src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_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_ElecNone_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: None
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_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     int              nvdwtype;
  72     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  73     int              *vdwtype;
  74     real             *vdwparam;
  75     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  76     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  77     __m128i          vfitab;
  78     __m128i          ifour       = _mm_set1_epi32(4);
  79     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  80     real             *vftab;
  81     __m128d          dummy_mask,cutoff_mask;
  82     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  83     __m128d          one     = _mm_set1_pd(1.0);
  84     __m128d          two     = _mm_set1_pd(2.0);
  85     x                = xx[0];
  86     f                = ff[0];
  87
  88     nri              = nlist->nri;
  89     iinr             = nlist->iinr;
  90     jindex           = nlist->jindex;
  91     jjnr             = nlist->jjnr;
  92     shiftidx         = nlist->shift;
  93     gid              = nlist->gid;
  94     shiftvec         = fr->shift_vec[0];
  95     fshift           = fr->fshift[0];
  96     nvdwtype         = fr->ntype;
  97     vdwparam         = fr->nbfp;
  98     vdwtype          = mdatoms->typeA;
  99
 100     vftab            = kernel_data->table_vdw->data;
 101     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 102
 103     /* Avoid stupid compiler warnings */
 104     jnrA = jnrB = 0;
 105     j_coord_offsetA = 0;
 106     j_coord_offsetB = 0;
 107
 108     outeriter        = 0;
 109     inneriter        = 0;
 110
 111     /* Start outer loop over neighborlists */
 112     for(iidx=0; iidx<nri; iidx++)
 113     {
 114         /* Load shift vector for this list */
 115         i_shift_offset   = DIM*shiftidx[iidx];
 116
 117         /* Load limits for loop over neighbors */
 118         j_index_start    = jindex[iidx];
 119         j_index_end      = jindex[iidx+1];
 120
 121         /* Get outer coordinate index */
 122         inr              = iinr[iidx];
 123         i_coord_offset   = DIM*inr;
 124
 125         /* Load i particle coords and add shift vector */
 126         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 127
 128         fix0             = _mm_setzero_pd();
 129         fiy0             = _mm_setzero_pd();
 130         fiz0             = _mm_setzero_pd();
 131
 132         /* Load parameters for i particles */
 133         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135         /* Reset potential sums */
 136         vvdwsum          = _mm_setzero_pd();
 137
 138         /* Start inner kernel loop */
 139         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 140         {
 141
 142             /* Get j neighbor index, and coordinate index */
 143             jnrA             = jjnr[jidx];
 144             jnrB             = jjnr[jidx+1];
 145             j_coord_offsetA  = DIM*jnrA;
 146             j_coord_offsetB  = DIM*jnrB;
 147
 148             /* load j atom coordinates */
 149             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 150                                               &jx0,&jy0,&jz0);
 151
 152             /* Calculate displacement vector */
 153             dx00             = _mm_sub_pd(ix0,jx0);
 154             dy00             = _mm_sub_pd(iy0,jy0);
 155             dz00             = _mm_sub_pd(iz0,jz0);
 156
 157             /* Calculate squared distance and things based on it */
 158             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 159
 160             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 161
 162             /* Load parameters for j particles */
 163             vdwjidx0A        = 2*vdwtype[jnrA+0];
 164             vdwjidx0B        = 2*vdwtype[jnrB+0];
 165
 166             /**************************
 167              * CALCULATE INTERACTIONS *
 168              **************************/
 169
 170             r00              = _mm_mul_pd(rsq00,rinv00);
 171
 172             /* Compute parameters for interactions between i and j atoms */
 173             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 174                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 175
 176             /* Calculate table index by multiplying r with table scale and truncate to integer */
 177             rt               = _mm_mul_pd(r00,vftabscale);
 178             vfitab           = _mm_cvttpd_epi32(rt);
 179 #ifdef __XOP__
 180             vfeps            = _mm_frcz_pd(rt);
 181 #else
 182             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 183 #endif
 184             twovfeps         = _mm_add_pd(vfeps,vfeps);
 185             vfitab           = _mm_slli_epi32(vfitab,3);
 186
 187             /* CUBIC SPLINE TABLE DISPERSION */
 188             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 189             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 190             GMX_MM_TRANSPOSE2_PD(Y,F);
 191             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 192             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 193             GMX_MM_TRANSPOSE2_PD(G,H);
 194             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 195             VV               = _mm_macc_pd(vfeps,Fp,Y);
 196             vvdw6            = _mm_mul_pd(c6_00,VV);
 197             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 198             fvdw6            = _mm_mul_pd(c6_00,FF);
 199
 200             /* CUBIC SPLINE TABLE REPULSION */
 201             vfitab           = _mm_add_epi32(vfitab,ifour);
 202             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 203             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 204             GMX_MM_TRANSPOSE2_PD(Y,F);
 205             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 206             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 207             GMX_MM_TRANSPOSE2_PD(G,H);
 208             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 209             VV               = _mm_macc_pd(vfeps,Fp,Y);
 210             vvdw12           = _mm_mul_pd(c12_00,VV);
 211             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 212             fvdw12           = _mm_mul_pd(c12_00,FF);
 213             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 214             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 215
 216             /* Update potential sum for this i atom from the interaction with this j atom. */
 217             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 218
 219             fscal            = fvdw;
 220
 221             /* Update vectorial force */
 222             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 223             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 224             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 225
 226             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 227                                                    _mm_mul_pd(dx00,fscal),
 228                                                    _mm_mul_pd(dy00,fscal),
 229                                                    _mm_mul_pd(dz00,fscal));
 230
 231             /* Inner loop uses 59 flops */
 232         }
 233
 234         if(jidx<j_index_end)
 235         {
 236
 237             jnrA             = jjnr[jidx];
 238             j_coord_offsetA  = DIM*jnrA;
 239
 240             /* load j atom coordinates */
 241             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 242                                               &jx0,&jy0,&jz0);
 243
 244             /* Calculate displacement vector */
 245             dx00             = _mm_sub_pd(ix0,jx0);
 246             dy00             = _mm_sub_pd(iy0,jy0);
 247             dz00             = _mm_sub_pd(iz0,jz0);
 248
 249             /* Calculate squared distance and things based on it */
 250             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 251
 252             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 253
 254             /* Load parameters for j particles */
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             r00              = _mm_mul_pd(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 265
 266             /* Calculate table index by multiplying r with table scale and truncate to integer */
 267             rt               = _mm_mul_pd(r00,vftabscale);
 268             vfitab           = _mm_cvttpd_epi32(rt);
 269 #ifdef __XOP__
 270             vfeps            = _mm_frcz_pd(rt);
 271 #else
 272             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 273 #endif
 274             twovfeps         = _mm_add_pd(vfeps,vfeps);
 275             vfitab           = _mm_slli_epi32(vfitab,3);
 276
 277             /* CUBIC SPLINE TABLE DISPERSION */
 278             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 279             F                = _mm_setzero_pd();
 280             GMX_MM_TRANSPOSE2_PD(Y,F);
 281             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 282             H                = _mm_setzero_pd();
 283             GMX_MM_TRANSPOSE2_PD(G,H);
 284             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 285             VV               = _mm_macc_pd(vfeps,Fp,Y);
 286             vvdw6            = _mm_mul_pd(c6_00,VV);
 287             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 288             fvdw6            = _mm_mul_pd(c6_00,FF);
 289
 290             /* CUBIC SPLINE TABLE REPULSION */
 291             vfitab           = _mm_add_epi32(vfitab,ifour);
 292             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 293             F                = _mm_setzero_pd();
 294             GMX_MM_TRANSPOSE2_PD(Y,F);
 295             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 296             H                = _mm_setzero_pd();
 297             GMX_MM_TRANSPOSE2_PD(G,H);
 298             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 299             VV               = _mm_macc_pd(vfeps,Fp,Y);
 300             vvdw12           = _mm_mul_pd(c12_00,VV);
 301             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 302             fvdw12           = _mm_mul_pd(c12_00,FF);
 303             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 304             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 308             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 309
 310             fscal            = fvdw;
 311
 312             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 313
 314             /* Update vectorial force */
 315             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 316             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 317             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 318
 319             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 320                                                    _mm_mul_pd(dx00,fscal),
 321                                                    _mm_mul_pd(dy00,fscal),
 322                                                    _mm_mul_pd(dz00,fscal));
 323
 324             /* Inner loop uses 59 flops */
 325         }
 326
 327         /* End of innermost loop */
 328
 329         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 330                                               f+i_coord_offset,fshift+i_shift_offset);
 331
 332         ggid                        = gid[iidx];
 333         /* Update potential energies */
 334         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 335
 336         /* Increment number of inner iterations */
 337         inneriter                  += j_index_end - j_index_start;
 338
 339         /* Outer loop uses 7 flops */
 340     }
 341
 342     /* Increment number of outer iterations */
 343     outeriter        += nri;
 344
 345     /* Update outer/inner flops */
 346
 347     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 348 }
 349 /*
 350  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 351  * Electrostatics interaction: None
 352  * VdW interaction:            CubicSplineTable
 353  * Geometry:                   Particle-Particle
 354  * Calculate force/pot:        Force
 355  */
 356 void
 357 nb_kernel_ElecNone_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 358                     (t_nblist * gmx_restrict                nlist,
 359                      rvec * gmx_restrict                    xx,
 360                      rvec * gmx_restrict                    ff,
 361                      t_forcerec * gmx_restrict              fr,
 362                      t_mdatoms * gmx_restrict               mdatoms,
 363                      nb_kernel_data_t * gmx_restrict        kernel_data,
 364                      t_nrnb * gmx_restrict                  nrnb)
 365 {
 366     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 367      * just 0 for non-waters.
 368      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 369      * jnr indices corresponding to data put in the four positions in the SIMD register.
 370      */
 371     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 372     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 373     int              jnrA,jnrB;
 374     int              j_coord_offsetA,j_coord_offsetB;
 375     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 376     real             rcutoff_scalar;
 377     real             *shiftvec,*fshift,*x,*f;
 378     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 379     int              vdwioffset0;
 380     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 381     int              vdwjidx0A,vdwjidx0B;
 382     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 383     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 384     int              nvdwtype;
 385     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 386     int              *vdwtype;
 387     real             *vdwparam;
 388     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 389     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 390     __m128i          vfitab;
 391     __m128i          ifour       = _mm_set1_epi32(4);
 392     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 393     real             *vftab;
 394     __m128d          dummy_mask,cutoff_mask;
 395     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 396     __m128d          one     = _mm_set1_pd(1.0);
 397     __m128d          two     = _mm_set1_pd(2.0);
 398     x                = xx[0];
 399     f                = ff[0];
 400
 401     nri              = nlist->nri;
 402     iinr             = nlist->iinr;
 403     jindex           = nlist->jindex;
 404     jjnr             = nlist->jjnr;
 405     shiftidx         = nlist->shift;
 406     gid              = nlist->gid;
 407     shiftvec         = fr->shift_vec[0];
 408     fshift           = fr->fshift[0];
 409     nvdwtype         = fr->ntype;
 410     vdwparam         = fr->nbfp;
 411     vdwtype          = mdatoms->typeA;
 412
 413     vftab            = kernel_data->table_vdw->data;
 414     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 415
 416     /* Avoid stupid compiler warnings */
 417     jnrA = jnrB = 0;
 418     j_coord_offsetA = 0;
 419     j_coord_offsetB = 0;
 420
 421     outeriter        = 0;
 422     inneriter        = 0;
 423
 424     /* Start outer loop over neighborlists */
 425     for(iidx=0; iidx<nri; iidx++)
 426     {
 427         /* Load shift vector for this list */
 428         i_shift_offset   = DIM*shiftidx[iidx];
 429
 430         /* Load limits for loop over neighbors */
 431         j_index_start    = jindex[iidx];
 432         j_index_end      = jindex[iidx+1];
 433
 434         /* Get outer coordinate index */
 435         inr              = iinr[iidx];
 436         i_coord_offset   = DIM*inr;
 437
 438         /* Load i particle coords and add shift vector */
 439         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 440
 441         fix0             = _mm_setzero_pd();
 442         fiy0             = _mm_setzero_pd();
 443         fiz0             = _mm_setzero_pd();
 444
 445         /* Load parameters for i particles */
 446         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 447
 448         /* Start inner kernel loop */
 449         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 450         {
 451
 452             /* Get j neighbor index, and coordinate index */
 453             jnrA             = jjnr[jidx];
 454             jnrB             = jjnr[jidx+1];
 455             j_coord_offsetA  = DIM*jnrA;
 456             j_coord_offsetB  = DIM*jnrB;
 457
 458             /* load j atom coordinates */
 459             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 460                                               &jx0,&jy0,&jz0);
 461
 462             /* Calculate displacement vector */
 463             dx00             = _mm_sub_pd(ix0,jx0);
 464             dy00             = _mm_sub_pd(iy0,jy0);
 465             dz00             = _mm_sub_pd(iz0,jz0);
 466
 467             /* Calculate squared distance and things based on it */
 468             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 469
 470             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 471
 472             /* Load parameters for j particles */
 473             vdwjidx0A        = 2*vdwtype[jnrA+0];
 474             vdwjidx0B        = 2*vdwtype[jnrB+0];
 475
 476             /**************************
 477              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             r00              = _mm_mul_pd(rsq00,rinv00);
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 484                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 485
 486             /* Calculate table index by multiplying r with table scale and truncate to integer */
 487             rt               = _mm_mul_pd(r00,vftabscale);
 488             vfitab           = _mm_cvttpd_epi32(rt);
 489 #ifdef __XOP__
 490             vfeps            = _mm_frcz_pd(rt);
 491 #else
 492             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 493 #endif
 494             twovfeps         = _mm_add_pd(vfeps,vfeps);
 495             vfitab           = _mm_slli_epi32(vfitab,3);
 496
 497             /* CUBIC SPLINE TABLE DISPERSION */
 498             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 499             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 500             GMX_MM_TRANSPOSE2_PD(Y,F);
 501             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 502             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 503             GMX_MM_TRANSPOSE2_PD(G,H);
 504             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 505             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 506             fvdw6            = _mm_mul_pd(c6_00,FF);
 507
 508             /* CUBIC SPLINE TABLE REPULSION */
 509             vfitab           = _mm_add_epi32(vfitab,ifour);
 510             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 511             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 512             GMX_MM_TRANSPOSE2_PD(Y,F);
 513             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 514             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 515             GMX_MM_TRANSPOSE2_PD(G,H);
 516             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 517             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 518             fvdw12           = _mm_mul_pd(c12_00,FF);
 519             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 520
 521             fscal            = fvdw;
 522
 523             /* Update vectorial force */
 524             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 525             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 526             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 527
 528             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 529                                                    _mm_mul_pd(dx00,fscal),
 530                                                    _mm_mul_pd(dy00,fscal),
 531                                                    _mm_mul_pd(dz00,fscal));
 532
 533             /* Inner loop uses 51 flops */
 534         }
 535
 536         if(jidx<j_index_end)
 537         {
 538
 539             jnrA             = jjnr[jidx];
 540             j_coord_offsetA  = DIM*jnrA;
 541
 542             /* load j atom coordinates */
 543             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 544                                               &jx0,&jy0,&jz0);
 545
 546             /* Calculate displacement vector */
 547             dx00             = _mm_sub_pd(ix0,jx0);
 548             dy00             = _mm_sub_pd(iy0,jy0);
 549             dz00             = _mm_sub_pd(iz0,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 553
 554             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 555
 556             /* Load parameters for j particles */
 557             vdwjidx0A        = 2*vdwtype[jnrA+0];
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r00              = _mm_mul_pd(rsq00,rinv00);
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 567
 568             /* Calculate table index by multiplying r with table scale and truncate to integer */
 569             rt               = _mm_mul_pd(r00,vftabscale);
 570             vfitab           = _mm_cvttpd_epi32(rt);
 571 #ifdef __XOP__
 572             vfeps            = _mm_frcz_pd(rt);
 573 #else
 574             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 575 #endif
 576             twovfeps         = _mm_add_pd(vfeps,vfeps);
 577             vfitab           = _mm_slli_epi32(vfitab,3);
 578
 579             /* CUBIC SPLINE TABLE DISPERSION */
 580             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 581             F                = _mm_setzero_pd();
 582             GMX_MM_TRANSPOSE2_PD(Y,F);
 583             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 584             H                = _mm_setzero_pd();
 585             GMX_MM_TRANSPOSE2_PD(G,H);
 586             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 587             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 588             fvdw6            = _mm_mul_pd(c6_00,FF);
 589
 590             /* CUBIC SPLINE TABLE REPULSION */
 591             vfitab           = _mm_add_epi32(vfitab,ifour);
 592             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 593             F                = _mm_setzero_pd();
 594             GMX_MM_TRANSPOSE2_PD(Y,F);
 595             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 596             H                = _mm_setzero_pd();
 597             GMX_MM_TRANSPOSE2_PD(G,H);
 598             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 599             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 600             fvdw12           = _mm_mul_pd(c12_00,FF);
 601             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 602
 603             fscal            = fvdw;
 604
 605             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 606
 607             /* Update vectorial force */
 608             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 609             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 610             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 611
 612             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 613                                                    _mm_mul_pd(dx00,fscal),
 614                                                    _mm_mul_pd(dy00,fscal),
 615                                                    _mm_mul_pd(dz00,fscal));
 616
 617             /* Inner loop uses 51 flops */
 618         }
 619
 620         /* End of innermost loop */
 621
 622         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 623                                               f+i_coord_offset,fshift+i_shift_offset);
 624
 625         /* Increment number of inner iterations */
 626         inneriter                  += j_index_end - j_index_start;
 627
 628         /* Outer loop uses 6 flops */
 629     }
 630
 631     /* Increment number of outer iterations */
 632     outeriter        += nri;
 633
 634     /* Update outer/inner flops */
 635
 636     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*51);
 637 }