src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwNone_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 avx_128_fma_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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     __m128d          dummy_mask,cutoff_mask;
  90     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  91     __m128d          one     = _mm_set1_pd(1.0);
  92     __m128d          two     = _mm_set1_pd(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_pd(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 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           = gmx_mm_invsqrt_pd(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             /* COULOMB ELECTROSTATICS */
 179             velec            = _mm_mul_pd(qq00,rinv00);
 180             felec            = _mm_mul_pd(velec,rinvsq00);
 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             /* Update vectorial force */
 188             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 189             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 190             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 191
 192             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 193                                                    _mm_mul_pd(dx00,fscal),
 194                                                    _mm_mul_pd(dy00,fscal),
 195                                                    _mm_mul_pd(dz00,fscal));
 196
 197             /* Inner loop uses 31 flops */
 198         }
 199
 200         if(jidx<j_index_end)
 201         {
 202
 203             jnrA             = jjnr[jidx];
 204             j_coord_offsetA  = DIM*jnrA;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_pd(ix0,jx0);
 212             dy00             = _mm_sub_pd(iy0,jy0);
 213             dz00             = _mm_sub_pd(iz0,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 217
 218             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 219
 220             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 221
 222             /* Load parameters for j particles */
 223             jq0              = _mm_load_sd(charge+jnrA+0);
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm_mul_pd(iq0,jq0);
 231
 232             /* COULOMB ELECTROSTATICS */
 233             velec            = _mm_mul_pd(qq00,rinv00);
 234             felec            = _mm_mul_pd(velec,rinvsq00);
 235
 236             /* Update potential sum for this i atom from the interaction with this j atom. */
 237             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 238             velecsum         = _mm_add_pd(velecsum,velec);
 239
 240             fscal            = felec;
 241
 242             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 243
 244             /* Update vectorial force */
 245             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 246             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 247             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 248
 249             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 250                                                    _mm_mul_pd(dx00,fscal),
 251                                                    _mm_mul_pd(dy00,fscal),
 252                                                    _mm_mul_pd(dz00,fscal));
 253
 254             /* Inner loop uses 31 flops */
 255         }
 256
 257         /* End of innermost loop */
 258
 259         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 260                                               f+i_coord_offset,fshift+i_shift_offset);
 261
 262         ggid                        = gid[iidx];
 263         /* Update potential energies */
 264         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 265
 266         /* Increment number of inner iterations */
 267         inneriter                  += j_index_end - j_index_start;
 268
 269         /* Outer loop uses 8 flops */
 270     }
 271
 272     /* Increment number of outer iterations */
 273     outeriter        += nri;
 274
 275     /* Update outer/inner flops */
 276
 277     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*31);
 278 }
 279 /*
 280  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_double
 281  * Electrostatics interaction: Coulomb
 282  * VdW interaction:            None
 283  * Geometry:                   Particle-Particle
 284  * Calculate force/pot:        Force
 285  */
 286 void
 287 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_double
 288                     (t_nblist                    * gmx_restrict       nlist,
 289                      rvec                        * gmx_restrict          xx,
 290                      rvec                        * gmx_restrict          ff,
 291                      t_forcerec                  * gmx_restrict          fr,
 292                      t_mdatoms                   * gmx_restrict     mdatoms,
 293                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 294                      t_nrnb                      * gmx_restrict        nrnb)
 295 {
 296     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 297      * just 0 for non-waters.
 298      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 299      * jnr indices corresponding to data put in the four positions in the SIMD register.
 300      */
 301     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 302     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 303     int              jnrA,jnrB;
 304     int              j_coord_offsetA,j_coord_offsetB;
 305     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 306     real             rcutoff_scalar;
 307     real             *shiftvec,*fshift,*x,*f;
 308     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 309     int              vdwioffset0;
 310     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 311     int              vdwjidx0A,vdwjidx0B;
 312     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 313     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 314     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 315     real             *charge;
 316     __m128d          dummy_mask,cutoff_mask;
 317     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 318     __m128d          one     = _mm_set1_pd(1.0);
 319     __m128d          two     = _mm_set1_pd(2.0);
 320     x                = xx[0];
 321     f                = ff[0];
 322
 323     nri              = nlist->nri;
 324     iinr             = nlist->iinr;
 325     jindex           = nlist->jindex;
 326     jjnr             = nlist->jjnr;
 327     shiftidx         = nlist->shift;
 328     gid              = nlist->gid;
 329     shiftvec         = fr->shift_vec[0];
 330     fshift           = fr->fshift[0];
 331     facel            = _mm_set1_pd(fr->epsfac);
 332     charge           = mdatoms->chargeA;
 333
 334     /* Avoid stupid compiler warnings */
 335     jnrA = jnrB = 0;
 336     j_coord_offsetA = 0;
 337     j_coord_offsetB = 0;
 338
 339     outeriter        = 0;
 340     inneriter        = 0;
 341
 342     /* Start outer loop over neighborlists */
 343     for(iidx=0; iidx<nri; iidx++)
 344     {
 345         /* Load shift vector for this list */
 346         i_shift_offset   = DIM*shiftidx[iidx];
 347
 348         /* Load limits for loop over neighbors */
 349         j_index_start    = jindex[iidx];
 350         j_index_end      = jindex[iidx+1];
 351
 352         /* Get outer coordinate index */
 353         inr              = iinr[iidx];
 354         i_coord_offset   = DIM*inr;
 355
 356         /* Load i particle coords and add shift vector */
 357         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 358
 359         fix0             = _mm_setzero_pd();
 360         fiy0             = _mm_setzero_pd();
 361         fiz0             = _mm_setzero_pd();
 362
 363         /* Load parameters for i particles */
 364         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 365
 366         /* Start inner kernel loop */
 367         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 368         {
 369
 370             /* Get j neighbor index, and coordinate index */
 371             jnrA             = jjnr[jidx];
 372             jnrB             = jjnr[jidx+1];
 373             j_coord_offsetA  = DIM*jnrA;
 374             j_coord_offsetB  = DIM*jnrB;
 375
 376             /* load j atom coordinates */
 377             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 378                                               &jx0,&jy0,&jz0);
 379
 380             /* Calculate displacement vector */
 381             dx00             = _mm_sub_pd(ix0,jx0);
 382             dy00             = _mm_sub_pd(iy0,jy0);
 383             dz00             = _mm_sub_pd(iz0,jz0);
 384
 385             /* Calculate squared distance and things based on it */
 386             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 387
 388             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 389
 390             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 391
 392             /* Load parameters for j particles */
 393             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             /* Compute parameters for interactions between i and j atoms */
 400             qq00             = _mm_mul_pd(iq0,jq0);
 401
 402             /* COULOMB ELECTROSTATICS */
 403             velec            = _mm_mul_pd(qq00,rinv00);
 404             felec            = _mm_mul_pd(velec,rinvsq00);
 405
 406             fscal            = felec;
 407
 408             /* Update vectorial force */
 409             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 410             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 411             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 412
 413             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 414                                                    _mm_mul_pd(dx00,fscal),
 415                                                    _mm_mul_pd(dy00,fscal),
 416                                                    _mm_mul_pd(dz00,fscal));
 417
 418             /* Inner loop uses 30 flops */
 419         }
 420
 421         if(jidx<j_index_end)
 422         {
 423
 424             jnrA             = jjnr[jidx];
 425             j_coord_offsetA  = DIM*jnrA;
 426
 427             /* load j atom coordinates */
 428             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 429                                               &jx0,&jy0,&jz0);
 430
 431             /* Calculate displacement vector */
 432             dx00             = _mm_sub_pd(ix0,jx0);
 433             dy00             = _mm_sub_pd(iy0,jy0);
 434             dz00             = _mm_sub_pd(iz0,jz0);
 435
 436             /* Calculate squared distance and things based on it */
 437             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 438
 439             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 440
 441             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 442
 443             /* Load parameters for j particles */
 444             jq0              = _mm_load_sd(charge+jnrA+0);
 445
 446             /**************************
 447              * CALCULATE INTERACTIONS *
 448              **************************/
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq00             = _mm_mul_pd(iq0,jq0);
 452
 453             /* COULOMB ELECTROSTATICS */
 454             velec            = _mm_mul_pd(qq00,rinv00);
 455             felec            = _mm_mul_pd(velec,rinvsq00);
 456
 457             fscal            = felec;
 458
 459             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 460
 461             /* Update vectorial force */
 462             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 463             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 464             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 465
 466             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 467                                                    _mm_mul_pd(dx00,fscal),
 468                                                    _mm_mul_pd(dy00,fscal),
 469                                                    _mm_mul_pd(dz00,fscal));
 470
 471             /* Inner loop uses 30 flops */
 472         }
 473
 474         /* End of innermost loop */
 475
 476         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 477                                               f+i_coord_offset,fshift+i_shift_offset);
 478
 479         /* Increment number of inner iterations */
 480         inneriter                  += j_index_end - j_index_start;
 481
 482         /* Outer loop uses 7 flops */
 483     }
 484
 485     /* Increment number of outer iterations */
 486     outeriter        += nri;
 487
 488     /* Update outer/inner flops */
 489
 490     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 491 }