src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_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_ElecEwSw_VdwNone_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_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     __m128i          ewitab;
  74     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  75     real             *ewtab;
  76     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  77     real             rswitch_scalar,d_scalar;
  78     __m128d          dummy_mask,cutoff_mask;
  79     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  80     __m128d          one     = _mm_set1_pd(1.0);
  81     __m128d          two     = _mm_set1_pd(2.0);
  82     x                = xx[0];
  83     f                = ff[0];
  84
  85     nri              = nlist->nri;
  86     iinr             = nlist->iinr;
  87     jindex           = nlist->jindex;
  88     jjnr             = nlist->jjnr;
  89     shiftidx         = nlist->shift;
  90     gid              = nlist->gid;
  91     shiftvec         = fr->shift_vec[0];
  92     fshift           = fr->fshift[0];
  93     facel            = _mm_set1_pd(fr->epsfac);
  94     charge           = mdatoms->chargeA;
  95
  96     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
  97     ewtab            = fr->ic->tabq_coul_FDV0;
  98     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
  99     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 100
 101     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 102     rcutoff_scalar   = fr->rcoulomb;
 103     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 104     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 105
 106     rswitch_scalar   = fr->rcoulomb_switch;
 107     rswitch          = _mm_set1_pd(rswitch_scalar);
 108     /* Setup switch parameters */
 109     d_scalar         = rcutoff_scalar-rswitch_scalar;
 110     d                = _mm_set1_pd(d_scalar);
 111     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 112     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 113     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 114     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 115     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 116     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _mm_setzero_pd();
 144         fiy0             = _mm_setzero_pd();
 145         fiz0             = _mm_setzero_pd();
 146
 147         /* Load parameters for i particles */
 148         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm_setzero_pd();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162
 163             /* load j atom coordinates */
 164             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_pd(ix0,jx0);
 169             dy00             = _mm_sub_pd(iy0,jy0);
 170             dz00             = _mm_sub_pd(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 174
 175             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 176
 177             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 178
 179             /* Load parameters for j particles */
 180             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 181
 182             /**************************
 183              * CALCULATE INTERACTIONS *
 184              **************************/
 185
 186             if (gmx_mm_any_lt(rsq00,rcutoff2))
 187             {
 188
 189             r00              = _mm_mul_pd(rsq00,rinv00);
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm_mul_pd(iq0,jq0);
 193
 194             /* EWALD ELECTROSTATICS */
 195
 196             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 197             ewrt             = _mm_mul_pd(r00,ewtabscale);
 198             ewitab           = _mm_cvttpd_epi32(ewrt);
 199 #ifdef __XOP__
 200             eweps            = _mm_frcz_pd(ewrt);
 201 #else
 202             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 203 #endif
 204             twoeweps         = _mm_add_pd(eweps,eweps);
 205             ewitab           = _mm_slli_epi32(ewitab,2);
 206             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 207             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 208             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 209             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 210             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 211             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 212             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 213             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 214             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 215             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 216
 217             d                = _mm_sub_pd(r00,rswitch);
 218             d                = _mm_max_pd(d,_mm_setzero_pd());
 219             d2               = _mm_mul_pd(d,d);
 220             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 221
 222             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 223
 224             /* Evaluate switch function */
 225             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 226             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 227             velec            = _mm_mul_pd(velec,sw);
 228             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 229
 230             /* Update potential sum for this i atom from the interaction with this j atom. */
 231             velec            = _mm_and_pd(velec,cutoff_mask);
 232             velecsum         = _mm_add_pd(velecsum,velec);
 233
 234             fscal            = felec;
 235
 236             fscal            = _mm_and_pd(fscal,cutoff_mask);
 237
 238             /* Update vectorial force */
 239             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 240             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 241             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 242
 243             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 244                                                    _mm_mul_pd(dx00,fscal),
 245                                                    _mm_mul_pd(dy00,fscal),
 246                                                    _mm_mul_pd(dz00,fscal));
 247
 248             }
 249
 250             /* Inner loop uses 68 flops */
 251         }
 252
 253         if(jidx<j_index_end)
 254         {
 255
 256             jnrA             = jjnr[jidx];
 257             j_coord_offsetA  = DIM*jnrA;
 258
 259             /* load j atom coordinates */
 260             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 261                                               &jx0,&jy0,&jz0);
 262
 263             /* Calculate displacement vector */
 264             dx00             = _mm_sub_pd(ix0,jx0);
 265             dy00             = _mm_sub_pd(iy0,jy0);
 266             dz00             = _mm_sub_pd(iz0,jz0);
 267
 268             /* Calculate squared distance and things based on it */
 269             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 270
 271             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 272
 273             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 274
 275             /* Load parameters for j particles */
 276             jq0              = _mm_load_sd(charge+jnrA+0);
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             if (gmx_mm_any_lt(rsq00,rcutoff2))
 283             {
 284
 285             r00              = _mm_mul_pd(rsq00,rinv00);
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq00             = _mm_mul_pd(iq0,jq0);
 289
 290             /* EWALD ELECTROSTATICS */
 291
 292             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 293             ewrt             = _mm_mul_pd(r00,ewtabscale);
 294             ewitab           = _mm_cvttpd_epi32(ewrt);
 295 #ifdef __XOP__
 296             eweps            = _mm_frcz_pd(ewrt);
 297 #else
 298             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 299 #endif
 300             twoeweps         = _mm_add_pd(eweps,eweps);
 301             ewitab           = _mm_slli_epi32(ewitab,2);
 302             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 303             ewtabD           = _mm_setzero_pd();
 304             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 305             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 306             ewtabFn          = _mm_setzero_pd();
 307             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 308             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 309             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 310             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 311             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 312
 313             d                = _mm_sub_pd(r00,rswitch);
 314             d                = _mm_max_pd(d,_mm_setzero_pd());
 315             d2               = _mm_mul_pd(d,d);
 316             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 317
 318             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 319
 320             /* Evaluate switch function */
 321             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 322             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 323             velec            = _mm_mul_pd(velec,sw);
 324             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velec            = _mm_and_pd(velec,cutoff_mask);
 328             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 329             velecsum         = _mm_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             fscal            = _mm_and_pd(fscal,cutoff_mask);
 334
 335             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 336
 337             /* Update vectorial force */
 338             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 339             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 340             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 341
 342             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 343                                                    _mm_mul_pd(dx00,fscal),
 344                                                    _mm_mul_pd(dy00,fscal),
 345                                                    _mm_mul_pd(dz00,fscal));
 346
 347             }
 348
 349             /* Inner loop uses 68 flops */
 350         }
 351
 352         /* End of innermost loop */
 353
 354         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 355                                               f+i_coord_offset,fshift+i_shift_offset);
 356
 357         ggid                        = gid[iidx];
 358         /* Update potential energies */
 359         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 360
 361         /* Increment number of inner iterations */
 362         inneriter                  += j_index_end - j_index_start;
 363
 364         /* Outer loop uses 8 flops */
 365     }
 366
 367     /* Increment number of outer iterations */
 368     outeriter        += nri;
 369
 370     /* Update outer/inner flops */
 371
 372     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*68);
 373 }
 374 /*
 375  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_double
 376  * Electrostatics interaction: Ewald
 377  * VdW interaction:            None
 378  * Geometry:                   Particle-Particle
 379  * Calculate force/pot:        Force
 380  */
 381 void
 382 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_128_fma_double
 383                     (t_nblist * gmx_restrict                nlist,
 384                      rvec * gmx_restrict                    xx,
 385                      rvec * gmx_restrict                    ff,
 386                      t_forcerec * gmx_restrict              fr,
 387                      t_mdatoms * gmx_restrict               mdatoms,
 388                      nb_kernel_data_t * gmx_restrict        kernel_data,
 389                      t_nrnb * gmx_restrict                  nrnb)
 390 {
 391     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 392      * just 0 for non-waters.
 393      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 394      * jnr indices corresponding to data put in the four positions in the SIMD register.
 395      */
 396     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 397     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 398     int              jnrA,jnrB;
 399     int              j_coord_offsetA,j_coord_offsetB;
 400     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 401     real             rcutoff_scalar;
 402     real             *shiftvec,*fshift,*x,*f;
 403     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 404     int              vdwioffset0;
 405     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 406     int              vdwjidx0A,vdwjidx0B;
 407     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 408     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 409     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 410     real             *charge;
 411     __m128i          ewitab;
 412     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 413     real             *ewtab;
 414     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 415     real             rswitch_scalar,d_scalar;
 416     __m128d          dummy_mask,cutoff_mask;
 417     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 418     __m128d          one     = _mm_set1_pd(1.0);
 419     __m128d          two     = _mm_set1_pd(2.0);
 420     x                = xx[0];
 421     f                = ff[0];
 422
 423     nri              = nlist->nri;
 424     iinr             = nlist->iinr;
 425     jindex           = nlist->jindex;
 426     jjnr             = nlist->jjnr;
 427     shiftidx         = nlist->shift;
 428     gid              = nlist->gid;
 429     shiftvec         = fr->shift_vec[0];
 430     fshift           = fr->fshift[0];
 431     facel            = _mm_set1_pd(fr->epsfac);
 432     charge           = mdatoms->chargeA;
 433
 434     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 435     ewtab            = fr->ic->tabq_coul_FDV0;
 436     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 437     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 438
 439     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 440     rcutoff_scalar   = fr->rcoulomb;
 441     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 442     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 443
 444     rswitch_scalar   = fr->rcoulomb_switch;
 445     rswitch          = _mm_set1_pd(rswitch_scalar);
 446     /* Setup switch parameters */
 447     d_scalar         = rcutoff_scalar-rswitch_scalar;
 448     d                = _mm_set1_pd(d_scalar);
 449     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 450     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 451     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 452     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 453     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 454     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 455
 456     /* Avoid stupid compiler warnings */
 457     jnrA = jnrB = 0;
 458     j_coord_offsetA = 0;
 459     j_coord_offsetB = 0;
 460
 461     outeriter        = 0;
 462     inneriter        = 0;
 463
 464     /* Start outer loop over neighborlists */
 465     for(iidx=0; iidx<nri; iidx++)
 466     {
 467         /* Load shift vector for this list */
 468         i_shift_offset   = DIM*shiftidx[iidx];
 469
 470         /* Load limits for loop over neighbors */
 471         j_index_start    = jindex[iidx];
 472         j_index_end      = jindex[iidx+1];
 473
 474         /* Get outer coordinate index */
 475         inr              = iinr[iidx];
 476         i_coord_offset   = DIM*inr;
 477
 478         /* Load i particle coords and add shift vector */
 479         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 480
 481         fix0             = _mm_setzero_pd();
 482         fiy0             = _mm_setzero_pd();
 483         fiz0             = _mm_setzero_pd();
 484
 485         /* Load parameters for i particles */
 486         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 487
 488         /* Start inner kernel loop */
 489         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 490         {
 491
 492             /* Get j neighbor index, and coordinate index */
 493             jnrA             = jjnr[jidx];
 494             jnrB             = jjnr[jidx+1];
 495             j_coord_offsetA  = DIM*jnrA;
 496             j_coord_offsetB  = DIM*jnrB;
 497
 498             /* load j atom coordinates */
 499             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 500                                               &jx0,&jy0,&jz0);
 501
 502             /* Calculate displacement vector */
 503             dx00             = _mm_sub_pd(ix0,jx0);
 504             dy00             = _mm_sub_pd(iy0,jy0);
 505             dz00             = _mm_sub_pd(iz0,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 509
 510             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 511
 512             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 513
 514             /* Load parameters for j particles */
 515             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             if (gmx_mm_any_lt(rsq00,rcutoff2))
 522             {
 523
 524             r00              = _mm_mul_pd(rsq00,rinv00);
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             qq00             = _mm_mul_pd(iq0,jq0);
 528
 529             /* EWALD ELECTROSTATICS */
 530
 531             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 532             ewrt             = _mm_mul_pd(r00,ewtabscale);
 533             ewitab           = _mm_cvttpd_epi32(ewrt);
 534 #ifdef __XOP__
 535             eweps            = _mm_frcz_pd(ewrt);
 536 #else
 537             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 538 #endif
 539             twoeweps         = _mm_add_pd(eweps,eweps);
 540             ewitab           = _mm_slli_epi32(ewitab,2);
 541             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 542             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 543             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 544             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 545             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 546             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 547             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 548             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 549             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 550             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 551
 552             d                = _mm_sub_pd(r00,rswitch);
 553             d                = _mm_max_pd(d,_mm_setzero_pd());
 554             d2               = _mm_mul_pd(d,d);
 555             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 556
 557             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 558
 559             /* Evaluate switch function */
 560             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 561             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 562             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm_and_pd(fscal,cutoff_mask);
 567
 568             /* Update vectorial force */
 569             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 570             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 571             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 572
 573             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 574                                                    _mm_mul_pd(dx00,fscal),
 575                                                    _mm_mul_pd(dy00,fscal),
 576                                                    _mm_mul_pd(dz00,fscal));
 577
 578             }
 579
 580             /* Inner loop uses 65 flops */
 581         }
 582
 583         if(jidx<j_index_end)
 584         {
 585
 586             jnrA             = jjnr[jidx];
 587             j_coord_offsetA  = DIM*jnrA;
 588
 589             /* load j atom coordinates */
 590             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 591                                               &jx0,&jy0,&jz0);
 592
 593             /* Calculate displacement vector */
 594             dx00             = _mm_sub_pd(ix0,jx0);
 595             dy00             = _mm_sub_pd(iy0,jy0);
 596             dz00             = _mm_sub_pd(iz0,jz0);
 597
 598             /* Calculate squared distance and things based on it */
 599             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 600
 601             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 602
 603             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 604
 605             /* Load parameters for j particles */
 606             jq0              = _mm_load_sd(charge+jnrA+0);
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             if (gmx_mm_any_lt(rsq00,rcutoff2))
 613             {
 614
 615             r00              = _mm_mul_pd(rsq00,rinv00);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq00             = _mm_mul_pd(iq0,jq0);
 619
 620             /* EWALD ELECTROSTATICS */
 621
 622             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 623             ewrt             = _mm_mul_pd(r00,ewtabscale);
 624             ewitab           = _mm_cvttpd_epi32(ewrt);
 625 #ifdef __XOP__
 626             eweps            = _mm_frcz_pd(ewrt);
 627 #else
 628             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 629 #endif
 630             twoeweps         = _mm_add_pd(eweps,eweps);
 631             ewitab           = _mm_slli_epi32(ewitab,2);
 632             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 633             ewtabD           = _mm_setzero_pd();
 634             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 635             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 636             ewtabFn          = _mm_setzero_pd();
 637             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 638             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 639             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 640             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 641             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 642
 643             d                = _mm_sub_pd(r00,rswitch);
 644             d                = _mm_max_pd(d,_mm_setzero_pd());
 645             d2               = _mm_mul_pd(d,d);
 646             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 647
 648             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 649
 650             /* Evaluate switch function */
 651             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 652             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 653             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm_and_pd(fscal,cutoff_mask);
 658
 659             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 660
 661             /* Update vectorial force */
 662             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 663             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 664             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 665
 666             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 667                                                    _mm_mul_pd(dx00,fscal),
 668                                                    _mm_mul_pd(dy00,fscal),
 669                                                    _mm_mul_pd(dz00,fscal));
 670
 671             }
 672
 673             /* Inner loop uses 65 flops */
 674         }
 675
 676         /* End of innermost loop */
 677
 678         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 679                                               f+i_coord_offset,fshift+i_shift_offset);
 680
 681         /* Increment number of inner iterations */
 682         inneriter                  += j_index_end - j_index_start;
 683
 684         /* Outer loop uses 7 flops */
 685     }
 686
 687     /* Increment number of outer iterations */
 688     outeriter        += nri;
 689
 690     /* Update outer/inner flops */
 691
 692     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*65);
 693 }