src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwNone_GeomP1P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomP1P1_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     __m128d          dummy_mask,cutoff_mask;
  74     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  75     __m128d          one     = _mm_set1_pd(1.0);
  76     __m128d          two     = _mm_set1_pd(2.0);
  77     x                = xx[0];
  78     f                = ff[0];
  79
  80     nri              = nlist->nri;
  81     iinr             = nlist->iinr;
  82     jindex           = nlist->jindex;
  83     jjnr             = nlist->jjnr;
  84     shiftidx         = nlist->shift;
  85     gid              = nlist->gid;
  86     shiftvec         = fr->shift_vec[0];
  87     fshift           = fr->fshift[0];
  88     facel            = _mm_set1_pd(fr->epsfac);
  89     charge           = mdatoms->chargeA;
  90
  91     /* Avoid stupid compiler warnings */
  92     jnrA = jnrB = 0;
  93     j_coord_offsetA = 0;
  94     j_coord_offsetB = 0;
  95
  96     outeriter        = 0;
  97     inneriter        = 0;
  98
  99     /* Start outer loop over neighborlists */
 100     for(iidx=0; iidx<nri; iidx++)
 101     {
 102         /* Load shift vector for this list */
 103         i_shift_offset   = DIM*shiftidx[iidx];
 104
 105         /* Load limits for loop over neighbors */
 106         j_index_start    = jindex[iidx];
 107         j_index_end      = jindex[iidx+1];
 108
 109         /* Get outer coordinate index */
 110         inr              = iinr[iidx];
 111         i_coord_offset   = DIM*inr;
 112
 113         /* Load i particle coords and add shift vector */
 114         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 115
 116         fix0             = _mm_setzero_pd();
 117         fiy0             = _mm_setzero_pd();
 118         fiz0             = _mm_setzero_pd();
 119
 120         /* Load parameters for i particles */
 121         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 122
 123         /* Reset potential sums */
 124         velecsum         = _mm_setzero_pd();
 125
 126         /* Start inner kernel loop */
 127         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 128         {
 129
 130             /* Get j neighbor index, and coordinate index */
 131             jnrA             = jjnr[jidx];
 132             jnrB             = jjnr[jidx+1];
 133             j_coord_offsetA  = DIM*jnrA;
 134             j_coord_offsetB  = DIM*jnrB;
 135
 136             /* load j atom coordinates */
 137             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 138                                               &jx0,&jy0,&jz0);
 139
 140             /* Calculate displacement vector */
 141             dx00             = _mm_sub_pd(ix0,jx0);
 142             dy00             = _mm_sub_pd(iy0,jy0);
 143             dz00             = _mm_sub_pd(iz0,jz0);
 144
 145             /* Calculate squared distance and things based on it */
 146             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 147
 148             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 149
 150             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 151
 152             /* Load parameters for j particles */
 153             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 154
 155             /**************************
 156              * CALCULATE INTERACTIONS *
 157              **************************/
 158
 159             /* Compute parameters for interactions between i and j atoms */
 160             qq00             = _mm_mul_pd(iq0,jq0);
 161
 162             /* COULOMB ELECTROSTATICS */
 163             velec            = _mm_mul_pd(qq00,rinv00);
 164             felec            = _mm_mul_pd(velec,rinvsq00);
 165
 166             /* Update potential sum for this i atom from the interaction with this j atom. */
 167             velecsum         = _mm_add_pd(velecsum,velec);
 168
 169             fscal            = felec;
 170
 171             /* Update vectorial force */
 172             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 173             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 174             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 175
 176             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 177                                                    _mm_mul_pd(dx00,fscal),
 178                                                    _mm_mul_pd(dy00,fscal),
 179                                                    _mm_mul_pd(dz00,fscal));
 180
 181             /* Inner loop uses 31 flops */
 182         }
 183
 184         if(jidx<j_index_end)
 185         {
 186
 187             jnrA             = jjnr[jidx];
 188             j_coord_offsetA  = DIM*jnrA;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_pd(ix0,jx0);
 196             dy00             = _mm_sub_pd(iy0,jy0);
 197             dz00             = _mm_sub_pd(iz0,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 201
 202             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 203
 204             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 205
 206             /* Load parameters for j particles */
 207             jq0              = _mm_load_sd(charge+jnrA+0);
 208
 209             /**************************
 210              * CALCULATE INTERACTIONS *
 211              **************************/
 212
 213             /* Compute parameters for interactions between i and j atoms */
 214             qq00             = _mm_mul_pd(iq0,jq0);
 215
 216             /* COULOMB ELECTROSTATICS */
 217             velec            = _mm_mul_pd(qq00,rinv00);
 218             felec            = _mm_mul_pd(velec,rinvsq00);
 219
 220             /* Update potential sum for this i atom from the interaction with this j atom. */
 221             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 222             velecsum         = _mm_add_pd(velecsum,velec);
 223
 224             fscal            = felec;
 225
 226             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 227
 228             /* Update vectorial force */
 229             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 230             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 231             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 232
 233             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 234                                                    _mm_mul_pd(dx00,fscal),
 235                                                    _mm_mul_pd(dy00,fscal),
 236                                                    _mm_mul_pd(dz00,fscal));
 237
 238             /* Inner loop uses 31 flops */
 239         }
 240
 241         /* End of innermost loop */
 242
 243         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 244                                               f+i_coord_offset,fshift+i_shift_offset);
 245
 246         ggid                        = gid[iidx];
 247         /* Update potential energies */
 248         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 249
 250         /* Increment number of inner iterations */
 251         inneriter                  += j_index_end - j_index_start;
 252
 253         /* Outer loop uses 8 flops */
 254     }
 255
 256     /* Increment number of outer iterations */
 257     outeriter        += nri;
 258
 259     /* Update outer/inner flops */
 260
 261     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*31);
 262 }
 263 /*
 264  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_double
 265  * Electrostatics interaction: Coulomb
 266  * VdW interaction:            None
 267  * Geometry:                   Particle-Particle
 268  * Calculate force/pot:        Force
 269  */
 270 void
 271 nb_kernel_ElecCoul_VdwNone_GeomP1P1_F_avx_128_fma_double
 272                     (t_nblist * gmx_restrict                nlist,
 273                      rvec * gmx_restrict                    xx,
 274                      rvec * gmx_restrict                    ff,
 275                      t_forcerec * gmx_restrict              fr,
 276                      t_mdatoms * gmx_restrict               mdatoms,
 277                      nb_kernel_data_t * gmx_restrict        kernel_data,
 278                      t_nrnb * gmx_restrict                  nrnb)
 279 {
 280     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 281      * just 0 for non-waters.
 282      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 283      * jnr indices corresponding to data put in the four positions in the SIMD register.
 284      */
 285     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 286     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 287     int              jnrA,jnrB;
 288     int              j_coord_offsetA,j_coord_offsetB;
 289     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 290     real             rcutoff_scalar;
 291     real             *shiftvec,*fshift,*x,*f;
 292     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 293     int              vdwioffset0;
 294     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 295     int              vdwjidx0A,vdwjidx0B;
 296     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 297     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 298     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 299     real             *charge;
 300     __m128d          dummy_mask,cutoff_mask;
 301     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 302     __m128d          one     = _mm_set1_pd(1.0);
 303     __m128d          two     = _mm_set1_pd(2.0);
 304     x                = xx[0];
 305     f                = ff[0];
 306
 307     nri              = nlist->nri;
 308     iinr             = nlist->iinr;
 309     jindex           = nlist->jindex;
 310     jjnr             = nlist->jjnr;
 311     shiftidx         = nlist->shift;
 312     gid              = nlist->gid;
 313     shiftvec         = fr->shift_vec[0];
 314     fshift           = fr->fshift[0];
 315     facel            = _mm_set1_pd(fr->epsfac);
 316     charge           = mdatoms->chargeA;
 317
 318     /* Avoid stupid compiler warnings */
 319     jnrA = jnrB = 0;
 320     j_coord_offsetA = 0;
 321     j_coord_offsetB = 0;
 322
 323     outeriter        = 0;
 324     inneriter        = 0;
 325
 326     /* Start outer loop over neighborlists */
 327     for(iidx=0; iidx<nri; iidx++)
 328     {
 329         /* Load shift vector for this list */
 330         i_shift_offset   = DIM*shiftidx[iidx];
 331
 332         /* Load limits for loop over neighbors */
 333         j_index_start    = jindex[iidx];
 334         j_index_end      = jindex[iidx+1];
 335
 336         /* Get outer coordinate index */
 337         inr              = iinr[iidx];
 338         i_coord_offset   = DIM*inr;
 339
 340         /* Load i particle coords and add shift vector */
 341         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 342
 343         fix0             = _mm_setzero_pd();
 344         fiy0             = _mm_setzero_pd();
 345         fiz0             = _mm_setzero_pd();
 346
 347         /* Load parameters for i particles */
 348         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 349
 350         /* Start inner kernel loop */
 351         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 352         {
 353
 354             /* Get j neighbor index, and coordinate index */
 355             jnrA             = jjnr[jidx];
 356             jnrB             = jjnr[jidx+1];
 357             j_coord_offsetA  = DIM*jnrA;
 358             j_coord_offsetB  = DIM*jnrB;
 359
 360             /* load j atom coordinates */
 361             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 362                                               &jx0,&jy0,&jz0);
 363
 364             /* Calculate displacement vector */
 365             dx00             = _mm_sub_pd(ix0,jx0);
 366             dy00             = _mm_sub_pd(iy0,jy0);
 367             dz00             = _mm_sub_pd(iz0,jz0);
 368
 369             /* Calculate squared distance and things based on it */
 370             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 371
 372             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 373
 374             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 375
 376             /* Load parameters for j particles */
 377             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq00             = _mm_mul_pd(iq0,jq0);
 385
 386             /* COULOMB ELECTROSTATICS */
 387             velec            = _mm_mul_pd(qq00,rinv00);
 388             felec            = _mm_mul_pd(velec,rinvsq00);
 389
 390             fscal            = felec;
 391
 392             /* Update vectorial force */
 393             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 394             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 395             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 396
 397             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 398                                                    _mm_mul_pd(dx00,fscal),
 399                                                    _mm_mul_pd(dy00,fscal),
 400                                                    _mm_mul_pd(dz00,fscal));
 401
 402             /* Inner loop uses 30 flops */
 403         }
 404
 405         if(jidx<j_index_end)
 406         {
 407
 408             jnrA             = jjnr[jidx];
 409             j_coord_offsetA  = DIM*jnrA;
 410
 411             /* load j atom coordinates */
 412             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 413                                               &jx0,&jy0,&jz0);
 414
 415             /* Calculate displacement vector */
 416             dx00             = _mm_sub_pd(ix0,jx0);
 417             dy00             = _mm_sub_pd(iy0,jy0);
 418             dz00             = _mm_sub_pd(iz0,jz0);
 419
 420             /* Calculate squared distance and things based on it */
 421             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 422
 423             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 424
 425             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 426
 427             /* Load parameters for j particles */
 428             jq0              = _mm_load_sd(charge+jnrA+0);
 429
 430             /**************************
 431              * CALCULATE INTERACTIONS *
 432              **************************/
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             qq00             = _mm_mul_pd(iq0,jq0);
 436
 437             /* COULOMB ELECTROSTATICS */
 438             velec            = _mm_mul_pd(qq00,rinv00);
 439             felec            = _mm_mul_pd(velec,rinvsq00);
 440
 441             fscal            = felec;
 442
 443             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 444
 445             /* Update vectorial force */
 446             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 447             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 448             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 449
 450             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 451                                                    _mm_mul_pd(dx00,fscal),
 452                                                    _mm_mul_pd(dy00,fscal),
 453                                                    _mm_mul_pd(dz00,fscal));
 454
 455             /* Inner loop uses 30 flops */
 456         }
 457
 458         /* End of innermost loop */
 459
 460         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 461                                               f+i_coord_offset,fshift+i_shift_offset);
 462
 463         /* Increment number of inner iterations */
 464         inneriter                  += j_index_end - j_index_start;
 465
 466         /* Outer loop uses 7 flops */
 467     }
 468
 469     /* Increment number of outer iterations */
 470     outeriter        += nri;
 471
 472     /* Update outer/inner flops */
 473
 474     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 475 }