src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecNone_VdwLJ_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_ElecNone_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: None
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecNone_VdwLJ_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     int              nvdwtype;
  72     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  73     int              *vdwtype;
  74     real             *vdwparam;
  75     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  76     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  77     __m128d          dummy_mask,cutoff_mask;
  78     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  79     __m128d          one     = _mm_set1_pd(1.0);
  80     __m128d          two     = _mm_set1_pd(2.0);
  81     x                = xx[0];
  82     f                = ff[0];
  83
  84     nri              = nlist->nri;
  85     iinr             = nlist->iinr;
  86     jindex           = nlist->jindex;
  87     jjnr             = nlist->jjnr;
  88     shiftidx         = nlist->shift;
  89     gid              = nlist->gid;
  90     shiftvec         = fr->shift_vec[0];
  91     fshift           = fr->fshift[0];
  92     nvdwtype         = fr->ntype;
  93     vdwparam         = fr->nbfp;
  94     vdwtype          = mdatoms->typeA;
  95
  96     /* Avoid stupid compiler warnings */
  97     jnrA = jnrB = 0;
  98     j_coord_offsetA = 0;
  99     j_coord_offsetB = 0;
 100
 101     outeriter        = 0;
 102     inneriter        = 0;
 103
 104     /* Start outer loop over neighborlists */
 105     for(iidx=0; iidx<nri; iidx++)
 106     {
 107         /* Load shift vector for this list */
 108         i_shift_offset   = DIM*shiftidx[iidx];
 109
 110         /* Load limits for loop over neighbors */
 111         j_index_start    = jindex[iidx];
 112         j_index_end      = jindex[iidx+1];
 113
 114         /* Get outer coordinate index */
 115         inr              = iinr[iidx];
 116         i_coord_offset   = DIM*inr;
 117
 118         /* Load i particle coords and add shift vector */
 119         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 120
 121         fix0             = _mm_setzero_pd();
 122         fiy0             = _mm_setzero_pd();
 123         fiz0             = _mm_setzero_pd();
 124
 125         /* Load parameters for i particles */
 126         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 127
 128         /* Reset potential sums */
 129         vvdwsum          = _mm_setzero_pd();
 130
 131         /* Start inner kernel loop */
 132         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 133         {
 134
 135             /* Get j neighbor index, and coordinate index */
 136             jnrA             = jjnr[jidx];
 137             jnrB             = jjnr[jidx+1];
 138             j_coord_offsetA  = DIM*jnrA;
 139             j_coord_offsetB  = DIM*jnrB;
 140
 141             /* load j atom coordinates */
 142             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 143                                               &jx0,&jy0,&jz0);
 144
 145             /* Calculate displacement vector */
 146             dx00             = _mm_sub_pd(ix0,jx0);
 147             dy00             = _mm_sub_pd(iy0,jy0);
 148             dz00             = _mm_sub_pd(iz0,jz0);
 149
 150             /* Calculate squared distance and things based on it */
 151             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 152
 153             rinvsq00         = gmx_mm_inv_pd(rsq00);
 154
 155             /* Load parameters for j particles */
 156             vdwjidx0A        = 2*vdwtype[jnrA+0];
 157             vdwjidx0B        = 2*vdwtype[jnrB+0];
 158
 159             /**************************
 160              * CALCULATE INTERACTIONS *
 161              **************************/
 162
 163             /* Compute parameters for interactions between i and j atoms */
 164             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 165                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 166
 167             /* LENNARD-JONES DISPERSION/REPULSION */
 168
 169             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 170             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 171             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 172             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 173             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 174
 175             /* Update potential sum for this i atom from the interaction with this j atom. */
 176             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 177
 178             fscal            = fvdw;
 179
 180             /* Update vectorial force */
 181             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 182             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 183             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 184
 185             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 186                                                    _mm_mul_pd(dx00,fscal),
 187                                                    _mm_mul_pd(dy00,fscal),
 188                                                    _mm_mul_pd(dz00,fscal));
 189
 190             /* Inner loop uses 35 flops */
 191         }
 192
 193         if(jidx<j_index_end)
 194         {
 195
 196             jnrA             = jjnr[jidx];
 197             j_coord_offsetA  = DIM*jnrA;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_pd(ix0,jx0);
 205             dy00             = _mm_sub_pd(iy0,jy0);
 206             dz00             = _mm_sub_pd(iz0,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 210
 211             rinvsq00         = gmx_mm_inv_pd(rsq00);
 212
 213             /* Load parameters for j particles */
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215
 216             /**************************
 217              * CALCULATE INTERACTIONS *
 218              **************************/
 219
 220             /* Compute parameters for interactions between i and j atoms */
 221             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 222
 223             /* LENNARD-JONES DISPERSION/REPULSION */
 224
 225             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 226             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 227             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 228             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 229             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 233             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 234
 235             fscal            = fvdw;
 236
 237             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 238
 239             /* Update vectorial force */
 240             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 241             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 242             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 243
 244             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 245                                                    _mm_mul_pd(dx00,fscal),
 246                                                    _mm_mul_pd(dy00,fscal),
 247                                                    _mm_mul_pd(dz00,fscal));
 248
 249             /* Inner loop uses 35 flops */
 250         }
 251
 252         /* End of innermost loop */
 253
 254         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 255                                               f+i_coord_offset,fshift+i_shift_offset);
 256
 257         ggid                        = gid[iidx];
 258         /* Update potential energies */
 259         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 260
 261         /* Increment number of inner iterations */
 262         inneriter                  += j_index_end - j_index_start;
 263
 264         /* Outer loop uses 7 flops */
 265     }
 266
 267     /* Increment number of outer iterations */
 268     outeriter        += nri;
 269
 270     /* Update outer/inner flops */
 271
 272     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*35);
 273 }
 274 /*
 275  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_double
 276  * Electrostatics interaction: None
 277  * VdW interaction:            LennardJones
 278  * Geometry:                   Particle-Particle
 279  * Calculate force/pot:        Force
 280  */
 281 void
 282 nb_kernel_ElecNone_VdwLJ_GeomP1P1_F_avx_128_fma_double
 283                     (t_nblist * gmx_restrict                nlist,
 284                      rvec * gmx_restrict                    xx,
 285                      rvec * gmx_restrict                    ff,
 286                      t_forcerec * gmx_restrict              fr,
 287                      t_mdatoms * gmx_restrict               mdatoms,
 288                      nb_kernel_data_t * gmx_restrict        kernel_data,
 289                      t_nrnb * gmx_restrict                  nrnb)
 290 {
 291     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 292      * just 0 for non-waters.
 293      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 294      * jnr indices corresponding to data put in the four positions in the SIMD register.
 295      */
 296     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 297     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 298     int              jnrA,jnrB;
 299     int              j_coord_offsetA,j_coord_offsetB;
 300     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 301     real             rcutoff_scalar;
 302     real             *shiftvec,*fshift,*x,*f;
 303     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 304     int              vdwioffset0;
 305     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 306     int              vdwjidx0A,vdwjidx0B;
 307     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 308     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 309     int              nvdwtype;
 310     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 311     int              *vdwtype;
 312     real             *vdwparam;
 313     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 314     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 315     __m128d          dummy_mask,cutoff_mask;
 316     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 317     __m128d          one     = _mm_set1_pd(1.0);
 318     __m128d          two     = _mm_set1_pd(2.0);
 319     x                = xx[0];
 320     f                = ff[0];
 321
 322     nri              = nlist->nri;
 323     iinr             = nlist->iinr;
 324     jindex           = nlist->jindex;
 325     jjnr             = nlist->jjnr;
 326     shiftidx         = nlist->shift;
 327     gid              = nlist->gid;
 328     shiftvec         = fr->shift_vec[0];
 329     fshift           = fr->fshift[0];
 330     nvdwtype         = fr->ntype;
 331     vdwparam         = fr->nbfp;
 332     vdwtype          = mdatoms->typeA;
 333
 334     /* Avoid stupid compiler warnings */
 335     jnrA = jnrB = 0;
 336     j_coord_offsetA = 0;
 337     j_coord_offsetB = 0;
 338
 339     outeriter        = 0;
 340     inneriter        = 0;
 341
 342     /* Start outer loop over neighborlists */
 343     for(iidx=0; iidx<nri; iidx++)
 344     {
 345         /* Load shift vector for this list */
 346         i_shift_offset   = DIM*shiftidx[iidx];
 347
 348         /* Load limits for loop over neighbors */
 349         j_index_start    = jindex[iidx];
 350         j_index_end      = jindex[iidx+1];
 351
 352         /* Get outer coordinate index */
 353         inr              = iinr[iidx];
 354         i_coord_offset   = DIM*inr;
 355
 356         /* Load i particle coords and add shift vector */
 357         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 358
 359         fix0             = _mm_setzero_pd();
 360         fiy0             = _mm_setzero_pd();
 361         fiz0             = _mm_setzero_pd();
 362
 363         /* Load parameters for i particles */
 364         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 365
 366         /* Start inner kernel loop */
 367         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 368         {
 369
 370             /* Get j neighbor index, and coordinate index */
 371             jnrA             = jjnr[jidx];
 372             jnrB             = jjnr[jidx+1];
 373             j_coord_offsetA  = DIM*jnrA;
 374             j_coord_offsetB  = DIM*jnrB;
 375
 376             /* load j atom coordinates */
 377             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 378                                               &jx0,&jy0,&jz0);
 379
 380             /* Calculate displacement vector */
 381             dx00             = _mm_sub_pd(ix0,jx0);
 382             dy00             = _mm_sub_pd(iy0,jy0);
 383             dz00             = _mm_sub_pd(iz0,jz0);
 384
 385             /* Calculate squared distance and things based on it */
 386             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 387
 388             rinvsq00         = gmx_mm_inv_pd(rsq00);
 389
 390             /* Load parameters for j particles */
 391             vdwjidx0A        = 2*vdwtype[jnrA+0];
 392             vdwjidx0B        = 2*vdwtype[jnrB+0];
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             /* Compute parameters for interactions between i and j atoms */
 399             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 400                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 401
 402             /* LENNARD-JONES DISPERSION/REPULSION */
 403
 404             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 405             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 406
 407             fscal            = fvdw;
 408
 409             /* Update vectorial force */
 410             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 411             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 412             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 413
 414             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 415                                                    _mm_mul_pd(dx00,fscal),
 416                                                    _mm_mul_pd(dy00,fscal),
 417                                                    _mm_mul_pd(dz00,fscal));
 418
 419             /* Inner loop uses 30 flops */
 420         }
 421
 422         if(jidx<j_index_end)
 423         {
 424
 425             jnrA             = jjnr[jidx];
 426             j_coord_offsetA  = DIM*jnrA;
 427
 428             /* load j atom coordinates */
 429             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 430                                               &jx0,&jy0,&jz0);
 431
 432             /* Calculate displacement vector */
 433             dx00             = _mm_sub_pd(ix0,jx0);
 434             dy00             = _mm_sub_pd(iy0,jy0);
 435             dz00             = _mm_sub_pd(iz0,jz0);
 436
 437             /* Calculate squared distance and things based on it */
 438             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 439
 440             rinvsq00         = gmx_mm_inv_pd(rsq00);
 441
 442             /* Load parameters for j particles */
 443             vdwjidx0A        = 2*vdwtype[jnrA+0];
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             /* Compute parameters for interactions between i and j atoms */
 450             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 451
 452             /* LENNARD-JONES DISPERSION/REPULSION */
 453
 454             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 455             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 456
 457             fscal            = fvdw;
 458
 459             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 460
 461             /* Update vectorial force */
 462             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 463             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 464             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 465
 466             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 467                                                    _mm_mul_pd(dx00,fscal),
 468                                                    _mm_mul_pd(dy00,fscal),
 469                                                    _mm_mul_pd(dz00,fscal));
 470
 471             /* Inner loop uses 30 flops */
 472         }
 473
 474         /* End of innermost loop */
 475
 476         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 477                                               f+i_coord_offset,fshift+i_shift_offset);
 478
 479         /* Increment number of inner iterations */
 480         inneriter                  += j_index_end - j_index_start;
 481
 482         /* Outer loop uses 6 flops */
 483     }
 484
 485     /* Increment number of outer iterations */
 486     outeriter        += nri;
 487
 488     /* Update outer/inner flops */
 489
 490     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*30);
 491 }