src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecRF_VdwNone_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     __m128d          dummy_mask,cutoff_mask;
  87     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  88     __m128d          one     = _mm_set1_pd(1.0);
  89     __m128d          two     = _mm_set1_pd(2.0);
  90     x                = xx[0];
  91     f                = ff[0];
  92
  93     nri              = nlist->nri;
  94     iinr             = nlist->iinr;
  95     jindex           = nlist->jindex;
  96     jjnr             = nlist->jjnr;
  97     shiftidx         = nlist->shift;
  98     gid              = nlist->gid;
  99     shiftvec         = fr->shift_vec[0];
 100     fshift           = fr->fshift[0];
 101     facel            = _mm_set1_pd(fr->ic->epsfac);
 102     charge           = mdatoms->chargeA;
 103     krf              = _mm_set1_pd(fr->ic->k_rf);
 104     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 105     crf              = _mm_set1_pd(fr->ic->c_rf);
 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
 139         /* Reset potential sums */
 140         velecsum         = _mm_setzero_pd();
 141
 142         /* Start inner kernel loop */
 143         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 144         {
 145
 146             /* Get j neighbor index, and coordinate index */
 147             jnrA             = jjnr[jidx];
 148             jnrB             = jjnr[jidx+1];
 149             j_coord_offsetA  = DIM*jnrA;
 150             j_coord_offsetB  = DIM*jnrB;
 151
 152             /* load j atom coordinates */
 153             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 154                                               &jx0,&jy0,&jz0);
 155
 156             /* Calculate displacement vector */
 157             dx00             = _mm_sub_pd(ix0,jx0);
 158             dy00             = _mm_sub_pd(iy0,jy0);
 159             dz00             = _mm_sub_pd(iz0,jz0);
 160
 161             /* Calculate squared distance and things based on it */
 162             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 163
 164             rinv00           = sse2_invsqrt_d(rsq00);
 165
 166             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 167
 168             /* Load parameters for j particles */
 169             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 170
 171             /**************************
 172              * CALCULATE INTERACTIONS *
 173              **************************/
 174
 175             /* Compute parameters for interactions between i and j atoms */
 176             qq00             = _mm_mul_pd(iq0,jq0);
 177
 178             /* REACTION-FIELD ELECTROSTATICS */
 179             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 180             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 181
 182             /* Update potential sum for this i atom from the interaction with this j atom. */
 183             velecsum         = _mm_add_pd(velecsum,velec);
 184
 185             fscal            = felec;
 186
 187             /* Calculate temporary vectorial force */
 188             tx               = _mm_mul_pd(fscal,dx00);
 189             ty               = _mm_mul_pd(fscal,dy00);
 190             tz               = _mm_mul_pd(fscal,dz00);
 191
 192             /* Update vectorial force */
 193             fix0             = _mm_add_pd(fix0,tx);
 194             fiy0             = _mm_add_pd(fiy0,ty);
 195             fiz0             = _mm_add_pd(fiz0,tz);
 196
 197             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 198
 199             /* Inner loop uses 32 flops */
 200         }
 201
 202         if(jidx<j_index_end)
 203         {
 204
 205             jnrA             = jjnr[jidx];
 206             j_coord_offsetA  = DIM*jnrA;
 207
 208             /* load j atom coordinates */
 209             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 210                                               &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx00             = _mm_sub_pd(ix0,jx0);
 214             dy00             = _mm_sub_pd(iy0,jy0);
 215             dz00             = _mm_sub_pd(iz0,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 219
 220             rinv00           = sse2_invsqrt_d(rsq00);
 221
 222             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 223
 224             /* Load parameters for j particles */
 225             jq0              = _mm_load_sd(charge+jnrA+0);
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm_mul_pd(iq0,jq0);
 233
 234             /* REACTION-FIELD ELECTROSTATICS */
 235             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 236             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 237
 238             /* Update potential sum for this i atom from the interaction with this j atom. */
 239             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 240             velecsum         = _mm_add_pd(velecsum,velec);
 241
 242             fscal            = felec;
 243
 244             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 245
 246             /* Calculate temporary vectorial force */
 247             tx               = _mm_mul_pd(fscal,dx00);
 248             ty               = _mm_mul_pd(fscal,dy00);
 249             tz               = _mm_mul_pd(fscal,dz00);
 250
 251             /* Update vectorial force */
 252             fix0             = _mm_add_pd(fix0,tx);
 253             fiy0             = _mm_add_pd(fiy0,ty);
 254             fiz0             = _mm_add_pd(fiz0,tz);
 255
 256             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 257
 258             /* Inner loop uses 32 flops */
 259         }
 260
 261         /* End of innermost loop */
 262
 263         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 264                                               f+i_coord_offset,fshift+i_shift_offset);
 265
 266         ggid                        = gid[iidx];
 267         /* Update potential energies */
 268         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 269
 270         /* Increment number of inner iterations */
 271         inneriter                  += j_index_end - j_index_start;
 272
 273         /* Outer loop uses 8 flops */
 274     }
 275
 276     /* Increment number of outer iterations */
 277     outeriter        += nri;
 278
 279     /* Update outer/inner flops */
 280
 281     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 282 }
 283 /*
 284  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_double
 285  * Electrostatics interaction: ReactionField
 286  * VdW interaction:            None
 287  * Geometry:                   Particle-Particle
 288  * Calculate force/pot:        Force
 289  */
 290 void
 291 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_double
 292                     (t_nblist                    * gmx_restrict       nlist,
 293                      rvec                        * gmx_restrict          xx,
 294                      rvec                        * gmx_restrict          ff,
 295                      struct t_forcerec           * gmx_restrict          fr,
 296                      t_mdatoms                   * gmx_restrict     mdatoms,
 297                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 298                      t_nrnb                      * gmx_restrict        nrnb)
 299 {
 300     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 301      * just 0 for non-waters.
 302      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 303      * jnr indices corresponding to data put in the four positions in the SIMD register.
 304      */
 305     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 306     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 307     int              jnrA,jnrB;
 308     int              j_coord_offsetA,j_coord_offsetB;
 309     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 310     real             rcutoff_scalar;
 311     real             *shiftvec,*fshift,*x,*f;
 312     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 313     int              vdwioffset0;
 314     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 315     int              vdwjidx0A,vdwjidx0B;
 316     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 317     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 318     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 319     real             *charge;
 320     __m128d          dummy_mask,cutoff_mask;
 321     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 322     __m128d          one     = _mm_set1_pd(1.0);
 323     __m128d          two     = _mm_set1_pd(2.0);
 324     x                = xx[0];
 325     f                = ff[0];
 326
 327     nri              = nlist->nri;
 328     iinr             = nlist->iinr;
 329     jindex           = nlist->jindex;
 330     jjnr             = nlist->jjnr;
 331     shiftidx         = nlist->shift;
 332     gid              = nlist->gid;
 333     shiftvec         = fr->shift_vec[0];
 334     fshift           = fr->fshift[0];
 335     facel            = _mm_set1_pd(fr->ic->epsfac);
 336     charge           = mdatoms->chargeA;
 337     krf              = _mm_set1_pd(fr->ic->k_rf);
 338     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 339     crf              = _mm_set1_pd(fr->ic->c_rf);
 340
 341     /* Avoid stupid compiler warnings */
 342     jnrA = jnrB = 0;
 343     j_coord_offsetA = 0;
 344     j_coord_offsetB = 0;
 345
 346     outeriter        = 0;
 347     inneriter        = 0;
 348
 349     /* Start outer loop over neighborlists */
 350     for(iidx=0; iidx<nri; iidx++)
 351     {
 352         /* Load shift vector for this list */
 353         i_shift_offset   = DIM*shiftidx[iidx];
 354
 355         /* Load limits for loop over neighbors */
 356         j_index_start    = jindex[iidx];
 357         j_index_end      = jindex[iidx+1];
 358
 359         /* Get outer coordinate index */
 360         inr              = iinr[iidx];
 361         i_coord_offset   = DIM*inr;
 362
 363         /* Load i particle coords and add shift vector */
 364         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 365
 366         fix0             = _mm_setzero_pd();
 367         fiy0             = _mm_setzero_pd();
 368         fiz0             = _mm_setzero_pd();
 369
 370         /* Load parameters for i particles */
 371         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 372
 373         /* Start inner kernel loop */
 374         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 375         {
 376
 377             /* Get j neighbor index, and coordinate index */
 378             jnrA             = jjnr[jidx];
 379             jnrB             = jjnr[jidx+1];
 380             j_coord_offsetA  = DIM*jnrA;
 381             j_coord_offsetB  = DIM*jnrB;
 382
 383             /* load j atom coordinates */
 384             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 385                                               &jx0,&jy0,&jz0);
 386
 387             /* Calculate displacement vector */
 388             dx00             = _mm_sub_pd(ix0,jx0);
 389             dy00             = _mm_sub_pd(iy0,jy0);
 390             dz00             = _mm_sub_pd(iz0,jz0);
 391
 392             /* Calculate squared distance and things based on it */
 393             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 394
 395             rinv00           = sse2_invsqrt_d(rsq00);
 396
 397             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 398
 399             /* Load parameters for j particles */
 400             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 401
 402             /**************************
 403              * CALCULATE INTERACTIONS *
 404              **************************/
 405
 406             /* Compute parameters for interactions between i and j atoms */
 407             qq00             = _mm_mul_pd(iq0,jq0);
 408
 409             /* REACTION-FIELD ELECTROSTATICS */
 410             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 411
 412             fscal            = felec;
 413
 414             /* Calculate temporary vectorial force */
 415             tx               = _mm_mul_pd(fscal,dx00);
 416             ty               = _mm_mul_pd(fscal,dy00);
 417             tz               = _mm_mul_pd(fscal,dz00);
 418
 419             /* Update vectorial force */
 420             fix0             = _mm_add_pd(fix0,tx);
 421             fiy0             = _mm_add_pd(fiy0,ty);
 422             fiz0             = _mm_add_pd(fiz0,tz);
 423
 424             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 425
 426             /* Inner loop uses 27 flops */
 427         }
 428
 429         if(jidx<j_index_end)
 430         {
 431
 432             jnrA             = jjnr[jidx];
 433             j_coord_offsetA  = DIM*jnrA;
 434
 435             /* load j atom coordinates */
 436             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 437                                               &jx0,&jy0,&jz0);
 438
 439             /* Calculate displacement vector */
 440             dx00             = _mm_sub_pd(ix0,jx0);
 441             dy00             = _mm_sub_pd(iy0,jy0);
 442             dz00             = _mm_sub_pd(iz0,jz0);
 443
 444             /* Calculate squared distance and things based on it */
 445             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 446
 447             rinv00           = sse2_invsqrt_d(rsq00);
 448
 449             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 450
 451             /* Load parameters for j particles */
 452             jq0              = _mm_load_sd(charge+jnrA+0);
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             /* Compute parameters for interactions between i and j atoms */
 459             qq00             = _mm_mul_pd(iq0,jq0);
 460
 461             /* REACTION-FIELD ELECTROSTATICS */
 462             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 463
 464             fscal            = felec;
 465
 466             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 467
 468             /* Calculate temporary vectorial force */
 469             tx               = _mm_mul_pd(fscal,dx00);
 470             ty               = _mm_mul_pd(fscal,dy00);
 471             tz               = _mm_mul_pd(fscal,dz00);
 472
 473             /* Update vectorial force */
 474             fix0             = _mm_add_pd(fix0,tx);
 475             fiy0             = _mm_add_pd(fiy0,ty);
 476             fiz0             = _mm_add_pd(fiz0,tz);
 477
 478             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 479
 480             /* Inner loop uses 27 flops */
 481         }
 482
 483         /* End of innermost loop */
 484
 485         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 486                                               f+i_coord_offset,fshift+i_shift_offset);
 487
 488         /* Increment number of inner iterations */
 489         inneriter                  += j_index_end - j_index_start;
 490
 491         /* Outer loop uses 7 flops */
 492     }
 493
 494     /* Increment number of outer iterations */
 495     outeriter        += nri;
 496
 497     /* Update outer/inner flops */
 498
 499     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 500 }