src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_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_ElecRF_VdwNone_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     krf              = _mm_set1_pd(fr->ic->k_rf);
  91     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
  92     crf              = _mm_set1_pd(fr->ic->c_rf);
  93
  94     /* Avoid stupid compiler warnings */
  95     jnrA = jnrB = 0;
  96     j_coord_offsetA = 0;
  97     j_coord_offsetB = 0;
  98
  99     outeriter        = 0;
 100     inneriter        = 0;
 101
 102     /* Start outer loop over neighborlists */
 103     for(iidx=0; iidx<nri; iidx++)
 104     {
 105         /* Load shift vector for this list */
 106         i_shift_offset   = DIM*shiftidx[iidx];
 107
 108         /* Load limits for loop over neighbors */
 109         j_index_start    = jindex[iidx];
 110         j_index_end      = jindex[iidx+1];
 111
 112         /* Get outer coordinate index */
 113         inr              = iinr[iidx];
 114         i_coord_offset   = DIM*inr;
 115
 116         /* Load i particle coords and add shift vector */
 117         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 118
 119         fix0             = _mm_setzero_pd();
 120         fiy0             = _mm_setzero_pd();
 121         fiz0             = _mm_setzero_pd();
 122
 123         /* Load parameters for i particles */
 124         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 125
 126         /* Reset potential sums */
 127         velecsum         = _mm_setzero_pd();
 128
 129         /* Start inner kernel loop */
 130         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 131         {
 132
 133             /* Get j neighbor index, and coordinate index */
 134             jnrA             = jjnr[jidx];
 135             jnrB             = jjnr[jidx+1];
 136             j_coord_offsetA  = DIM*jnrA;
 137             j_coord_offsetB  = DIM*jnrB;
 138
 139             /* load j atom coordinates */
 140             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 141                                               &jx0,&jy0,&jz0);
 142
 143             /* Calculate displacement vector */
 144             dx00             = _mm_sub_pd(ix0,jx0);
 145             dy00             = _mm_sub_pd(iy0,jy0);
 146             dz00             = _mm_sub_pd(iz0,jz0);
 147
 148             /* Calculate squared distance and things based on it */
 149             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 150
 151             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 152
 153             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 154
 155             /* Load parameters for j particles */
 156             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 157
 158             /**************************
 159              * CALCULATE INTERACTIONS *
 160              **************************/
 161
 162             /* Compute parameters for interactions between i and j atoms */
 163             qq00             = _mm_mul_pd(iq0,jq0);
 164
 165             /* REACTION-FIELD ELECTROSTATICS */
 166             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 167             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 168
 169             /* Update potential sum for this i atom from the interaction with this j atom. */
 170             velecsum         = _mm_add_pd(velecsum,velec);
 171
 172             fscal            = felec;
 173
 174             /* Update vectorial force */
 175             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 176             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 177             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 178
 179             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 180                                                    _mm_mul_pd(dx00,fscal),
 181                                                    _mm_mul_pd(dy00,fscal),
 182                                                    _mm_mul_pd(dz00,fscal));
 183
 184             /* Inner loop uses 35 flops */
 185         }
 186
 187         if(jidx<j_index_end)
 188         {
 189
 190             jnrA             = jjnr[jidx];
 191             j_coord_offsetA  = DIM*jnrA;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_pd(ix0,jx0);
 199             dy00             = _mm_sub_pd(iy0,jy0);
 200             dz00             = _mm_sub_pd(iz0,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 204
 205             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 206
 207             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 208
 209             /* Load parameters for j particles */
 210             jq0              = _mm_load_sd(charge+jnrA+0);
 211
 212             /**************************
 213              * CALCULATE INTERACTIONS *
 214              **************************/
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq00             = _mm_mul_pd(iq0,jq0);
 218
 219             /* REACTION-FIELD ELECTROSTATICS */
 220             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 221             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 222
 223             /* Update potential sum for this i atom from the interaction with this j atom. */
 224             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 225             velecsum         = _mm_add_pd(velecsum,velec);
 226
 227             fscal            = felec;
 228
 229             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 230
 231             /* Update vectorial force */
 232             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 233             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 234             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 235
 236             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 237                                                    _mm_mul_pd(dx00,fscal),
 238                                                    _mm_mul_pd(dy00,fscal),
 239                                                    _mm_mul_pd(dz00,fscal));
 240
 241             /* Inner loop uses 35 flops */
 242         }
 243
 244         /* End of innermost loop */
 245
 246         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 247                                               f+i_coord_offset,fshift+i_shift_offset);
 248
 249         ggid                        = gid[iidx];
 250         /* Update potential energies */
 251         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 252
 253         /* Increment number of inner iterations */
 254         inneriter                  += j_index_end - j_index_start;
 255
 256         /* Outer loop uses 8 flops */
 257     }
 258
 259     /* Increment number of outer iterations */
 260     outeriter        += nri;
 261
 262     /* Update outer/inner flops */
 263
 264     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*35);
 265 }
 266 /*
 267  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_128_fma_double
 268  * Electrostatics interaction: ReactionField
 269  * VdW interaction:            None
 270  * Geometry:                   Particle-Particle
 271  * Calculate force/pot:        Force
 272  */
 273 void
 274 nb_kernel_ElecRF_VdwNone_GeomP1P1_F_avx_128_fma_double
 275                     (t_nblist * gmx_restrict                nlist,
 276                      rvec * gmx_restrict                    xx,
 277                      rvec * gmx_restrict                    ff,
 278                      t_forcerec * gmx_restrict              fr,
 279                      t_mdatoms * gmx_restrict               mdatoms,
 280                      nb_kernel_data_t * gmx_restrict        kernel_data,
 281                      t_nrnb * gmx_restrict                  nrnb)
 282 {
 283     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 284      * just 0 for non-waters.
 285      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 286      * jnr indices corresponding to data put in the four positions in the SIMD register.
 287      */
 288     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 289     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 290     int              jnrA,jnrB;
 291     int              j_coord_offsetA,j_coord_offsetB;
 292     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 293     real             rcutoff_scalar;
 294     real             *shiftvec,*fshift,*x,*f;
 295     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 296     int              vdwioffset0;
 297     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 298     int              vdwjidx0A,vdwjidx0B;
 299     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 300     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 301     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 302     real             *charge;
 303     __m128d          dummy_mask,cutoff_mask;
 304     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 305     __m128d          one     = _mm_set1_pd(1.0);
 306     __m128d          two     = _mm_set1_pd(2.0);
 307     x                = xx[0];
 308     f                = ff[0];
 309
 310     nri              = nlist->nri;
 311     iinr             = nlist->iinr;
 312     jindex           = nlist->jindex;
 313     jjnr             = nlist->jjnr;
 314     shiftidx         = nlist->shift;
 315     gid              = nlist->gid;
 316     shiftvec         = fr->shift_vec[0];
 317     fshift           = fr->fshift[0];
 318     facel            = _mm_set1_pd(fr->epsfac);
 319     charge           = mdatoms->chargeA;
 320     krf              = _mm_set1_pd(fr->ic->k_rf);
 321     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 322     crf              = _mm_set1_pd(fr->ic->c_rf);
 323
 324     /* Avoid stupid compiler warnings */
 325     jnrA = jnrB = 0;
 326     j_coord_offsetA = 0;
 327     j_coord_offsetB = 0;
 328
 329     outeriter        = 0;
 330     inneriter        = 0;
 331
 332     /* Start outer loop over neighborlists */
 333     for(iidx=0; iidx<nri; iidx++)
 334     {
 335         /* Load shift vector for this list */
 336         i_shift_offset   = DIM*shiftidx[iidx];
 337
 338         /* Load limits for loop over neighbors */
 339         j_index_start    = jindex[iidx];
 340         j_index_end      = jindex[iidx+1];
 341
 342         /* Get outer coordinate index */
 343         inr              = iinr[iidx];
 344         i_coord_offset   = DIM*inr;
 345
 346         /* Load i particle coords and add shift vector */
 347         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 348
 349         fix0             = _mm_setzero_pd();
 350         fiy0             = _mm_setzero_pd();
 351         fiz0             = _mm_setzero_pd();
 352
 353         /* Load parameters for i particles */
 354         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 355
 356         /* Start inner kernel loop */
 357         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 358         {
 359
 360             /* Get j neighbor index, and coordinate index */
 361             jnrA             = jjnr[jidx];
 362             jnrB             = jjnr[jidx+1];
 363             j_coord_offsetA  = DIM*jnrA;
 364             j_coord_offsetB  = DIM*jnrB;
 365
 366             /* load j atom coordinates */
 367             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 368                                               &jx0,&jy0,&jz0);
 369
 370             /* Calculate displacement vector */
 371             dx00             = _mm_sub_pd(ix0,jx0);
 372             dy00             = _mm_sub_pd(iy0,jy0);
 373             dz00             = _mm_sub_pd(iz0,jz0);
 374
 375             /* Calculate squared distance and things based on it */
 376             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 377
 378             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 379
 380             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 381
 382             /* Load parameters for j particles */
 383             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq00             = _mm_mul_pd(iq0,jq0);
 391
 392             /* REACTION-FIELD ELECTROSTATICS */
 393             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 394
 395             fscal            = felec;
 396
 397             /* Update vectorial force */
 398             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 399             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 400             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 401
 402             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 403                                                    _mm_mul_pd(dx00,fscal),
 404                                                    _mm_mul_pd(dy00,fscal),
 405                                                    _mm_mul_pd(dz00,fscal));
 406
 407             /* Inner loop uses 30 flops */
 408         }
 409
 410         if(jidx<j_index_end)
 411         {
 412
 413             jnrA             = jjnr[jidx];
 414             j_coord_offsetA  = DIM*jnrA;
 415
 416             /* load j atom coordinates */
 417             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 418                                               &jx0,&jy0,&jz0);
 419
 420             /* Calculate displacement vector */
 421             dx00             = _mm_sub_pd(ix0,jx0);
 422             dy00             = _mm_sub_pd(iy0,jy0);
 423             dz00             = _mm_sub_pd(iz0,jz0);
 424
 425             /* Calculate squared distance and things based on it */
 426             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 427
 428             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 429
 430             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 431
 432             /* Load parameters for j particles */
 433             jq0              = _mm_load_sd(charge+jnrA+0);
 434
 435             /**************************
 436              * CALCULATE INTERACTIONS *
 437              **************************/
 438
 439             /* Compute parameters for interactions between i and j atoms */
 440             qq00             = _mm_mul_pd(iq0,jq0);
 441
 442             /* REACTION-FIELD ELECTROSTATICS */
 443             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 444
 445             fscal            = felec;
 446
 447             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 448
 449             /* Update vectorial force */
 450             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 451             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 452             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 453
 454             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 455                                                    _mm_mul_pd(dx00,fscal),
 456                                                    _mm_mul_pd(dy00,fscal),
 457                                                    _mm_mul_pd(dz00,fscal));
 458
 459             /* Inner loop uses 30 flops */
 460         }
 461
 462         /* End of innermost loop */
 463
 464         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 465                                               f+i_coord_offset,fshift+i_shift_offset);
 466
 467         /* Increment number of inner iterations */
 468         inneriter                  += j_index_end - j_index_start;
 469
 470         /* Outer loop uses 7 flops */
 471     }
 472
 473     /* Increment number of outer iterations */
 474     outeriter        += nri;
 475
 476     /* Update outer/inner flops */
 477
 478     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*30);
 479 }