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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwNone_GeomP1P1_VF_sse2_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     krf              = _mm_set1_pd(fr->ic->k_rf);
 107     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 108     crf              = _mm_set1_pd(fr->ic->c_rf);
 109
 110     /* Avoid stupid compiler warnings */
 111     jnrA = jnrB = 0;
 112     j_coord_offsetA = 0;
 113     j_coord_offsetB = 0;
 114
 115     outeriter        = 0;
 116     inneriter        = 0;
 117
 118     /* Start outer loop over neighborlists */
 119     for(iidx=0; iidx<nri; iidx++)
 120     {
 121         /* Load shift vector for this list */
 122         i_shift_offset   = DIM*shiftidx[iidx];
 123
 124         /* Load limits for loop over neighbors */
 125         j_index_start    = jindex[iidx];
 126         j_index_end      = jindex[iidx+1];
 127
 128         /* Get outer coordinate index */
 129         inr              = iinr[iidx];
 130         i_coord_offset   = DIM*inr;
 131
 132         /* Load i particle coords and add shift vector */
 133         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 134
 135         fix0             = _mm_setzero_pd();
 136         fiy0             = _mm_setzero_pd();
 137         fiz0             = _mm_setzero_pd();
 138
 139         /* Load parameters for i particles */
 140         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 141
 142         /* Reset potential sums */
 143         velecsum         = _mm_setzero_pd();
 144
 145         /* Start inner kernel loop */
 146         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 147         {
 148
 149             /* Get j neighbor index, and coordinate index */
 150             jnrA             = jjnr[jidx];
 151             jnrB             = jjnr[jidx+1];
 152             j_coord_offsetA  = DIM*jnrA;
 153             j_coord_offsetB  = DIM*jnrB;
 154
 155             /* load j atom coordinates */
 156             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 157                                               &jx0,&jy0,&jz0);
 158
 159             /* Calculate displacement vector */
 160             dx00             = _mm_sub_pd(ix0,jx0);
 161             dy00             = _mm_sub_pd(iy0,jy0);
 162             dz00             = _mm_sub_pd(iz0,jz0);
 163
 164             /* Calculate squared distance and things based on it */
 165             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 166
 167             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 168
 169             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 170
 171             /* Load parameters for j particles */
 172             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 173
 174             /**************************
 175              * CALCULATE INTERACTIONS *
 176              **************************/
 177
 178             /* Compute parameters for interactions between i and j atoms */
 179             qq00             = _mm_mul_pd(iq0,jq0);
 180
 181             /* REACTION-FIELD ELECTROSTATICS */
 182             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 183             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 184
 185             /* Update potential sum for this i atom from the interaction with this j atom. */
 186             velecsum         = _mm_add_pd(velecsum,velec);
 187
 188             fscal            = felec;
 189
 190             /* Calculate temporary vectorial force */
 191             tx               = _mm_mul_pd(fscal,dx00);
 192             ty               = _mm_mul_pd(fscal,dy00);
 193             tz               = _mm_mul_pd(fscal,dz00);
 194
 195             /* Update vectorial force */
 196             fix0             = _mm_add_pd(fix0,tx);
 197             fiy0             = _mm_add_pd(fiy0,ty);
 198             fiz0             = _mm_add_pd(fiz0,tz);
 199
 200             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 201
 202             /* Inner loop uses 32 flops */
 203         }
 204
 205         if(jidx<j_index_end)
 206         {
 207
 208             jnrA             = jjnr[jidx];
 209             j_coord_offsetA  = DIM*jnrA;
 210
 211             /* load j atom coordinates */
 212             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 213                                               &jx0,&jy0,&jz0);
 214
 215             /* Calculate displacement vector */
 216             dx00             = _mm_sub_pd(ix0,jx0);
 217             dy00             = _mm_sub_pd(iy0,jy0);
 218             dz00             = _mm_sub_pd(iz0,jz0);
 219
 220             /* Calculate squared distance and things based on it */
 221             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 222
 223             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 224
 225             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 226
 227             /* Load parameters for j particles */
 228             jq0              = _mm_load_sd(charge+jnrA+0);
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm_mul_pd(iq0,jq0);
 236
 237             /* REACTION-FIELD ELECTROSTATICS */
 238             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 239             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 240
 241             /* Update potential sum for this i atom from the interaction with this j atom. */
 242             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 243             velecsum         = _mm_add_pd(velecsum,velec);
 244
 245             fscal            = felec;
 246
 247             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 248
 249             /* Calculate temporary vectorial force */
 250             tx               = _mm_mul_pd(fscal,dx00);
 251             ty               = _mm_mul_pd(fscal,dy00);
 252             tz               = _mm_mul_pd(fscal,dz00);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_add_pd(fix0,tx);
 256             fiy0             = _mm_add_pd(fiy0,ty);
 257             fiz0             = _mm_add_pd(fiz0,tz);
 258
 259             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 260
 261             /* Inner loop uses 32 flops */
 262         }
 263
 264         /* End of innermost loop */
 265
 266         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 267                                               f+i_coord_offset,fshift+i_shift_offset);
 268
 269         ggid                        = gid[iidx];
 270         /* Update potential energies */
 271         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 272
 273         /* Increment number of inner iterations */
 274         inneriter                  += j_index_end - j_index_start;
 275
 276         /* Outer loop uses 8 flops */
 277     }
 278
 279     /* Increment number of outer iterations */
 280     outeriter        += nri;
 281
 282     /* Update outer/inner flops */
 283
 284     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*32);
 285 }
 286 /*
 287  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_double
 288  * Electrostatics interaction: ReactionField
 289  * VdW interaction:            None
 290  * Geometry:                   Particle-Particle
 291  * Calculate force/pot:        Force
 292  */
 293 void
 294 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_sse2_double
 295                     (t_nblist                    * gmx_restrict       nlist,
 296                      rvec                        * gmx_restrict          xx,
 297                      rvec                        * gmx_restrict          ff,
 298                      t_forcerec                  * gmx_restrict          fr,
 299                      t_mdatoms                   * gmx_restrict     mdatoms,
 300                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 301                      t_nrnb                      * gmx_restrict        nrnb)
 302 {
 303     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 304      * just 0 for non-waters.
 305      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 306      * jnr indices corresponding to data put in the four positions in the SIMD register.
 307      */
 308     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 309     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 310     int              jnrA,jnrB;
 311     int              j_coord_offsetA,j_coord_offsetB;
 312     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 313     real             rcutoff_scalar;
 314     real             *shiftvec,*fshift,*x,*f;
 315     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 316     int              vdwioffset0;
 317     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 318     int              vdwjidx0A,vdwjidx0B;
 319     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 320     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 321     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 322     real             *charge;
 323     __m128d          dummy_mask,cutoff_mask;
 324     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 325     __m128d          one     = _mm_set1_pd(1.0);
 326     __m128d          two     = _mm_set1_pd(2.0);
 327     x                = xx[0];
 328     f                = ff[0];
 329
 330     nri              = nlist->nri;
 331     iinr             = nlist->iinr;
 332     jindex           = nlist->jindex;
 333     jjnr             = nlist->jjnr;
 334     shiftidx         = nlist->shift;
 335     gid              = nlist->gid;
 336     shiftvec         = fr->shift_vec[0];
 337     fshift           = fr->fshift[0];
 338     facel            = _mm_set1_pd(fr->epsfac);
 339     charge           = mdatoms->chargeA;
 340     krf              = _mm_set1_pd(fr->ic->k_rf);
 341     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 342     crf              = _mm_set1_pd(fr->ic->c_rf);
 343
 344     /* Avoid stupid compiler warnings */
 345     jnrA = jnrB = 0;
 346     j_coord_offsetA = 0;
 347     j_coord_offsetB = 0;
 348
 349     outeriter        = 0;
 350     inneriter        = 0;
 351
 352     /* Start outer loop over neighborlists */
 353     for(iidx=0; iidx<nri; iidx++)
 354     {
 355         /* Load shift vector for this list */
 356         i_shift_offset   = DIM*shiftidx[iidx];
 357
 358         /* Load limits for loop over neighbors */
 359         j_index_start    = jindex[iidx];
 360         j_index_end      = jindex[iidx+1];
 361
 362         /* Get outer coordinate index */
 363         inr              = iinr[iidx];
 364         i_coord_offset   = DIM*inr;
 365
 366         /* Load i particle coords and add shift vector */
 367         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 368
 369         fix0             = _mm_setzero_pd();
 370         fiy0             = _mm_setzero_pd();
 371         fiz0             = _mm_setzero_pd();
 372
 373         /* Load parameters for i particles */
 374         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 375
 376         /* Start inner kernel loop */
 377         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 378         {
 379
 380             /* Get j neighbor index, and coordinate index */
 381             jnrA             = jjnr[jidx];
 382             jnrB             = jjnr[jidx+1];
 383             j_coord_offsetA  = DIM*jnrA;
 384             j_coord_offsetB  = DIM*jnrB;
 385
 386             /* load j atom coordinates */
 387             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 388                                               &jx0,&jy0,&jz0);
 389
 390             /* Calculate displacement vector */
 391             dx00             = _mm_sub_pd(ix0,jx0);
 392             dy00             = _mm_sub_pd(iy0,jy0);
 393             dz00             = _mm_sub_pd(iz0,jz0);
 394
 395             /* Calculate squared distance and things based on it */
 396             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 397
 398             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 399
 400             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 401
 402             /* Load parameters for j particles */
 403             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             /* Compute parameters for interactions between i and j atoms */
 410             qq00             = _mm_mul_pd(iq0,jq0);
 411
 412             /* REACTION-FIELD ELECTROSTATICS */
 413             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 414
 415             fscal            = felec;
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm_mul_pd(fscal,dx00);
 419             ty               = _mm_mul_pd(fscal,dy00);
 420             tz               = _mm_mul_pd(fscal,dz00);
 421
 422             /* Update vectorial force */
 423             fix0             = _mm_add_pd(fix0,tx);
 424             fiy0             = _mm_add_pd(fiy0,ty);
 425             fiz0             = _mm_add_pd(fiz0,tz);
 426
 427             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 428
 429             /* Inner loop uses 27 flops */
 430         }
 431
 432         if(jidx<j_index_end)
 433         {
 434
 435             jnrA             = jjnr[jidx];
 436             j_coord_offsetA  = DIM*jnrA;
 437
 438             /* load j atom coordinates */
 439             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 440                                               &jx0,&jy0,&jz0);
 441
 442             /* Calculate displacement vector */
 443             dx00             = _mm_sub_pd(ix0,jx0);
 444             dy00             = _mm_sub_pd(iy0,jy0);
 445             dz00             = _mm_sub_pd(iz0,jz0);
 446
 447             /* Calculate squared distance and things based on it */
 448             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 449
 450             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 451
 452             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 453
 454             /* Load parameters for j particles */
 455             jq0              = _mm_load_sd(charge+jnrA+0);
 456
 457             /**************************
 458              * CALCULATE INTERACTIONS *
 459              **************************/
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             qq00             = _mm_mul_pd(iq0,jq0);
 463
 464             /* REACTION-FIELD ELECTROSTATICS */
 465             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 466
 467             fscal            = felec;
 468
 469             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 470
 471             /* Calculate temporary vectorial force */
 472             tx               = _mm_mul_pd(fscal,dx00);
 473             ty               = _mm_mul_pd(fscal,dy00);
 474             tz               = _mm_mul_pd(fscal,dz00);
 475
 476             /* Update vectorial force */
 477             fix0             = _mm_add_pd(fix0,tx);
 478             fiy0             = _mm_add_pd(fiy0,ty);
 479             fiz0             = _mm_add_pd(fiz0,tz);
 480
 481             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 482
 483             /* Inner loop uses 27 flops */
 484         }
 485
 486         /* End of innermost loop */
 487
 488         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 489                                               f+i_coord_offset,fshift+i_shift_offset);
 490
 491         /* Increment number of inner iterations */
 492         inneriter                  += j_index_end - j_index_start;
 493
 494         /* Outer loop uses 7 flops */
 495     }
 496
 497     /* Increment number of outer iterations */
 498     outeriter        += nri;
 499
 500     /* Update outer/inner flops */
 501
 502     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*27);
 503 }