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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_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
 104     /* Avoid stupid compiler warnings */
 105     jnrA = jnrB = 0;
 106     j_coord_offsetA = 0;
 107     j_coord_offsetB = 0;
 108
 109     outeriter        = 0;
 110     inneriter        = 0;
 111
 112     /* Start outer loop over neighborlists */
 113     for(iidx=0; iidx<nri; iidx++)
 114     {
 115         /* Load shift vector for this list */
 116         i_shift_offset   = DIM*shiftidx[iidx];
 117
 118         /* Load limits for loop over neighbors */
 119         j_index_start    = jindex[iidx];
 120         j_index_end      = jindex[iidx+1];
 121
 122         /* Get outer coordinate index */
 123         inr              = iinr[iidx];
 124         i_coord_offset   = DIM*inr;
 125
 126         /* Load i particle coords and add shift vector */
 127         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 128
 129         fix0             = _mm_setzero_pd();
 130         fiy0             = _mm_setzero_pd();
 131         fiz0             = _mm_setzero_pd();
 132
 133         /* Load parameters for i particles */
 134         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 135
 136         /* Reset potential sums */
 137         velecsum         = _mm_setzero_pd();
 138
 139         /* Start inner kernel loop */
 140         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 141         {
 142
 143             /* Get j neighbor index, and coordinate index */
 144             jnrA             = jjnr[jidx];
 145             jnrB             = jjnr[jidx+1];
 146             j_coord_offsetA  = DIM*jnrA;
 147             j_coord_offsetB  = DIM*jnrB;
 148
 149             /* load j atom coordinates */
 150             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 151                                               &jx0,&jy0,&jz0);
 152
 153             /* Calculate displacement vector */
 154             dx00             = _mm_sub_pd(ix0,jx0);
 155             dy00             = _mm_sub_pd(iy0,jy0);
 156             dz00             = _mm_sub_pd(iz0,jz0);
 157
 158             /* Calculate squared distance and things based on it */
 159             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 160
 161             rinv00           = avx128fma_invsqrt_d(rsq00);
 162
 163             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 164
 165             /* Load parameters for j particles */
 166             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 167
 168             /**************************
 169              * CALCULATE INTERACTIONS *
 170              **************************/
 171
 172             /* Compute parameters for interactions between i and j atoms */
 173             qq00             = _mm_mul_pd(iq0,jq0);
 174
 175             /* COULOMB ELECTROSTATICS */
 176             velec            = _mm_mul_pd(qq00,rinv00);
 177             felec            = _mm_mul_pd(velec,rinvsq00);
 178
 179             /* Update potential sum for this i atom from the interaction with this j atom. */
 180             velecsum         = _mm_add_pd(velecsum,velec);
 181
 182             fscal            = felec;
 183
 184             /* Update vectorial force */
 185             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 186             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 187             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 188
 189             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 190                                                    _mm_mul_pd(dx00,fscal),
 191                                                    _mm_mul_pd(dy00,fscal),
 192                                                    _mm_mul_pd(dz00,fscal));
 193
 194             /* Inner loop uses 31 flops */
 195         }
 196
 197         if(jidx<j_index_end)
 198         {
 199
 200             jnrA             = jjnr[jidx];
 201             j_coord_offsetA  = DIM*jnrA;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_pd(ix0,jx0);
 209             dy00             = _mm_sub_pd(iy0,jy0);
 210             dz00             = _mm_sub_pd(iz0,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 214
 215             rinv00           = avx128fma_invsqrt_d(rsq00);
 216
 217             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 218
 219             /* Load parameters for j particles */
 220             jq0              = _mm_load_sd(charge+jnrA+0);
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq00             = _mm_mul_pd(iq0,jq0);
 228
 229             /* COULOMB ELECTROSTATICS */
 230             velec            = _mm_mul_pd(qq00,rinv00);
 231             felec            = _mm_mul_pd(velec,rinvsq00);
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 235             velecsum         = _mm_add_pd(velecsum,velec);
 236
 237             fscal            = felec;
 238
 239             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 240
 241             /* Update vectorial force */
 242             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 243             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 244             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 245
 246             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 247                                                    _mm_mul_pd(dx00,fscal),
 248                                                    _mm_mul_pd(dy00,fscal),
 249                                                    _mm_mul_pd(dz00,fscal));
 250
 251             /* Inner loop uses 31 flops */
 252         }
 253
 254         /* End of innermost loop */
 255
 256         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 257                                               f+i_coord_offset,fshift+i_shift_offset);
 258
 259         ggid                        = gid[iidx];
 260         /* Update potential energies */
 261         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 262
 263         /* Increment number of inner iterations */
 264         inneriter                  += j_index_end - j_index_start;
 265
 266         /* Outer loop uses 8 flops */
 267     }
 268
 269     /* Increment number of outer iterations */
 270     outeriter        += nri;
 271
 272     /* Update outer/inner flops */
 273
 274     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*31);
 275 }
 276 /*
 277  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_double
 278  * Electrostatics interaction: Coulomb
 279  * VdW interaction:            None
 280  * Geometry:                   Particle-Particle
 281  * Calculate force/pot:        Force
 282  */
 283 void
 284 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_double
 285                     (t_nblist                    * gmx_restrict       nlist,
 286                      rvec                        * gmx_restrict          xx,
 287                      rvec                        * gmx_restrict          ff,
 288                      struct t_forcerec           * gmx_restrict          fr,
 289                      t_mdatoms                   * gmx_restrict     mdatoms,
 290                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 291                      t_nrnb                      * gmx_restrict        nrnb)
 292 {
 293     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 294      * just 0 for non-waters.
 295      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 296      * jnr indices corresponding to data put in the four positions in the SIMD register.
 297      */
 298     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 299     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 300     int              jnrA,jnrB;
 301     int              j_coord_offsetA,j_coord_offsetB;
 302     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 303     real             rcutoff_scalar;
 304     real             *shiftvec,*fshift,*x,*f;
 305     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 306     int              vdwioffset0;
 307     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 308     int              vdwjidx0A,vdwjidx0B;
 309     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 310     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 311     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 312     real             *charge;
 313     __m128d          dummy_mask,cutoff_mask;
 314     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 315     __m128d          one     = _mm_set1_pd(1.0);
 316     __m128d          two     = _mm_set1_pd(2.0);
 317     x                = xx[0];
 318     f                = ff[0];
 319
 320     nri              = nlist->nri;
 321     iinr             = nlist->iinr;
 322     jindex           = nlist->jindex;
 323     jjnr             = nlist->jjnr;
 324     shiftidx         = nlist->shift;
 325     gid              = nlist->gid;
 326     shiftvec         = fr->shift_vec[0];
 327     fshift           = fr->fshift[0];
 328     facel            = _mm_set1_pd(fr->ic->epsfac);
 329     charge           = mdatoms->chargeA;
 330
 331     /* Avoid stupid compiler warnings */
 332     jnrA = jnrB = 0;
 333     j_coord_offsetA = 0;
 334     j_coord_offsetB = 0;
 335
 336     outeriter        = 0;
 337     inneriter        = 0;
 338
 339     /* Start outer loop over neighborlists */
 340     for(iidx=0; iidx<nri; iidx++)
 341     {
 342         /* Load shift vector for this list */
 343         i_shift_offset   = DIM*shiftidx[iidx];
 344
 345         /* Load limits for loop over neighbors */
 346         j_index_start    = jindex[iidx];
 347         j_index_end      = jindex[iidx+1];
 348
 349         /* Get outer coordinate index */
 350         inr              = iinr[iidx];
 351         i_coord_offset   = DIM*inr;
 352
 353         /* Load i particle coords and add shift vector */
 354         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 355
 356         fix0             = _mm_setzero_pd();
 357         fiy0             = _mm_setzero_pd();
 358         fiz0             = _mm_setzero_pd();
 359
 360         /* Load parameters for i particles */
 361         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 362
 363         /* Start inner kernel loop */
 364         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 365         {
 366
 367             /* Get j neighbor index, and coordinate index */
 368             jnrA             = jjnr[jidx];
 369             jnrB             = jjnr[jidx+1];
 370             j_coord_offsetA  = DIM*jnrA;
 371             j_coord_offsetB  = DIM*jnrB;
 372
 373             /* load j atom coordinates */
 374             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 375                                               &jx0,&jy0,&jz0);
 376
 377             /* Calculate displacement vector */
 378             dx00             = _mm_sub_pd(ix0,jx0);
 379             dy00             = _mm_sub_pd(iy0,jy0);
 380             dz00             = _mm_sub_pd(iz0,jz0);
 381
 382             /* Calculate squared distance and things based on it */
 383             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 384
 385             rinv00           = avx128fma_invsqrt_d(rsq00);
 386
 387             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 388
 389             /* Load parameters for j particles */
 390             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             /* Compute parameters for interactions between i and j atoms */
 397             qq00             = _mm_mul_pd(iq0,jq0);
 398
 399             /* COULOMB ELECTROSTATICS */
 400             velec            = _mm_mul_pd(qq00,rinv00);
 401             felec            = _mm_mul_pd(velec,rinvsq00);
 402
 403             fscal            = felec;
 404
 405             /* Update vectorial force */
 406             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 407             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 408             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 409
 410             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 411                                                    _mm_mul_pd(dx00,fscal),
 412                                                    _mm_mul_pd(dy00,fscal),
 413                                                    _mm_mul_pd(dz00,fscal));
 414
 415             /* Inner loop uses 30 flops */
 416         }
 417
 418         if(jidx<j_index_end)
 419         {
 420
 421             jnrA             = jjnr[jidx];
 422             j_coord_offsetA  = DIM*jnrA;
 423
 424             /* load j atom coordinates */
 425             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 426                                               &jx0,&jy0,&jz0);
 427
 428             /* Calculate displacement vector */
 429             dx00             = _mm_sub_pd(ix0,jx0);
 430             dy00             = _mm_sub_pd(iy0,jy0);
 431             dz00             = _mm_sub_pd(iz0,jz0);
 432
 433             /* Calculate squared distance and things based on it */
 434             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 435
 436             rinv00           = avx128fma_invsqrt_d(rsq00);
 437
 438             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 439
 440             /* Load parameters for j particles */
 441             jq0              = _mm_load_sd(charge+jnrA+0);
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             /* Compute parameters for interactions between i and j atoms */
 448             qq00             = _mm_mul_pd(iq0,jq0);
 449
 450             /* COULOMB ELECTROSTATICS */
 451             velec            = _mm_mul_pd(qq00,rinv00);
 452             felec            = _mm_mul_pd(velec,rinvsq00);
 453
 454             fscal            = felec;
 455
 456             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 457
 458             /* Update vectorial force */
 459             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 460             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 461             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 462
 463             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 464                                                    _mm_mul_pd(dx00,fscal),
 465                                                    _mm_mul_pd(dy00,fscal),
 466                                                    _mm_mul_pd(dz00,fscal));
 467
 468             /* Inner loop uses 30 flops */
 469         }
 470
 471         /* End of innermost loop */
 472
 473         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 474                                               f+i_coord_offset,fshift+i_shift_offset);
 475
 476         /* Increment number of inner iterations */
 477         inneriter                  += j_index_end - j_index_start;
 478
 479         /* Outer loop uses 7 flops */
 480     }
 481
 482     /* Increment number of outer iterations */
 483     outeriter        += nri;
 484
 485     /* Update outer/inner flops */
 486
 487     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 488 }