src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_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_ElecRFCut_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_ElecRFCut_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
  95     rcutoff_scalar   = fr->rcoulomb;
  96     rcutoff          = _mm_set1_pd(rcutoff_scalar);
  97     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
  98
  99     /* Avoid stupid compiler warnings */
 100     jnrA = jnrB = 0;
 101     j_coord_offsetA = 0;
 102     j_coord_offsetB = 0;
 103
 104     outeriter        = 0;
 105     inneriter        = 0;
 106
 107     /* Start outer loop over neighborlists */
 108     for(iidx=0; iidx<nri; iidx++)
 109     {
 110         /* Load shift vector for this list */
 111         i_shift_offset   = DIM*shiftidx[iidx];
 112
 113         /* Load limits for loop over neighbors */
 114         j_index_start    = jindex[iidx];
 115         j_index_end      = jindex[iidx+1];
 116
 117         /* Get outer coordinate index */
 118         inr              = iinr[iidx];
 119         i_coord_offset   = DIM*inr;
 120
 121         /* Load i particle coords and add shift vector */
 122         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 123
 124         fix0             = _mm_setzero_pd();
 125         fiy0             = _mm_setzero_pd();
 126         fiz0             = _mm_setzero_pd();
 127
 128         /* Load parameters for i particles */
 129         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 130
 131         /* Reset potential sums */
 132         velecsum         = _mm_setzero_pd();
 133
 134         /* Start inner kernel loop */
 135         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 136         {
 137
 138             /* Get j neighbor index, and coordinate index */
 139             jnrA             = jjnr[jidx];
 140             jnrB             = jjnr[jidx+1];
 141             j_coord_offsetA  = DIM*jnrA;
 142             j_coord_offsetB  = DIM*jnrB;
 143
 144             /* load j atom coordinates */
 145             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 146                                               &jx0,&jy0,&jz0);
 147
 148             /* Calculate displacement vector */
 149             dx00             = _mm_sub_pd(ix0,jx0);
 150             dy00             = _mm_sub_pd(iy0,jy0);
 151             dz00             = _mm_sub_pd(iz0,jz0);
 152
 153             /* Calculate squared distance and things based on it */
 154             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 155
 156             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 157
 158             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 159
 160             /* Load parameters for j particles */
 161             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 162
 163             /**************************
 164              * CALCULATE INTERACTIONS *
 165              **************************/
 166
 167             if (gmx_mm_any_lt(rsq00,rcutoff2))
 168             {
 169
 170             /* Compute parameters for interactions between i and j atoms */
 171             qq00             = _mm_mul_pd(iq0,jq0);
 172
 173             /* REACTION-FIELD ELECTROSTATICS */
 174             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 175             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 176
 177             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 178
 179             /* Update potential sum for this i atom from the interaction with this j atom. */
 180             velec            = _mm_and_pd(velec,cutoff_mask);
 181             velecsum         = _mm_add_pd(velecsum,velec);
 182
 183             fscal            = felec;
 184
 185             fscal            = _mm_and_pd(fscal,cutoff_mask);
 186
 187             /* Update vectorial force */
 188             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 189             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 190             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 191
 192             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 193                                                    _mm_mul_pd(dx00,fscal),
 194                                                    _mm_mul_pd(dy00,fscal),
 195                                                    _mm_mul_pd(dz00,fscal));
 196
 197             }
 198
 199             /* Inner loop uses 39 flops */
 200         }
 201
 202         if(jidx<j_index_end)
 203         {
 204
 205             jnrA             = jjnr[jidx];
 206             j_coord_offsetA  = DIM*jnrA;
 207
 208             /* load j atom coordinates */
 209             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 210                                               &jx0,&jy0,&jz0);
 211
 212             /* Calculate displacement vector */
 213             dx00             = _mm_sub_pd(ix0,jx0);
 214             dy00             = _mm_sub_pd(iy0,jy0);
 215             dz00             = _mm_sub_pd(iz0,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 219
 220             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 221
 222             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 223
 224             /* Load parameters for j particles */
 225             jq0              = _mm_load_sd(charge+jnrA+0);
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             if (gmx_mm_any_lt(rsq00,rcutoff2))
 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_macc_pd(krf,rsq00,rinv00),crf));
 239             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 240
 241             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velec            = _mm_and_pd(velec,cutoff_mask);
 245             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 246             velecsum         = _mm_add_pd(velecsum,velec);
 247
 248             fscal            = felec;
 249
 250             fscal            = _mm_and_pd(fscal,cutoff_mask);
 251
 252             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 256             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 257             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 258
 259             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 260                                                    _mm_mul_pd(dx00,fscal),
 261                                                    _mm_mul_pd(dy00,fscal),
 262                                                    _mm_mul_pd(dz00,fscal));
 263
 264             }
 265
 266             /* Inner loop uses 39 flops */
 267         }
 268
 269         /* End of innermost loop */
 270
 271         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 272                                               f+i_coord_offset,fshift+i_shift_offset);
 273
 274         ggid                        = gid[iidx];
 275         /* Update potential energies */
 276         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 277
 278         /* Increment number of inner iterations */
 279         inneriter                  += j_index_end - j_index_start;
 280
 281         /* Outer loop uses 8 flops */
 282     }
 283
 284     /* Increment number of outer iterations */
 285     outeriter        += nri;
 286
 287     /* Update outer/inner flops */
 288
 289     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*39);
 290 }
 291 /*
 292  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_128_fma_double
 293  * Electrostatics interaction: ReactionField
 294  * VdW interaction:            None
 295  * Geometry:                   Particle-Particle
 296  * Calculate force/pot:        Force
 297  */
 298 void
 299 nb_kernel_ElecRFCut_VdwNone_GeomP1P1_F_avx_128_fma_double
 300                     (t_nblist * gmx_restrict                nlist,
 301                      rvec * gmx_restrict                    xx,
 302                      rvec * gmx_restrict                    ff,
 303                      t_forcerec * gmx_restrict              fr,
 304                      t_mdatoms * gmx_restrict               mdatoms,
 305                      nb_kernel_data_t * gmx_restrict        kernel_data,
 306                      t_nrnb * gmx_restrict                  nrnb)
 307 {
 308     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 309      * just 0 for non-waters.
 310      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 311      * jnr indices corresponding to data put in the four positions in the SIMD register.
 312      */
 313     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 314     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 315     int              jnrA,jnrB;
 316     int              j_coord_offsetA,j_coord_offsetB;
 317     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 318     real             rcutoff_scalar;
 319     real             *shiftvec,*fshift,*x,*f;
 320     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 321     int              vdwioffset0;
 322     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 323     int              vdwjidx0A,vdwjidx0B;
 324     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 325     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 326     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 327     real             *charge;
 328     __m128d          dummy_mask,cutoff_mask;
 329     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 330     __m128d          one     = _mm_set1_pd(1.0);
 331     __m128d          two     = _mm_set1_pd(2.0);
 332     x                = xx[0];
 333     f                = ff[0];
 334
 335     nri              = nlist->nri;
 336     iinr             = nlist->iinr;
 337     jindex           = nlist->jindex;
 338     jjnr             = nlist->jjnr;
 339     shiftidx         = nlist->shift;
 340     gid              = nlist->gid;
 341     shiftvec         = fr->shift_vec[0];
 342     fshift           = fr->fshift[0];
 343     facel            = _mm_set1_pd(fr->epsfac);
 344     charge           = mdatoms->chargeA;
 345     krf              = _mm_set1_pd(fr->ic->k_rf);
 346     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 347     crf              = _mm_set1_pd(fr->ic->c_rf);
 348
 349     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 350     rcutoff_scalar   = fr->rcoulomb;
 351     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 352     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 353
 354     /* Avoid stupid compiler warnings */
 355     jnrA = jnrB = 0;
 356     j_coord_offsetA = 0;
 357     j_coord_offsetB = 0;
 358
 359     outeriter        = 0;
 360     inneriter        = 0;
 361
 362     /* Start outer loop over neighborlists */
 363     for(iidx=0; iidx<nri; iidx++)
 364     {
 365         /* Load shift vector for this list */
 366         i_shift_offset   = DIM*shiftidx[iidx];
 367
 368         /* Load limits for loop over neighbors */
 369         j_index_start    = jindex[iidx];
 370         j_index_end      = jindex[iidx+1];
 371
 372         /* Get outer coordinate index */
 373         inr              = iinr[iidx];
 374         i_coord_offset   = DIM*inr;
 375
 376         /* Load i particle coords and add shift vector */
 377         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 378
 379         fix0             = _mm_setzero_pd();
 380         fiy0             = _mm_setzero_pd();
 381         fiz0             = _mm_setzero_pd();
 382
 383         /* Load parameters for i particles */
 384         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 385
 386         /* Start inner kernel loop */
 387         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 388         {
 389
 390             /* Get j neighbor index, and coordinate index */
 391             jnrA             = jjnr[jidx];
 392             jnrB             = jjnr[jidx+1];
 393             j_coord_offsetA  = DIM*jnrA;
 394             j_coord_offsetB  = DIM*jnrB;
 395
 396             /* load j atom coordinates */
 397             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 398                                               &jx0,&jy0,&jz0);
 399
 400             /* Calculate displacement vector */
 401             dx00             = _mm_sub_pd(ix0,jx0);
 402             dy00             = _mm_sub_pd(iy0,jy0);
 403             dz00             = _mm_sub_pd(iz0,jz0);
 404
 405             /* Calculate squared distance and things based on it */
 406             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 407
 408             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 409
 410             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 411
 412             /* Load parameters for j particles */
 413             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 414
 415             /**************************
 416              * CALCULATE INTERACTIONS *
 417              **************************/
 418
 419             if (gmx_mm_any_lt(rsq00,rcutoff2))
 420             {
 421
 422             /* Compute parameters for interactions between i and j atoms */
 423             qq00             = _mm_mul_pd(iq0,jq0);
 424
 425             /* REACTION-FIELD ELECTROSTATICS */
 426             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 427
 428             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 429
 430             fscal            = felec;
 431
 432             fscal            = _mm_and_pd(fscal,cutoff_mask);
 433
 434             /* Update vectorial force */
 435             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 436             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 437             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 438
 439             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 440                                                    _mm_mul_pd(dx00,fscal),
 441                                                    _mm_mul_pd(dy00,fscal),
 442                                                    _mm_mul_pd(dz00,fscal));
 443
 444             }
 445
 446             /* Inner loop uses 33 flops */
 447         }
 448
 449         if(jidx<j_index_end)
 450         {
 451
 452             jnrA             = jjnr[jidx];
 453             j_coord_offsetA  = DIM*jnrA;
 454
 455             /* load j atom coordinates */
 456             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 457                                               &jx0,&jy0,&jz0);
 458
 459             /* Calculate displacement vector */
 460             dx00             = _mm_sub_pd(ix0,jx0);
 461             dy00             = _mm_sub_pd(iy0,jy0);
 462             dz00             = _mm_sub_pd(iz0,jz0);
 463
 464             /* Calculate squared distance and things based on it */
 465             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 466
 467             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 468
 469             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 470
 471             /* Load parameters for j particles */
 472             jq0              = _mm_load_sd(charge+jnrA+0);
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             if (gmx_mm_any_lt(rsq00,rcutoff2))
 479             {
 480
 481             /* Compute parameters for interactions between i and j atoms */
 482             qq00             = _mm_mul_pd(iq0,jq0);
 483
 484             /* REACTION-FIELD ELECTROSTATICS */
 485             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 486
 487             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 488
 489             fscal            = felec;
 490
 491             fscal            = _mm_and_pd(fscal,cutoff_mask);
 492
 493             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 494
 495             /* Update vectorial force */
 496             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 497             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 498             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 499
 500             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 501                                                    _mm_mul_pd(dx00,fscal),
 502                                                    _mm_mul_pd(dy00,fscal),
 503                                                    _mm_mul_pd(dz00,fscal));
 504
 505             }
 506
 507             /* Inner loop uses 33 flops */
 508         }
 509
 510         /* End of innermost loop */
 511
 512         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 513                                               f+i_coord_offset,fshift+i_shift_offset);
 514
 515         /* Increment number of inner iterations */
 516         inneriter                  += j_index_end - j_index_start;
 517
 518         /* Outer loop uses 7 flops */
 519     }
 520
 521     /* Increment number of outer iterations */
 522     outeriter        += nri;
 523
 524     /* Update outer/inner flops */
 525
 526     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*33);
 527 }