src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwNone_GeomP1P1_sse4_1_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 sse4_1_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_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_sse4_1_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_sse4_1_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           = sse41_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             /* Calculate temporary vectorial force */
 185             tx               = _mm_mul_pd(fscal,dx00);
 186             ty               = _mm_mul_pd(fscal,dy00);
 187             tz               = _mm_mul_pd(fscal,dz00);
 188
 189             /* Update vectorial force */
 190             fix0             = _mm_add_pd(fix0,tx);
 191             fiy0             = _mm_add_pd(fiy0,ty);
 192             fiz0             = _mm_add_pd(fiz0,tz);
 193
 194             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 195
 196             /* Inner loop uses 28 flops */
 197         }
 198
 199         if(jidx<j_index_end)
 200         {
 201
 202             jnrA             = jjnr[jidx];
 203             j_coord_offsetA  = DIM*jnrA;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_pd(ix0,jx0);
 211             dy00             = _mm_sub_pd(iy0,jy0);
 212             dz00             = _mm_sub_pd(iz0,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 216
 217             rinv00           = sse41_invsqrt_d(rsq00);
 218
 219             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 220
 221             /* Load parameters for j particles */
 222             jq0              = _mm_load_sd(charge+jnrA+0);
 223
 224             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             /* Compute parameters for interactions between i and j atoms */
 229             qq00             = _mm_mul_pd(iq0,jq0);
 230
 231             /* COULOMB ELECTROSTATICS */
 232             velec            = _mm_mul_pd(qq00,rinv00);
 233             felec            = _mm_mul_pd(velec,rinvsq00);
 234
 235             /* Update potential sum for this i atom from the interaction with this j atom. */
 236             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 237             velecsum         = _mm_add_pd(velecsum,velec);
 238
 239             fscal            = felec;
 240
 241             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 242
 243             /* Calculate temporary vectorial force */
 244             tx               = _mm_mul_pd(fscal,dx00);
 245             ty               = _mm_mul_pd(fscal,dy00);
 246             tz               = _mm_mul_pd(fscal,dz00);
 247
 248             /* Update vectorial force */
 249             fix0             = _mm_add_pd(fix0,tx);
 250             fiy0             = _mm_add_pd(fiy0,ty);
 251             fiz0             = _mm_add_pd(fiz0,tz);
 252
 253             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 254
 255             /* Inner loop uses 28 flops */
 256         }
 257
 258         /* End of innermost loop */
 259
 260         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 261                                               f+i_coord_offset,fshift+i_shift_offset);
 262
 263         ggid                        = gid[iidx];
 264         /* Update potential energies */
 265         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 266
 267         /* Increment number of inner iterations */
 268         inneriter                  += j_index_end - j_index_start;
 269
 270         /* Outer loop uses 8 flops */
 271     }
 272
 273     /* Increment number of outer iterations */
 274     outeriter        += nri;
 275
 276     /* Update outer/inner flops */
 277
 278     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*28);
 279 }
 280 /*
 281  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse4_1_double
 282  * Electrostatics interaction: Coulomb
 283  * VdW interaction:            None
 284  * Geometry:                   Particle-Particle
 285  * Calculate force/pot:        Force
 286  */
 287 void
 288 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_sse4_1_double
 289                     (t_nblist                    * gmx_restrict       nlist,
 290                      rvec                        * gmx_restrict          xx,
 291                      rvec                        * gmx_restrict          ff,
 292                      struct t_forcerec           * gmx_restrict          fr,
 293                      t_mdatoms                   * gmx_restrict     mdatoms,
 294                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 295                      t_nrnb                      * gmx_restrict        nrnb)
 296 {
 297     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 298      * just 0 for non-waters.
 299      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 300      * jnr indices corresponding to data put in the four positions in the SIMD register.
 301      */
 302     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 303     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 304     int              jnrA,jnrB;
 305     int              j_coord_offsetA,j_coord_offsetB;
 306     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 307     real             rcutoff_scalar;
 308     real             *shiftvec,*fshift,*x,*f;
 309     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 310     int              vdwioffset0;
 311     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 312     int              vdwjidx0A,vdwjidx0B;
 313     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 314     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 315     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 316     real             *charge;
 317     __m128d          dummy_mask,cutoff_mask;
 318     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 319     __m128d          one     = _mm_set1_pd(1.0);
 320     __m128d          two     = _mm_set1_pd(2.0);
 321     x                = xx[0];
 322     f                = ff[0];
 323
 324     nri              = nlist->nri;
 325     iinr             = nlist->iinr;
 326     jindex           = nlist->jindex;
 327     jjnr             = nlist->jjnr;
 328     shiftidx         = nlist->shift;
 329     gid              = nlist->gid;
 330     shiftvec         = fr->shift_vec[0];
 331     fshift           = fr->fshift[0];
 332     facel            = _mm_set1_pd(fr->ic->epsfac);
 333     charge           = mdatoms->chargeA;
 334
 335     /* Avoid stupid compiler warnings */
 336     jnrA = jnrB = 0;
 337     j_coord_offsetA = 0;
 338     j_coord_offsetB = 0;
 339
 340     outeriter        = 0;
 341     inneriter        = 0;
 342
 343     /* Start outer loop over neighborlists */
 344     for(iidx=0; iidx<nri; iidx++)
 345     {
 346         /* Load shift vector for this list */
 347         i_shift_offset   = DIM*shiftidx[iidx];
 348
 349         /* Load limits for loop over neighbors */
 350         j_index_start    = jindex[iidx];
 351         j_index_end      = jindex[iidx+1];
 352
 353         /* Get outer coordinate index */
 354         inr              = iinr[iidx];
 355         i_coord_offset   = DIM*inr;
 356
 357         /* Load i particle coords and add shift vector */
 358         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 359
 360         fix0             = _mm_setzero_pd();
 361         fiy0             = _mm_setzero_pd();
 362         fiz0             = _mm_setzero_pd();
 363
 364         /* Load parameters for i particles */
 365         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 366
 367         /* Start inner kernel loop */
 368         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 369         {
 370
 371             /* Get j neighbor index, and coordinate index */
 372             jnrA             = jjnr[jidx];
 373             jnrB             = jjnr[jidx+1];
 374             j_coord_offsetA  = DIM*jnrA;
 375             j_coord_offsetB  = DIM*jnrB;
 376
 377             /* load j atom coordinates */
 378             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 379                                               &jx0,&jy0,&jz0);
 380
 381             /* Calculate displacement vector */
 382             dx00             = _mm_sub_pd(ix0,jx0);
 383             dy00             = _mm_sub_pd(iy0,jy0);
 384             dz00             = _mm_sub_pd(iz0,jz0);
 385
 386             /* Calculate squared distance and things based on it */
 387             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 388
 389             rinv00           = sse41_invsqrt_d(rsq00);
 390
 391             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 392
 393             /* Load parameters for j particles */
 394             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             /* Compute parameters for interactions between i and j atoms */
 401             qq00             = _mm_mul_pd(iq0,jq0);
 402
 403             /* COULOMB ELECTROSTATICS */
 404             velec            = _mm_mul_pd(qq00,rinv00);
 405             felec            = _mm_mul_pd(velec,rinvsq00);
 406
 407             fscal            = felec;
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm_mul_pd(fscal,dx00);
 411             ty               = _mm_mul_pd(fscal,dy00);
 412             tz               = _mm_mul_pd(fscal,dz00);
 413
 414             /* Update vectorial force */
 415             fix0             = _mm_add_pd(fix0,tx);
 416             fiy0             = _mm_add_pd(fiy0,ty);
 417             fiz0             = _mm_add_pd(fiz0,tz);
 418
 419             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 420
 421             /* Inner loop uses 27 flops */
 422         }
 423
 424         if(jidx<j_index_end)
 425         {
 426
 427             jnrA             = jjnr[jidx];
 428             j_coord_offsetA  = DIM*jnrA;
 429
 430             /* load j atom coordinates */
 431             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 432                                               &jx0,&jy0,&jz0);
 433
 434             /* Calculate displacement vector */
 435             dx00             = _mm_sub_pd(ix0,jx0);
 436             dy00             = _mm_sub_pd(iy0,jy0);
 437             dz00             = _mm_sub_pd(iz0,jz0);
 438
 439             /* Calculate squared distance and things based on it */
 440             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 441
 442             rinv00           = sse41_invsqrt_d(rsq00);
 443
 444             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 445
 446             /* Load parameters for j particles */
 447             jq0              = _mm_load_sd(charge+jnrA+0);
 448
 449             /**************************
 450              * CALCULATE INTERACTIONS *
 451              **************************/
 452
 453             /* Compute parameters for interactions between i and j atoms */
 454             qq00             = _mm_mul_pd(iq0,jq0);
 455
 456             /* COULOMB ELECTROSTATICS */
 457             velec            = _mm_mul_pd(qq00,rinv00);
 458             felec            = _mm_mul_pd(velec,rinvsq00);
 459
 460             fscal            = felec;
 461
 462             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm_mul_pd(fscal,dx00);
 466             ty               = _mm_mul_pd(fscal,dy00);
 467             tz               = _mm_mul_pd(fscal,dz00);
 468
 469             /* Update vectorial force */
 470             fix0             = _mm_add_pd(fix0,tx);
 471             fiy0             = _mm_add_pd(fiy0,ty);
 472             fiz0             = _mm_add_pd(fiz0,tz);
 473
 474             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 475
 476             /* Inner loop uses 27 flops */
 477         }
 478
 479         /* End of innermost loop */
 480
 481         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 482                                               f+i_coord_offset,fshift+i_shift_offset);
 483
 484         /* Increment number of inner iterations */
 485         inneriter                  += j_index_end - j_index_start;
 486
 487         /* Outer loop uses 7 flops */
 488     }
 489
 490     /* Increment number of outer iterations */
 491     outeriter        += nri;
 492
 493     /* Update outer/inner flops */
 494
 495     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 496 }