src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecNone_VdwLJ_GeomP1P1_sse2_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_double.h"
  34 #include "kernelutil_x86_sse2_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_VF_sse2_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_VdwLJ_GeomP1P1_VF_sse2_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          dummy_mask,cutoff_mask;
  78     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  79     __m128d          one     = _mm_set1_pd(1.0);
  80     __m128d          two     = _mm_set1_pd(2.0);
  81     x                = xx[0];
  82     f                = ff[0];
  83
  84     nri              = nlist->nri;
  85     iinr             = nlist->iinr;
  86     jindex           = nlist->jindex;
  87     jjnr             = nlist->jjnr;
  88     shiftidx         = nlist->shift;
  89     gid              = nlist->gid;
  90     shiftvec         = fr->shift_vec[0];
  91     fshift           = fr->fshift[0];
  92     nvdwtype         = fr->ntype;
  93     vdwparam         = fr->nbfp;
  94     vdwtype          = mdatoms->typeA;
  95
  96     /* Avoid stupid compiler warnings */
  97     jnrA = jnrB = 0;
  98     j_coord_offsetA = 0;
  99     j_coord_offsetB = 0;
 100
 101     outeriter        = 0;
 102     inneriter        = 0;
 103
 104     /* Start outer loop over neighborlists */
 105     for(iidx=0; iidx<nri; iidx++)
 106     {
 107         /* Load shift vector for this list */
 108         i_shift_offset   = DIM*shiftidx[iidx];
 109
 110         /* Load limits for loop over neighbors */
 111         j_index_start    = jindex[iidx];
 112         j_index_end      = jindex[iidx+1];
 113
 114         /* Get outer coordinate index */
 115         inr              = iinr[iidx];
 116         i_coord_offset   = DIM*inr;
 117
 118         /* Load i particle coords and add shift vector */
 119         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 120
 121         fix0             = _mm_setzero_pd();
 122         fiy0             = _mm_setzero_pd();
 123         fiz0             = _mm_setzero_pd();
 124
 125         /* Load parameters for i particles */
 126         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 127
 128         /* Reset potential sums */
 129         vvdwsum          = _mm_setzero_pd();
 130
 131         /* Start inner kernel loop */
 132         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 133         {
 134
 135             /* Get j neighbor index, and coordinate index */
 136             jnrA             = jjnr[jidx];
 137             jnrB             = jjnr[jidx+1];
 138             j_coord_offsetA  = DIM*jnrA;
 139             j_coord_offsetB  = DIM*jnrB;
 140
 141             /* load j atom coordinates */
 142             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 143                                               &jx0,&jy0,&jz0);
 144
 145             /* Calculate displacement vector */
 146             dx00             = _mm_sub_pd(ix0,jx0);
 147             dy00             = _mm_sub_pd(iy0,jy0);
 148             dz00             = _mm_sub_pd(iz0,jz0);
 149
 150             /* Calculate squared distance and things based on it */
 151             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 152
 153             rinvsq00         = gmx_mm_inv_pd(rsq00);
 154
 155             /* Load parameters for j particles */
 156             vdwjidx0A        = 2*vdwtype[jnrA+0];
 157             vdwjidx0B        = 2*vdwtype[jnrB+0];
 158
 159             /**************************
 160              * CALCULATE INTERACTIONS *
 161              **************************/
 162
 163             /* Compute parameters for interactions between i and j atoms */
 164             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 165                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 166
 167             /* LENNARD-JONES DISPERSION/REPULSION */
 168
 169             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 170             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 171             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 172             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 173             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 174
 175             /* Update potential sum for this i atom from the interaction with this j atom. */
 176             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 177
 178             fscal            = fvdw;
 179
 180             /* Calculate temporary vectorial force */
 181             tx               = _mm_mul_pd(fscal,dx00);
 182             ty               = _mm_mul_pd(fscal,dy00);
 183             tz               = _mm_mul_pd(fscal,dz00);
 184
 185             /* Update vectorial force */
 186             fix0             = _mm_add_pd(fix0,tx);
 187             fiy0             = _mm_add_pd(fiy0,ty);
 188             fiz0             = _mm_add_pd(fiz0,tz);
 189
 190             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 191
 192             /* Inner loop uses 32 flops */
 193         }
 194
 195         if(jidx<j_index_end)
 196         {
 197
 198             jnrA             = jjnr[jidx];
 199             j_coord_offsetA  = DIM*jnrA;
 200
 201             /* load j atom coordinates */
 202             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_pd(ix0,jx0);
 207             dy00             = _mm_sub_pd(iy0,jy0);
 208             dz00             = _mm_sub_pd(iz0,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 212
 213             rinvsq00         = gmx_mm_inv_pd(rsq00);
 214
 215             /* Load parameters for j particles */
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             /* Compute parameters for interactions between i and j atoms */
 223             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 224
 225             /* LENNARD-JONES DISPERSION/REPULSION */
 226
 227             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 228             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 229             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 230             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 231             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 235             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 236
 237             fscal            = fvdw;
 238
 239             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 240
 241             /* Calculate temporary vectorial force */
 242             tx               = _mm_mul_pd(fscal,dx00);
 243             ty               = _mm_mul_pd(fscal,dy00);
 244             tz               = _mm_mul_pd(fscal,dz00);
 245
 246             /* Update vectorial force */
 247             fix0             = _mm_add_pd(fix0,tx);
 248             fiy0             = _mm_add_pd(fiy0,ty);
 249             fiz0             = _mm_add_pd(fiz0,tz);
 250
 251             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 252
 253             /* Inner loop uses 32 flops */
 254         }
 255
 256         /* End of innermost loop */
 257
 258         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 259                                               f+i_coord_offset,fshift+i_shift_offset);
 260
 261         ggid                        = gid[iidx];
 262         /* Update potential energies */
 263         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 264
 265         /* Increment number of inner iterations */
 266         inneriter                  += j_index_end - j_index_start;
 267
 268         /* Outer loop uses 7 flops */
 269     }
 270
 271     /* Increment number of outer iterations */
 272     outeriter        += nri;
 273
 274     /* Update outer/inner flops */
 275
 276     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*32);
 277 }
 278 /*
 279  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_sse2_double
 280  * Electrostatics interaction: None
 281  * VdW interaction:            LennardJones
 282  * Geometry:                   Particle-Particle
 283  * Calculate force/pot:        Force
 284  */
 285 void
 286 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_sse2_double
 287                     (t_nblist * gmx_restrict                nlist,
 288                      rvec * gmx_restrict                    xx,
 289                      rvec * gmx_restrict                    ff,
 290                      t_forcerec * gmx_restrict              fr,
 291                      t_mdatoms * gmx_restrict               mdatoms,
 292                      nb_kernel_data_t * gmx_restrict        kernel_data,
 293                      t_nrnb * gmx_restrict                  nrnb)
 294 {
 295     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 296      * just 0 for non-waters.
 297      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 298      * jnr indices corresponding to data put in the four positions in the SIMD register.
 299      */
 300     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 301     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 302     int              jnrA,jnrB;
 303     int              j_coord_offsetA,j_coord_offsetB;
 304     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 305     real             rcutoff_scalar;
 306     real             *shiftvec,*fshift,*x,*f;
 307     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 308     int              vdwioffset0;
 309     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 310     int              vdwjidx0A,vdwjidx0B;
 311     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 312     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 313     int              nvdwtype;
 314     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 315     int              *vdwtype;
 316     real             *vdwparam;
 317     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 318     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 319     __m128d          dummy_mask,cutoff_mask;
 320     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 321     __m128d          one     = _mm_set1_pd(1.0);
 322     __m128d          two     = _mm_set1_pd(2.0);
 323     x                = xx[0];
 324     f                = ff[0];
 325
 326     nri              = nlist->nri;
 327     iinr             = nlist->iinr;
 328     jindex           = nlist->jindex;
 329     jjnr             = nlist->jjnr;
 330     shiftidx         = nlist->shift;
 331     gid              = nlist->gid;
 332     shiftvec         = fr->shift_vec[0];
 333     fshift           = fr->fshift[0];
 334     nvdwtype         = fr->ntype;
 335     vdwparam         = fr->nbfp;
 336     vdwtype          = mdatoms->typeA;
 337
 338     /* Avoid stupid compiler warnings */
 339     jnrA = jnrB = 0;
 340     j_coord_offsetA = 0;
 341     j_coord_offsetB = 0;
 342
 343     outeriter        = 0;
 344     inneriter        = 0;
 345
 346     /* Start outer loop over neighborlists */
 347     for(iidx=0; iidx<nri; iidx++)
 348     {
 349         /* Load shift vector for this list */
 350         i_shift_offset   = DIM*shiftidx[iidx];
 351
 352         /* Load limits for loop over neighbors */
 353         j_index_start    = jindex[iidx];
 354         j_index_end      = jindex[iidx+1];
 355
 356         /* Get outer coordinate index */
 357         inr              = iinr[iidx];
 358         i_coord_offset   = DIM*inr;
 359
 360         /* Load i particle coords and add shift vector */
 361         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 362
 363         fix0             = _mm_setzero_pd();
 364         fiy0             = _mm_setzero_pd();
 365         fiz0             = _mm_setzero_pd();
 366
 367         /* Load parameters for i particles */
 368         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 369
 370         /* Start inner kernel loop */
 371         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 372         {
 373
 374             /* Get j neighbor index, and coordinate index */
 375             jnrA             = jjnr[jidx];
 376             jnrB             = jjnr[jidx+1];
 377             j_coord_offsetA  = DIM*jnrA;
 378             j_coord_offsetB  = DIM*jnrB;
 379
 380             /* load j atom coordinates */
 381             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 382                                               &jx0,&jy0,&jz0);
 383
 384             /* Calculate displacement vector */
 385             dx00             = _mm_sub_pd(ix0,jx0);
 386             dy00             = _mm_sub_pd(iy0,jy0);
 387             dz00             = _mm_sub_pd(iz0,jz0);
 388
 389             /* Calculate squared distance and things based on it */
 390             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 391
 392             rinvsq00         = gmx_mm_inv_pd(rsq00);
 393
 394             /* Load parameters for j particles */
 395             vdwjidx0A        = 2*vdwtype[jnrA+0];
 396             vdwjidx0B        = 2*vdwtype[jnrB+0];
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             /* Compute parameters for interactions between i and j atoms */
 403             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 404                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 405
 406             /* LENNARD-JONES DISPERSION/REPULSION */
 407
 408             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 409             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 410
 411             fscal            = fvdw;
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm_mul_pd(fscal,dx00);
 415             ty               = _mm_mul_pd(fscal,dy00);
 416             tz               = _mm_mul_pd(fscal,dz00);
 417
 418             /* Update vectorial force */
 419             fix0             = _mm_add_pd(fix0,tx);
 420             fiy0             = _mm_add_pd(fiy0,ty);
 421             fiz0             = _mm_add_pd(fiz0,tz);
 422
 423             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 424
 425             /* Inner loop uses 27 flops */
 426         }
 427
 428         if(jidx<j_index_end)
 429         {
 430
 431             jnrA             = jjnr[jidx];
 432             j_coord_offsetA  = DIM*jnrA;
 433
 434             /* load j atom coordinates */
 435             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 436                                               &jx0,&jy0,&jz0);
 437
 438             /* Calculate displacement vector */
 439             dx00             = _mm_sub_pd(ix0,jx0);
 440             dy00             = _mm_sub_pd(iy0,jy0);
 441             dz00             = _mm_sub_pd(iz0,jz0);
 442
 443             /* Calculate squared distance and things based on it */
 444             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 445
 446             rinvsq00         = gmx_mm_inv_pd(rsq00);
 447
 448             /* Load parameters for j particles */
 449             vdwjidx0A        = 2*vdwtype[jnrA+0];
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 457
 458             /* LENNARD-JONES DISPERSION/REPULSION */
 459
 460             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 461             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 462
 463             fscal            = fvdw;
 464
 465             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 466
 467             /* Calculate temporary vectorial force */
 468             tx               = _mm_mul_pd(fscal,dx00);
 469             ty               = _mm_mul_pd(fscal,dy00);
 470             tz               = _mm_mul_pd(fscal,dz00);
 471
 472             /* Update vectorial force */
 473             fix0             = _mm_add_pd(fix0,tx);
 474             fiy0             = _mm_add_pd(fiy0,ty);
 475             fiz0             = _mm_add_pd(fiz0,tz);
 476
 477             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 478
 479             /* Inner loop uses 27 flops */
 480         }
 481
 482         /* End of innermost loop */
 483
 484         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 485                                               f+i_coord_offset,fshift+i_shift_offset);
 486
 487         /* Increment number of inner iterations */
 488         inneriter                  += j_index_end - j_index_start;
 489
 490         /* Outer loop uses 6 flops */
 491     }
 492
 493     /* Increment number of outer iterations */
 494     outeriter        += nri;
 495
 496     /* Update outer/inner flops */
 497
 498     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*27);
 499 }