
 306             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 307             H                = _mm_setzero_pd();
 308             GMX_MM_TRANSPOSE2_PD(G,H);
 309             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 310             VV               = _mm_macc_pd(vfeps,Fp,Y);
 311             velec            = _mm_mul_pd(qq00,VV);
 312             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 313             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 314
 315             /* CUBIC SPLINE TABLE DISPERSION */
 316             vfitab           = _mm_add_epi32(vfitab,ifour);
 317             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 318             F                = _mm_setzero_pd();
 319             GMX_MM_TRANSPOSE2_PD(Y,F);
 320             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 321             H                = _mm_setzero_pd();
 322             GMX_MM_TRANSPOSE2_PD(G,H);
 323             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 324             VV               = _mm_macc_pd(vfeps,Fp,Y);
 325             vvdw6            = _mm_mul_pd(c6_00,VV);
 326             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 327             fvdw6            = _mm_mul_pd(c6_00,FF);
 328
 329             /* CUBIC SPLINE TABLE REPULSION */
 330             vfitab           = _mm_add_epi32(vfitab,ifour);
 331             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 332             F                = _mm_setzero_pd();
 333             GMX_MM_TRANSPOSE2_PD(Y,F);
 334             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 335             H                = _mm_setzero_pd();
 336             GMX_MM_TRANSPOSE2_PD(G,H);
 337             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 338             VV               = _mm_macc_pd(vfeps,Fp,Y);
 339             vvdw12           = _mm_mul_pd(c12_00,VV);
 340             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 341             fvdw12           = _mm_mul_pd(c12_00,FF);
 342             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 343             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 347             velecsum         = _mm_add_pd(velecsum,velec);
 348             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 349             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 350
 351             fscal            = _mm_add_pd(felec,fvdw);
 352
 353             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 354
 355             /* Update vectorial force */
 356             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 357             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 358             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 359
 360             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 361                                                    _mm_mul_pd(dx00,fscal),
 362                                                    _mm_mul_pd(dy00,fscal),
 363                                                    _mm_mul_pd(dz00,fscal));
 364
 365             /* Inner loop uses 76 flops */
 366         }
 367
 368         /* End of innermost loop */
 369
 370         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 371                                               f+i_coord_offset,fshift+i_shift_offset);
 372
 373         ggid                        = gid[iidx];
 374         /* Update potential energies */
 375         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 376         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 377
 378         /* Increment number of inner iterations */
 379         inneriter                  += j_index_end - j_index_start;
 380
 381         /* Outer loop uses 9 flops */
 382     }
 383
 384     /* Increment number of outer iterations */
 385     outeriter        += nri;
 386
 387     /* Update outer/inner flops */
 388
 389     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*76);
 390 }
 391 /*
 392  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 393  * Electrostatics interaction: CubicSplineTable
 394  * VdW interaction:            CubicSplineTable
 395  * Geometry:                   Particle-Particle
 396  * Calculate force/pot:        Force
 397  */
 398 void
 399 nb_kernel_ElecCSTab_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 400                     (t_nblist * gmx_restrict                nlist,
 401                      rvec * gmx_restrict                    xx,
 402                      rvec * gmx_restrict                    ff,
 403                      t_forcerec * gmx_restrict              fr,
 404                      t_mdatoms * gmx_restrict               mdatoms,
 405                      nb_kernel_data_t * gmx_restrict        kernel_data,
 406                      t_nrnb * gmx_restrict                  nrnb)
 407 {
 408     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 409      * just 0 for non-waters.
 410      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 411      * jnr indices corresponding to data put in the four positions in the SIMD register.
 412      */
 413     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 414     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 415     int              jnrA,jnrB;
 416     int              j_coord_offsetA,j_coord_offsetB;
 417     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 418     real             rcutoff_scalar;
 419     real             *shiftvec,*fshift,*x,*f;
 420     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 421     int              vdwioffset0;
 422     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 423     int              vdwjidx0A,vdwjidx0B;
 424     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 425     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 426     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 427     real             *charge;
 428     int              nvdwtype;
 429     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 430     int              *vdwtype;
 431     real             *vdwparam;
 432     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 433     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 434     __m128i          vfitab;
 435     __m128i          ifour       = _mm_set1_epi32(4);
 436     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 437     real             *vftab;
 438     __m128d          dummy_mask,cutoff_mask;
 439     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 440     __m128d          one     = _mm_set1_pd(1.0);
 441     __m128d          two     = _mm_set1_pd(2.0);
 442     x                = xx[0];
 443     f                = ff[0];
 444
 445     nri              = nlist->nri;
 446     iinr             = nlist->iinr;
 447     jindex           = nlist->jindex;
 448     jjnr             = nlist->jjnr;
 449     shiftidx         = nlist->shift;
 450     gid              = nlist->gid;
 451     shiftvec         = fr->shift_vec[0];
 452     fshift           = fr->fshift[0];
 453     facel            = _mm_set1_pd(fr->epsfac);
 454     charge           = mdatoms->chargeA;
 455     nvdwtype         = fr->ntype;
 456     vdwparam         = fr->nbfp;
 457     vdwtype          = mdatoms->typeA;
 458
 459     vftab            = kernel_data->table_elec_vdw->data;
 460     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 461
 462     /* Avoid stupid compiler warnings */
 463     jnrA = jnrB = 0;
 464     j_coord_offsetA = 0;
 465     j_coord_offsetB = 0;
 466
 467     outeriter        = 0;
 468     inneriter        = 0;
 469
 470     /* Start outer loop over neighborlists */
 471     for(iidx=0; iidx<nri; iidx++)
 472     {
 473         /* Load shift vector for this list */
 474         i_shift_offset   = DIM*shiftidx[iidx];
 475
 476         /* Load limits for loop over neighbors */
 477         j_index_start    = jindex[iidx];
 478         j_index_end      = jindex[iidx+1];
 479
 480         /* Get outer coordinate index */
 481         inr              = iinr[iidx];
 482         i_coord_offset   = DIM*inr;
 483
 484         /* Load i particle coords and add shift vector */
 485         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 486
 487         fix0             = _mm_setzero_pd();
 488         fiy0             = _mm_setzero_pd();
 489         fiz0             = _mm_setzero_pd();
 490
 491         /* Load parameters for i particles */
 492         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 493         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 494
 495         /* Start inner kernel loop */
 496         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 497         {
 498
 499             /* Get j neighbor index, and coordinate index */
 500             jnrA             = jjnr[jidx];
 501             jnrB             = jjnr[jidx+1];
 502             j_coord_offsetA  = DIM*jnrA;
 503             j_coord_offsetB  = DIM*jnrB;
 504
 505             /* load j atom coordinates */
 506             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 507                                               &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _mm_sub_pd(ix0,jx0);
 511             dy00             = _mm_sub_pd(iy0,jy0);
 512             dz00             = _mm_sub_pd(iz0,jz0);
 513
 514             /* Calculate squared distance and things based on it */
 515             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 516
 517             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522             vdwjidx0B        = 2*vdwtype[jnrB+0];
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r00              = _mm_mul_pd(rsq00,rinv00);
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq00             = _mm_mul_pd(iq0,jq0);
 532             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 533                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 534
 535             /* Calculate table index by multiplying r with table scale and truncate to integer */
 536             rt               = _mm_mul_pd(r00,vftabscale);
 537             vfitab           = _mm_cvttpd_epi32(rt);
 538 #ifdef __XOP__
 539             vfeps            = _mm_frcz_pd(rt);
 540 #else
 541             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 542 #endif
 543             twovfeps         = _mm_add_pd(vfeps,vfeps);
 544             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 545
 546             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 547             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 548             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 549             GMX_MM_TRANSPOSE2_PD(Y,F);
 550             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 551             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 552             GMX_MM_TRANSPOSE2_PD(G,H);
 553             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 554             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 555             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 556
 557             /* CUBIC SPLINE TABLE DISPERSION */
 558             vfitab           = _mm_add_epi32(vfitab,ifour);
 559             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 560             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 561             GMX_MM_TRANSPOSE2_PD(Y,F);
 562             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 563             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 564             GMX_MM_TRANSPOSE2_PD(G,H);
 565             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 566             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 567             fvdw6            = _mm_mul_pd(c6_00,FF);
 568
 569             /* CUBIC SPLINE TABLE REPULSION */
 570             vfitab           = _mm_add_epi32(vfitab,ifour);
 571             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 572             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 573             GMX_MM_TRANSPOSE2_PD(Y,F);
 574             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 575             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 576             GMX_MM_TRANSPOSE2_PD(G,H);
 577             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 578             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 579             fvdw12           = _mm_mul_pd(c12_00,FF);
 580             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 581
 582             fscal            = _mm_add_pd(felec,fvdw);
 583
 584             /* Update vectorial force */
 585             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 586             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 587             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 588
 589             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 590                                                    _mm_mul_pd(dx00,fscal),
 591                                                    _mm_mul_pd(dy00,fscal),
 592                                                    _mm_mul_pd(dz00,fscal));
 593
 594             /* Inner loop uses 64 flops */
 595         }
 596
 597         if(jidx<j_index_end)
 598         {
 599
 600             jnrA             = jjnr[jidx];
 601             j_coord_offsetA  = DIM*jnrA;
 602
 603             /* load j atom coordinates */
 604             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 605                                               &jx0,&jy0,&jz0);
 606
 607             /* Calculate displacement vector */
 608             dx00             = _mm_sub_pd(ix0,jx0);
 609             dy00             = _mm_sub_pd(iy0,jy0);
 610             dz00             = _mm_sub_pd(iz0,jz0);
 611
 612             /* Calculate squared distance and things based on it */
 613             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 614
 615             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 616
 617             /* Load parameters for j particles */
 618             jq0              = _mm_load_sd(charge+jnrA+0);
 619             vdwjidx0A        = 2*vdwtype[jnrA+0];
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             r00              = _mm_mul_pd(rsq00,rinv00);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq00             = _mm_mul_pd(iq0,jq0);
 629             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 630
 631             /* Calculate table index by multiplying r with table scale and truncate to integer */
 632             rt               = _mm_mul_pd(r00,vftabscale);
 633             vfitab           = _mm_cvttpd_epi32(rt);
 634 #ifdef __XOP__
 635             vfeps            = _mm_frcz_pd(rt);
 636 #else
 637             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 638 #endif
 639             twovfeps         = _mm_add_pd(vfeps,vfeps);
 640             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 641
 642             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 643             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 644             F                = _mm_setzero_pd();
 645             GMX_MM_TRANSPOSE2_PD(Y,F);
 646             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 647             H                = _mm_setzero_pd();
 648             GMX_MM_TRANSPOSE2_PD(G,H);
 649             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 650             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 651             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 652
 653             /* CUBIC SPLINE TABLE DISPERSION */
 654             vfitab           = _mm_add_epi32(vfitab,ifour);
 655             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 656             F                = _mm_setzero_pd();
 657             GMX_MM_TRANSPOSE2_PD(Y,F);
 658             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 659             H                = _mm_setzero_pd();
 660             GMX_MM_TRANSPOSE2_PD(G,H);
 661             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 662             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 663             fvdw6            = _mm_mul_pd(c6_00,FF);
 664
 665             /* CUBIC SPLINE TABLE REPULSION */
 666             vfitab           = _mm_add_epi32(vfitab,ifour);
 667             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 668             F                = _mm_setzero_pd();
 669             GMX_MM_TRANSPOSE2_PD(Y,F);
 670             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 671             H                = _mm_setzero_pd();
 672             GMX_MM_TRANSPOSE2_PD(G,H);
 673             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 674             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 675             fvdw12           = _mm_mul_pd(c12_00,FF);
 676             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 677
 678             fscal            = _mm_add_pd(felec,fvdw);
 679
 680             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 681
 682             /* Update vectorial force */
 683             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 684             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 685             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 686
 687             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 688                                                    _mm_mul_pd(dx00,fscal),
 689                                                    _mm_mul_pd(dy00,fscal),
 690                                                    _mm_mul_pd(dz00,fscal));
 691
 692             /* Inner loop uses 64 flops */
 693         }
 694
 695         /* End of innermost loop */
 696
 697         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 698                                               f+i_coord_offset,fshift+i_shift_offset);
 699
 700         /* Increment number of inner iterations */
 701         inneriter                  += j_index_end - j_index_start;
 702
 703         /* Outer loop uses 7 flops */
 704     }
 705
 706     /* Increment number of outer iterations */
 707     outeriter        += nri;
 708
 709     /* Update outer/inner flops */
 710
 711     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 712 }