src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_VdwLJSw_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_VdwLJSw_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: None
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_VdwLJSw_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     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  78     real             rswitch_scalar,d_scalar;
  79     __m128d          dummy_mask,cutoff_mask;
  80     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  81     __m128d          one     = _mm_set1_pd(1.0);
  82     __m128d          two     = _mm_set1_pd(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     nvdwtype         = fr->ntype;
  95     vdwparam         = fr->nbfp;
  96     vdwtype          = mdatoms->typeA;
  97
  98     rcutoff_scalar   = fr->rvdw;
  99     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 100     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 101
 102     rswitch_scalar   = fr->rvdw_switch;
 103     rswitch          = _mm_set1_pd(rswitch_scalar);
 104     /* Setup switch parameters */
 105     d_scalar         = rcutoff_scalar-rswitch_scalar;
 106     d                = _mm_set1_pd(d_scalar);
 107     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 108     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 109     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 110     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 111     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 112     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 113
 114     /* Avoid stupid compiler warnings */
 115     jnrA = jnrB = 0;
 116     j_coord_offsetA = 0;
 117     j_coord_offsetB = 0;
 118
 119     outeriter        = 0;
 120     inneriter        = 0;
 121
 122     /* Start outer loop over neighborlists */
 123     for(iidx=0; iidx<nri; iidx++)
 124     {
 125         /* Load shift vector for this list */
 126         i_shift_offset   = DIM*shiftidx[iidx];
 127
 128         /* Load limits for loop over neighbors */
 129         j_index_start    = jindex[iidx];
 130         j_index_end      = jindex[iidx+1];
 131
 132         /* Get outer coordinate index */
 133         inr              = iinr[iidx];
 134         i_coord_offset   = DIM*inr;
 135
 136         /* Load i particle coords and add shift vector */
 137         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 138
 139         fix0             = _mm_setzero_pd();
 140         fiy0             = _mm_setzero_pd();
 141         fiz0             = _mm_setzero_pd();
 142
 143         /* Load parameters for i particles */
 144         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 145
 146         /* Reset potential sums */
 147         vvdwsum          = _mm_setzero_pd();
 148
 149         /* Start inner kernel loop */
 150         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 151         {
 152
 153             /* Get j neighbor index, and coordinate index */
 154             jnrA             = jjnr[jidx];
 155             jnrB             = jjnr[jidx+1];
 156             j_coord_offsetA  = DIM*jnrA;
 157             j_coord_offsetB  = DIM*jnrB;
 158
 159             /* load j atom coordinates */
 160             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 161                                               &jx0,&jy0,&jz0);
 162
 163             /* Calculate displacement vector */
 164             dx00             = _mm_sub_pd(ix0,jx0);
 165             dy00             = _mm_sub_pd(iy0,jy0);
 166             dz00             = _mm_sub_pd(iz0,jz0);
 167
 168             /* Calculate squared distance and things based on it */
 169             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 170
 171             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 172
 173             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 174
 175             /* Load parameters for j particles */
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             if (gmx_mm_any_lt(rsq00,rcutoff2))
 184             {
 185
 186             r00              = _mm_mul_pd(rsq00,rinv00);
 187
 188             /* Compute parameters for interactions between i and j atoms */
 189             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 190                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 191
 192             /* LENNARD-JONES DISPERSION/REPULSION */
 193
 194             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 195             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 196             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 197             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 198             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 199
 200             d                = _mm_sub_pd(r00,rswitch);
 201             d                = _mm_max_pd(d,_mm_setzero_pd());
 202             d2               = _mm_mul_pd(d,d);
 203             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 204
 205             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 206
 207             /* Evaluate switch function */
 208             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 209             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 210             vvdw             = _mm_mul_pd(vvdw,sw);
 211             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 215             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 216
 217             fscal            = fvdw;
 218
 219             fscal            = _mm_and_pd(fscal,cutoff_mask);
 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             }
 232
 233             /* Inner loop uses 62 flops */
 234         }
 235
 236         if(jidx<j_index_end)
 237         {
 238
 239             jnrA             = jjnr[jidx];
 240             j_coord_offsetA  = DIM*jnrA;
 241
 242             /* load j atom coordinates */
 243             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 244                                               &jx0,&jy0,&jz0);
 245
 246             /* Calculate displacement vector */
 247             dx00             = _mm_sub_pd(ix0,jx0);
 248             dy00             = _mm_sub_pd(iy0,jy0);
 249             dz00             = _mm_sub_pd(iz0,jz0);
 250
 251             /* Calculate squared distance and things based on it */
 252             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 253
 254             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 255
 256             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 257
 258             /* Load parameters for j particles */
 259             vdwjidx0A        = 2*vdwtype[jnrA+0];
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             if (gmx_mm_any_lt(rsq00,rcutoff2))
 266             {
 267
 268             r00              = _mm_mul_pd(rsq00,rinv00);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 272
 273             /* LENNARD-JONES DISPERSION/REPULSION */
 274
 275             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 276             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 277             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 278             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 279             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 280
 281             d                = _mm_sub_pd(r00,rswitch);
 282             d                = _mm_max_pd(d,_mm_setzero_pd());
 283             d2               = _mm_mul_pd(d,d);
 284             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 285
 286             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 287
 288             /* Evaluate switch function */
 289             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 290             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 291             vvdw             = _mm_mul_pd(vvdw,sw);
 292             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 296             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 297             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 298
 299             fscal            = fvdw;
 300
 301             fscal            = _mm_and_pd(fscal,cutoff_mask);
 302
 303             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 304
 305             /* Update vectorial force */
 306             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 307             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 308             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 309
 310             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 311                                                    _mm_mul_pd(dx00,fscal),
 312                                                    _mm_mul_pd(dy00,fscal),
 313                                                    _mm_mul_pd(dz00,fscal));
 314
 315             }
 316
 317             /* Inner loop uses 62 flops */
 318         }
 319
 320         /* End of innermost loop */
 321
 322         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 323                                               f+i_coord_offset,fshift+i_shift_offset);
 324
 325         ggid                        = gid[iidx];
 326         /* Update potential energies */
 327         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 328
 329         /* Increment number of inner iterations */
 330         inneriter                  += j_index_end - j_index_start;
 331
 332         /* Outer loop uses 7 flops */
 333     }
 334
 335     /* Increment number of outer iterations */
 336     outeriter        += nri;
 337
 338     /* Update outer/inner flops */
 339
 340     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*62);
 341 }
 342 /*
 343  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 344  * Electrostatics interaction: None
 345  * VdW interaction:            LennardJones
 346  * Geometry:                   Particle-Particle
 347  * Calculate force/pot:        Force
 348  */
 349 void
 350 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_avx_128_fma_double
 351                     (t_nblist * gmx_restrict                nlist,
 352                      rvec * gmx_restrict                    xx,
 353                      rvec * gmx_restrict                    ff,
 354                      t_forcerec * gmx_restrict              fr,
 355                      t_mdatoms * gmx_restrict               mdatoms,
 356                      nb_kernel_data_t * gmx_restrict        kernel_data,
 357                      t_nrnb * gmx_restrict                  nrnb)
 358 {
 359     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 360      * just 0 for non-waters.
 361      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 362      * jnr indices corresponding to data put in the four positions in the SIMD register.
 363      */
 364     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 365     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 366     int              jnrA,jnrB;
 367     int              j_coord_offsetA,j_coord_offsetB;
 368     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 369     real             rcutoff_scalar;
 370     real             *shiftvec,*fshift,*x,*f;
 371     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 372     int              vdwioffset0;
 373     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 374     int              vdwjidx0A,vdwjidx0B;
 375     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 376     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 377     int              nvdwtype;
 378     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 379     int              *vdwtype;
 380     real             *vdwparam;
 381     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 382     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 383     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 384     real             rswitch_scalar,d_scalar;
 385     __m128d          dummy_mask,cutoff_mask;
 386     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 387     __m128d          one     = _mm_set1_pd(1.0);
 388     __m128d          two     = _mm_set1_pd(2.0);
 389     x                = xx[0];
 390     f                = ff[0];
 391
 392     nri              = nlist->nri;
 393     iinr             = nlist->iinr;
 394     jindex           = nlist->jindex;
 395     jjnr             = nlist->jjnr;
 396     shiftidx         = nlist->shift;
 397     gid              = nlist->gid;
 398     shiftvec         = fr->shift_vec[0];
 399     fshift           = fr->fshift[0];
 400     nvdwtype         = fr->ntype;
 401     vdwparam         = fr->nbfp;
 402     vdwtype          = mdatoms->typeA;
 403
 404     rcutoff_scalar   = fr->rvdw;
 405     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 406     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 407
 408     rswitch_scalar   = fr->rvdw_switch;
 409     rswitch          = _mm_set1_pd(rswitch_scalar);
 410     /* Setup switch parameters */
 411     d_scalar         = rcutoff_scalar-rswitch_scalar;
 412     d                = _mm_set1_pd(d_scalar);
 413     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 414     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 415     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 416     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 417     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 418     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 419
 420     /* Avoid stupid compiler warnings */
 421     jnrA = jnrB = 0;
 422     j_coord_offsetA = 0;
 423     j_coord_offsetB = 0;
 424
 425     outeriter        = 0;
 426     inneriter        = 0;
 427
 428     /* Start outer loop over neighborlists */
 429     for(iidx=0; iidx<nri; iidx++)
 430     {
 431         /* Load shift vector for this list */
 432         i_shift_offset   = DIM*shiftidx[iidx];
 433
 434         /* Load limits for loop over neighbors */
 435         j_index_start    = jindex[iidx];
 436         j_index_end      = jindex[iidx+1];
 437
 438         /* Get outer coordinate index */
 439         inr              = iinr[iidx];
 440         i_coord_offset   = DIM*inr;
 441
 442         /* Load i particle coords and add shift vector */
 443         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 444
 445         fix0             = _mm_setzero_pd();
 446         fiy0             = _mm_setzero_pd();
 447         fiz0             = _mm_setzero_pd();
 448
 449         /* Load parameters for i particles */
 450         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 451
 452         /* Start inner kernel loop */
 453         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 454         {
 455
 456             /* Get j neighbor index, and coordinate index */
 457             jnrA             = jjnr[jidx];
 458             jnrB             = jjnr[jidx+1];
 459             j_coord_offsetA  = DIM*jnrA;
 460             j_coord_offsetB  = DIM*jnrB;
 461
 462             /* load j atom coordinates */
 463             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 464                                               &jx0,&jy0,&jz0);
 465
 466             /* Calculate displacement vector */
 467             dx00             = _mm_sub_pd(ix0,jx0);
 468             dy00             = _mm_sub_pd(iy0,jy0);
 469             dz00             = _mm_sub_pd(iz0,jz0);
 470
 471             /* Calculate squared distance and things based on it */
 472             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 473
 474             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 475
 476             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 477
 478             /* Load parameters for j particles */
 479             vdwjidx0A        = 2*vdwtype[jnrA+0];
 480             vdwjidx0B        = 2*vdwtype[jnrB+0];
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             if (gmx_mm_any_lt(rsq00,rcutoff2))
 487             {
 488
 489             r00              = _mm_mul_pd(rsq00,rinv00);
 490
 491             /* Compute parameters for interactions between i and j atoms */
 492             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 493                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 494
 495             /* LENNARD-JONES DISPERSION/REPULSION */
 496
 497             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 498             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 499             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 500             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 501             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 502
 503             d                = _mm_sub_pd(r00,rswitch);
 504             d                = _mm_max_pd(d,_mm_setzero_pd());
 505             d2               = _mm_mul_pd(d,d);
 506             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 507
 508             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 509
 510             /* Evaluate switch function */
 511             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 512             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 513             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 514
 515             fscal            = fvdw;
 516
 517             fscal            = _mm_and_pd(fscal,cutoff_mask);
 518
 519             /* Update vectorial force */
 520             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 521             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 522             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 523
 524             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 525                                                    _mm_mul_pd(dx00,fscal),
 526                                                    _mm_mul_pd(dy00,fscal),
 527                                                    _mm_mul_pd(dz00,fscal));
 528
 529             }
 530
 531             /* Inner loop uses 59 flops */
 532         }
 533
 534         if(jidx<j_index_end)
 535         {
 536
 537             jnrA             = jjnr[jidx];
 538             j_coord_offsetA  = DIM*jnrA;
 539
 540             /* load j atom coordinates */
 541             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 542                                               &jx0,&jy0,&jz0);
 543
 544             /* Calculate displacement vector */
 545             dx00             = _mm_sub_pd(ix0,jx0);
 546             dy00             = _mm_sub_pd(iy0,jy0);
 547             dz00             = _mm_sub_pd(iz0,jz0);
 548
 549             /* Calculate squared distance and things based on it */
 550             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 551
 552             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 553
 554             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 555
 556             /* Load parameters for j particles */
 557             vdwjidx0A        = 2*vdwtype[jnrA+0];
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             if (gmx_mm_any_lt(rsq00,rcutoff2))
 564             {
 565
 566             r00              = _mm_mul_pd(rsq00,rinv00);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 570
 571             /* LENNARD-JONES DISPERSION/REPULSION */
 572
 573             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 574             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 575             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 576             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 577             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 578
 579             d                = _mm_sub_pd(r00,rswitch);
 580             d                = _mm_max_pd(d,_mm_setzero_pd());
 581             d2               = _mm_mul_pd(d,d);
 582             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 583
 584             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 585
 586             /* Evaluate switch function */
 587             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 588             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 589             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 590
 591             fscal            = fvdw;
 592
 593             fscal            = _mm_and_pd(fscal,cutoff_mask);
 594
 595             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 596
 597             /* Update vectorial force */
 598             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 599             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 600             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 601
 602             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 603                                                    _mm_mul_pd(dx00,fscal),
 604                                                    _mm_mul_pd(dy00,fscal),
 605                                                    _mm_mul_pd(dz00,fscal));
 606
 607             }
 608
 609             /* Inner loop uses 59 flops */
 610         }
 611
 612         /* End of innermost loop */
 613
 614         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 615                                               f+i_coord_offset,fshift+i_shift_offset);
 616
 617         /* Increment number of inner iterations */
 618         inneriter                  += j_index_end - j_index_start;
 619
 620         /* Outer loop uses 6 flops */
 621     }
 622
 623     /* Increment number of outer iterations */
 624     outeriter        += nri;
 625
 626     /* Update outer/inner flops */
 627
 628     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*59);
 629 }