src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single 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_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_128_fma_single
  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,C,D refer to j loop unrolling done with AVX_128, e.g. for the four 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,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          ewitab;
  92     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  93     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  94     real             *ewtab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 117     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 118     beta2            = _mm_mul_ps(beta,beta);
 119     beta3            = _mm_mul_ps(beta,beta2);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 127     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 128     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 129     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 130
 131     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 132     rcutoff_scalar   = fr->rcoulomb;
 133     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 134     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 135
 136     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 137     rvdw             = _mm_set1_ps(fr->rvdw);
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181         fix3             = _mm_setzero_ps();
 182         fiy3             = _mm_setzero_ps();
 183         fiz3             = _mm_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_ps();
 187         vvdwsum          = _mm_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               x+j_coord_offsetC,x+j_coord_offsetD,
 206                                               &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_ps(ix0,jx0);
 210             dy00             = _mm_sub_ps(iy0,jy0);
 211             dz00             = _mm_sub_ps(iz0,jz0);
 212             dx10             = _mm_sub_ps(ix1,jx0);
 213             dy10             = _mm_sub_ps(iy1,jy0);
 214             dz10             = _mm_sub_ps(iz1,jz0);
 215             dx20             = _mm_sub_ps(ix2,jx0);
 216             dy20             = _mm_sub_ps(iy2,jy0);
 217             dz20             = _mm_sub_ps(iz2,jz0);
 218             dx30             = _mm_sub_ps(ix3,jx0);
 219             dy30             = _mm_sub_ps(iy3,jy0);
 220             dz30             = _mm_sub_ps(iz3,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 227
 228             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 229             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 230             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 231
 232             rinvsq00         = gmx_mm_inv_ps(rsq00);
 233             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 234             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 235             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 239                                                               charge+jnrC+0,charge+jnrD+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244
 245             fjx0             = _mm_setzero_ps();
 246             fjy0             = _mm_setzero_ps();
 247             fjz0             = _mm_setzero_ps();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             if (gmx_mm_any_lt(rsq00,rcutoff2))
 254             {
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 258                                          vdwparam+vdwioffset0+vdwjidx0B,
 259                                          vdwparam+vdwioffset0+vdwjidx0C,
 260                                          vdwparam+vdwioffset0+vdwjidx0D,
 261                                          &c6_00,&c12_00);
 262
 263             /* LENNARD-JONES DISPERSION/REPULSION */
 264
 265             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 266             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 267             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 268             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 269                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 270             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 271
 272             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 276             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 277
 278             fscal            = fvdw;
 279
 280             fscal            = _mm_and_ps(fscal,cutoff_mask);
 281
 282              /* Update vectorial force */
 283             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 284             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 285             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 286
 287             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 288             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 289             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 290
 291             }
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             if (gmx_mm_any_lt(rsq10,rcutoff2))
 298             {
 299
 300             r10              = _mm_mul_ps(rsq10,rinv10);
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq10             = _mm_mul_ps(iq1,jq0);
 304
 305             /* EWALD ELECTROSTATICS */
 306
 307             /* Analytical PME correction */
 308             zeta2            = _mm_mul_ps(beta2,rsq10);
 309             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 310             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 311             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 312             felec            = _mm_mul_ps(qq10,felec);
 313             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 314             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 315             velec            = _mm_mul_ps(qq10,velec);
 316
 317             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm_and_ps(velec,cutoff_mask);
 321             velecsum         = _mm_add_ps(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm_and_ps(fscal,cutoff_mask);
 326
 327              /* Update vectorial force */
 328             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 329             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 330             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 331
 332             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 333             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 334             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 335
 336             }
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             if (gmx_mm_any_lt(rsq20,rcutoff2))
 343             {
 344
 345             r20              = _mm_mul_ps(rsq20,rinv20);
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq20             = _mm_mul_ps(iq2,jq0);
 349
 350             /* EWALD ELECTROSTATICS */
 351
 352             /* Analytical PME correction */
 353             zeta2            = _mm_mul_ps(beta2,rsq20);
 354             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 355             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 356             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 357             felec            = _mm_mul_ps(qq20,felec);
 358             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 359             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 360             velec            = _mm_mul_ps(qq20,velec);
 361
 362             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velec            = _mm_and_ps(velec,cutoff_mask);
 366             velecsum         = _mm_add_ps(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             fscal            = _mm_and_ps(fscal,cutoff_mask);
 371
 372              /* Update vectorial force */
 373             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 374             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 375             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 376
 377             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 378             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 379             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 380
 381             }
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             if (gmx_mm_any_lt(rsq30,rcutoff2))
 388             {
 389
 390             r30              = _mm_mul_ps(rsq30,rinv30);
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq30             = _mm_mul_ps(iq3,jq0);
 394
 395             /* EWALD ELECTROSTATICS */
 396
 397             /* Analytical PME correction */
 398             zeta2            = _mm_mul_ps(beta2,rsq30);
 399             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 400             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 401             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 402             felec            = _mm_mul_ps(qq30,felec);
 403             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 404             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 405             velec            = _mm_mul_ps(qq30,velec);
 406
 407             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm_and_ps(velec,cutoff_mask);
 411             velecsum         = _mm_add_ps(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             fscal            = _mm_and_ps(fscal,cutoff_mask);
 416
 417              /* Update vectorial force */
 418             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 419             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 420             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 421
 422             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 423             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 424             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 425
 426             }
 427
 428             fjptrA             = f+j_coord_offsetA;
 429             fjptrB             = f+j_coord_offsetB;
 430             fjptrC             = f+j_coord_offsetC;
 431             fjptrD             = f+j_coord_offsetD;
 432
 433             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 434
 435             /* Inner loop uses 143 flops */
 436         }
 437
 438         if(jidx<j_index_end)
 439         {
 440
 441             /* Get j neighbor index, and coordinate index */
 442             jnrlistA         = jjnr[jidx];
 443             jnrlistB         = jjnr[jidx+1];
 444             jnrlistC         = jjnr[jidx+2];
 445             jnrlistD         = jjnr[jidx+3];
 446             /* Sign of each element will be negative for non-real atoms.
 447              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 448              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 449              */
 450             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 451             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 452             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 453             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 454             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 455             j_coord_offsetA  = DIM*jnrA;
 456             j_coord_offsetB  = DIM*jnrB;
 457             j_coord_offsetC  = DIM*jnrC;
 458             j_coord_offsetD  = DIM*jnrD;
 459
 460             /* load j atom coordinates */
 461             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 462                                               x+j_coord_offsetC,x+j_coord_offsetD,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_ps(ix0,jx0);
 467             dy00             = _mm_sub_ps(iy0,jy0);
 468             dz00             = _mm_sub_ps(iz0,jz0);
 469             dx10             = _mm_sub_ps(ix1,jx0);
 470             dy10             = _mm_sub_ps(iy1,jy0);
 471             dz10             = _mm_sub_ps(iz1,jz0);
 472             dx20             = _mm_sub_ps(ix2,jx0);
 473             dy20             = _mm_sub_ps(iy2,jy0);
 474             dz20             = _mm_sub_ps(iz2,jz0);
 475             dx30             = _mm_sub_ps(ix3,jx0);
 476             dy30             = _mm_sub_ps(iy3,jy0);
 477             dz30             = _mm_sub_ps(iz3,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 481             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 482             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 483             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 484
 485             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 486             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 487             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 488
 489             rinvsq00         = gmx_mm_inv_ps(rsq00);
 490             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 491             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 492             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 493
 494             /* Load parameters for j particles */
 495             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 496                                                               charge+jnrC+0,charge+jnrD+0);
 497             vdwjidx0A        = 2*vdwtype[jnrA+0];
 498             vdwjidx0B        = 2*vdwtype[jnrB+0];
 499             vdwjidx0C        = 2*vdwtype[jnrC+0];
 500             vdwjidx0D        = 2*vdwtype[jnrD+0];
 501
 502             fjx0             = _mm_setzero_ps();
 503             fjy0             = _mm_setzero_ps();
 504             fjz0             = _mm_setzero_ps();
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             if (gmx_mm_any_lt(rsq00,rcutoff2))
 511             {
 512
 513             /* Compute parameters for interactions between i and j atoms */
 514             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 515                                          vdwparam+vdwioffset0+vdwjidx0B,
 516                                          vdwparam+vdwioffset0+vdwjidx0C,
 517                                          vdwparam+vdwioffset0+vdwjidx0D,
 518                                          &c6_00,&c12_00);
 519
 520             /* LENNARD-JONES DISPERSION/REPULSION */
 521
 522             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 523             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 524             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 525             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 526                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 527             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 528
 529             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 530
 531             /* Update potential sum for this i atom from the interaction with this j atom. */
 532             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 533             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 534             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 535
 536             fscal            = fvdw;
 537
 538             fscal            = _mm_and_ps(fscal,cutoff_mask);
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542              /* Update vectorial force */
 543             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 544             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 545             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 546
 547             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 548             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 549             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 550
 551             }
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_mm_any_lt(rsq10,rcutoff2))
 558             {
 559
 560             r10              = _mm_mul_ps(rsq10,rinv10);
 561             r10              = _mm_andnot_ps(dummy_mask,r10);
 562
 563             /* Compute parameters for interactions between i and j atoms */
 564             qq10             = _mm_mul_ps(iq1,jq0);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Analytical PME correction */
 569             zeta2            = _mm_mul_ps(beta2,rsq10);
 570             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 571             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 572             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 573             felec            = _mm_mul_ps(qq10,felec);
 574             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 575             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 576             velec            = _mm_mul_ps(qq10,velec);
 577
 578             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velec            = _mm_and_ps(velec,cutoff_mask);
 582             velec            = _mm_andnot_ps(dummy_mask,velec);
 583             velecsum         = _mm_add_ps(velecsum,velec);
 584
 585             fscal            = felec;
 586
 587             fscal            = _mm_and_ps(fscal,cutoff_mask);
 588
 589             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 590
 591              /* Update vectorial force */
 592             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 593             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 594             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 595
 596             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 597             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 598             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 599
 600             }
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             if (gmx_mm_any_lt(rsq20,rcutoff2))
 607             {
 608
 609             r20              = _mm_mul_ps(rsq20,rinv20);
 610             r20              = _mm_andnot_ps(dummy_mask,r20);
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq20             = _mm_mul_ps(iq2,jq0);
 614
 615             /* EWALD ELECTROSTATICS */
 616
 617             /* Analytical PME correction */
 618             zeta2            = _mm_mul_ps(beta2,rsq20);
 619             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 620             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 621             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 622             felec            = _mm_mul_ps(qq20,felec);
 623             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 624             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 625             velec            = _mm_mul_ps(qq20,velec);
 626
 627             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 628
 629             /* Update potential sum for this i atom from the interaction with this j atom. */
 630             velec            = _mm_and_ps(velec,cutoff_mask);
 631             velec            = _mm_andnot_ps(dummy_mask,velec);
 632             velecsum         = _mm_add_ps(velecsum,velec);
 633
 634             fscal            = felec;
 635
 636             fscal            = _mm_and_ps(fscal,cutoff_mask);
 637
 638             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 639
 640              /* Update vectorial force */
 641             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 642             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 643             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 644
 645             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 646             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 647             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 648
 649             }
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             if (gmx_mm_any_lt(rsq30,rcutoff2))
 656             {
 657
 658             r30              = _mm_mul_ps(rsq30,rinv30);
 659             r30              = _mm_andnot_ps(dummy_mask,r30);
 660
 661             /* Compute parameters for interactions between i and j atoms */
 662             qq30             = _mm_mul_ps(iq3,jq0);
 663
 664             /* EWALD ELECTROSTATICS */
 665
 666             /* Analytical PME correction */
 667             zeta2            = _mm_mul_ps(beta2,rsq30);
 668             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 669             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 670             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 671             felec            = _mm_mul_ps(qq30,felec);
 672             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 673             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 674             velec            = _mm_mul_ps(qq30,velec);
 675
 676             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 677
 678             /* Update potential sum for this i atom from the interaction with this j atom. */
 679             velec            = _mm_and_ps(velec,cutoff_mask);
 680             velec            = _mm_andnot_ps(dummy_mask,velec);
 681             velecsum         = _mm_add_ps(velecsum,velec);
 682
 683             fscal            = felec;
 684
 685             fscal            = _mm_and_ps(fscal,cutoff_mask);
 686
 687             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 688
 689              /* Update vectorial force */
 690             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 691             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 692             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 693
 694             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 695             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 696             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 697
 698             }
 699
 700             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 701             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 702             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 703             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 704
 705             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 706
 707             /* Inner loop uses 146 flops */
 708         }
 709
 710         /* End of innermost loop */
 711
 712         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 713                                               f+i_coord_offset,fshift+i_shift_offset);
 714
 715         ggid                        = gid[iidx];
 716         /* Update potential energies */
 717         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 718         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 719
 720         /* Increment number of inner iterations */
 721         inneriter                  += j_index_end - j_index_start;
 722
 723         /* Outer loop uses 26 flops */
 724     }
 725
 726     /* Increment number of outer iterations */
 727     outeriter        += nri;
 728
 729     /* Update outer/inner flops */
 730
 731     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*146);
 732 }
 733 /*
 734  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 735  * Electrostatics interaction: Ewald
 736  * VdW interaction:            LennardJones
 737  * Geometry:                   Water4-Particle
 738  * Calculate force/pot:        Force
 739  */
 740 void
 741 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 742                     (t_nblist * gmx_restrict                nlist,
 743                      rvec * gmx_restrict                    xx,
 744                      rvec * gmx_restrict                    ff,
 745                      t_forcerec * gmx_restrict              fr,
 746                      t_mdatoms * gmx_restrict               mdatoms,
 747                      nb_kernel_data_t * gmx_restrict        kernel_data,
 748                      t_nrnb * gmx_restrict                  nrnb)
 749 {
 750     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 751      * just 0 for non-waters.
 752      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 753      * jnr indices corresponding to data put in the four positions in the SIMD register.
 754      */
 755     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 756     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 757     int              jnrA,jnrB,jnrC,jnrD;
 758     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 759     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 760     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 761     real             rcutoff_scalar;
 762     real             *shiftvec,*fshift,*x,*f;
 763     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 764     real             scratch[4*DIM];
 765     __m128           fscal,rcutoff,rcutoff2,jidxall;
 766     int              vdwioffset0;
 767     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 768     int              vdwioffset1;
 769     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 770     int              vdwioffset2;
 771     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 772     int              vdwioffset3;
 773     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 774     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 775     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 776     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 777     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 778     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 779     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 780     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 781     real             *charge;
 782     int              nvdwtype;
 783     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 784     int              *vdwtype;
 785     real             *vdwparam;
 786     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 787     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 788     __m128i          ewitab;
 789     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 790     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 791     real             *ewtab;
 792     __m128           dummy_mask,cutoff_mask;
 793     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 794     __m128           one     = _mm_set1_ps(1.0);
 795     __m128           two     = _mm_set1_ps(2.0);
 796     x                = xx[0];
 797     f                = ff[0];
 798
 799     nri              = nlist->nri;
 800     iinr             = nlist->iinr;
 801     jindex           = nlist->jindex;
 802     jjnr             = nlist->jjnr;
 803     shiftidx         = nlist->shift;
 804     gid              = nlist->gid;
 805     shiftvec         = fr->shift_vec[0];
 806     fshift           = fr->fshift[0];
 807     facel            = _mm_set1_ps(fr->epsfac);
 808     charge           = mdatoms->chargeA;
 809     nvdwtype         = fr->ntype;
 810     vdwparam         = fr->nbfp;
 811     vdwtype          = mdatoms->typeA;
 812
 813     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 814     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 815     beta2            = _mm_mul_ps(beta,beta);
 816     beta3            = _mm_mul_ps(beta,beta2);
 817     ewtab            = fr->ic->tabq_coul_F;
 818     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 819     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 820
 821     /* Setup water-specific parameters */
 822     inr              = nlist->iinr[0];
 823     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 824     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 825     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 826     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 827
 828     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 829     rcutoff_scalar   = fr->rcoulomb;
 830     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 831     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 832
 833     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 834     rvdw             = _mm_set1_ps(fr->rvdw);
 835
 836     /* Avoid stupid compiler warnings */
 837     jnrA = jnrB = jnrC = jnrD = 0;
 838     j_coord_offsetA = 0;
 839     j_coord_offsetB = 0;
 840     j_coord_offsetC = 0;
 841     j_coord_offsetD = 0;
 842
 843     outeriter        = 0;
 844     inneriter        = 0;
 845
 846     for(iidx=0;iidx<4*DIM;iidx++)
 847     {
 848         scratch[iidx] = 0.0;
 849     }
 850
 851     /* Start outer loop over neighborlists */
 852     for(iidx=0; iidx<nri; iidx++)
 853     {
 854         /* Load shift vector for this list */
 855         i_shift_offset   = DIM*shiftidx[iidx];
 856
 857         /* Load limits for loop over neighbors */
 858         j_index_start    = jindex[iidx];
 859         j_index_end      = jindex[iidx+1];
 860
 861         /* Get outer coordinate index */
 862         inr              = iinr[iidx];
 863         i_coord_offset   = DIM*inr;
 864
 865         /* Load i particle coords and add shift vector */
 866         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 867                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 868
 869         fix0             = _mm_setzero_ps();
 870         fiy0             = _mm_setzero_ps();
 871         fiz0             = _mm_setzero_ps();
 872         fix1             = _mm_setzero_ps();
 873         fiy1             = _mm_setzero_ps();
 874         fiz1             = _mm_setzero_ps();
 875         fix2             = _mm_setzero_ps();
 876         fiy2             = _mm_setzero_ps();
 877         fiz2             = _mm_setzero_ps();
 878         fix3             = _mm_setzero_ps();
 879         fiy3             = _mm_setzero_ps();
 880         fiz3             = _mm_setzero_ps();
 881
 882         /* Start inner kernel loop */
 883         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 884         {
 885
 886             /* Get j neighbor index, and coordinate index */
 887             jnrA             = jjnr[jidx];
 888             jnrB             = jjnr[jidx+1];
 889             jnrC             = jjnr[jidx+2];
 890             jnrD             = jjnr[jidx+3];
 891             j_coord_offsetA  = DIM*jnrA;
 892             j_coord_offsetB  = DIM*jnrB;
 893             j_coord_offsetC  = DIM*jnrC;
 894             j_coord_offsetD  = DIM*jnrD;
 895
 896             /* load j atom coordinates */
 897             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 898                                               x+j_coord_offsetC,x+j_coord_offsetD,
 899                                               &jx0,&jy0,&jz0);
 900
 901             /* Calculate displacement vector */
 902             dx00             = _mm_sub_ps(ix0,jx0);
 903             dy00             = _mm_sub_ps(iy0,jy0);
 904             dz00             = _mm_sub_ps(iz0,jz0);
 905             dx10             = _mm_sub_ps(ix1,jx0);
 906             dy10             = _mm_sub_ps(iy1,jy0);
 907             dz10             = _mm_sub_ps(iz1,jz0);
 908             dx20             = _mm_sub_ps(ix2,jx0);
 909             dy20             = _mm_sub_ps(iy2,jy0);
 910             dz20             = _mm_sub_ps(iz2,jz0);
 911             dx30             = _mm_sub_ps(ix3,jx0);
 912             dy30             = _mm_sub_ps(iy3,jy0);
 913             dz30             = _mm_sub_ps(iz3,jz0);
 914
 915             /* Calculate squared distance and things based on it */
 916             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 917             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 918             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 919             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 920
 921             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 922             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 923             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 924
 925             rinvsq00         = gmx_mm_inv_ps(rsq00);
 926             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 927             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 928             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 929
 930             /* Load parameters for j particles */
 931             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 932                                                               charge+jnrC+0,charge+jnrD+0);
 933             vdwjidx0A        = 2*vdwtype[jnrA+0];
 934             vdwjidx0B        = 2*vdwtype[jnrB+0];
 935             vdwjidx0C        = 2*vdwtype[jnrC+0];
 936             vdwjidx0D        = 2*vdwtype[jnrD+0];
 937
 938             fjx0             = _mm_setzero_ps();
 939             fjy0             = _mm_setzero_ps();
 940             fjz0             = _mm_setzero_ps();
 941
 942             /**************************
 943              * CALCULATE INTERACTIONS *
 944              **************************/
 945
 946             if (gmx_mm_any_lt(rsq00,rcutoff2))
 947             {
 948
 949             /* Compute parameters for interactions between i and j atoms */
 950             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 951                                          vdwparam+vdwioffset0+vdwjidx0B,
 952                                          vdwparam+vdwioffset0+vdwjidx0C,
 953                                          vdwparam+vdwioffset0+vdwjidx0D,
 954                                          &c6_00,&c12_00);
 955
 956             /* LENNARD-JONES DISPERSION/REPULSION */
 957
 958             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 959             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 960
 961             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 962
 963             fscal            = fvdw;
 964
 965             fscal            = _mm_and_ps(fscal,cutoff_mask);
 966
 967              /* Update vectorial force */
 968             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 969             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 970             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 971
 972             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 973             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 974             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 975
 976             }
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             if (gmx_mm_any_lt(rsq10,rcutoff2))
 983             {
 984
 985             r10              = _mm_mul_ps(rsq10,rinv10);
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq10             = _mm_mul_ps(iq1,jq0);
 989
 990             /* EWALD ELECTROSTATICS */
 991
 992             /* Analytical PME correction */
 993             zeta2            = _mm_mul_ps(beta2,rsq10);
 994             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 995             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 996             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 997             felec            = _mm_mul_ps(qq10,felec);
 998
 999             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1000
1001             fscal            = felec;
1002
1003             fscal            = _mm_and_ps(fscal,cutoff_mask);
1004
1005              /* Update vectorial force */
1006             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1007             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1008             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1009
1010             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1011             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1012             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1013
1014             }
1015
1016             /**************************
1017              * CALCULATE INTERACTIONS *
1018              **************************/
1019
1020             if (gmx_mm_any_lt(rsq20,rcutoff2))
1021             {
1022
1023             r20              = _mm_mul_ps(rsq20,rinv20);
1024
1025             /* Compute parameters for interactions between i and j atoms */
1026             qq20             = _mm_mul_ps(iq2,jq0);
1027
1028             /* EWALD ELECTROSTATICS */
1029
1030             /* Analytical PME correction */
1031             zeta2            = _mm_mul_ps(beta2,rsq20);
1032             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1033             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1034             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1035             felec            = _mm_mul_ps(qq20,felec);
1036
1037             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1038
1039             fscal            = felec;
1040
1041             fscal            = _mm_and_ps(fscal,cutoff_mask);
1042
1043              /* Update vectorial force */
1044             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1045             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1046             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1047
1048             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1049             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1050             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1051
1052             }
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             if (gmx_mm_any_lt(rsq30,rcutoff2))
1059             {
1060
1061             r30              = _mm_mul_ps(rsq30,rinv30);
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             qq30             = _mm_mul_ps(iq3,jq0);
1065
1066             /* EWALD ELECTROSTATICS */
1067
1068             /* Analytical PME correction */
1069             zeta2            = _mm_mul_ps(beta2,rsq30);
1070             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1071             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1072             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1073             felec            = _mm_mul_ps(qq30,felec);
1074
1075             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1076
1077             fscal            = felec;
1078
1079             fscal            = _mm_and_ps(fscal,cutoff_mask);
1080
1081              /* Update vectorial force */
1082             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1083             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1084             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1085
1086             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1087             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1088             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1089
1090             }
1091
1092             fjptrA             = f+j_coord_offsetA;
1093             fjptrB             = f+j_coord_offsetB;
1094             fjptrC             = f+j_coord_offsetC;
1095             fjptrD             = f+j_coord_offsetD;
1096
1097             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1098
1099             /* Inner loop uses 126 flops */
1100         }
1101
1102         if(jidx<j_index_end)
1103         {
1104
1105             /* Get j neighbor index, and coordinate index */
1106             jnrlistA         = jjnr[jidx];
1107             jnrlistB         = jjnr[jidx+1];
1108             jnrlistC         = jjnr[jidx+2];
1109             jnrlistD         = jjnr[jidx+3];
1110             /* Sign of each element will be negative for non-real atoms.
1111              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1112              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1113              */
1114             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1115             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1116             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1117             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1118             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1119             j_coord_offsetA  = DIM*jnrA;
1120             j_coord_offsetB  = DIM*jnrB;
1121             j_coord_offsetC  = DIM*jnrC;
1122             j_coord_offsetD  = DIM*jnrD;
1123
1124             /* load j atom coordinates */
1125             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1126                                               x+j_coord_offsetC,x+j_coord_offsetD,
1127                                               &jx0,&jy0,&jz0);
1128
1129             /* Calculate displacement vector */
1130             dx00             = _mm_sub_ps(ix0,jx0);
1131             dy00             = _mm_sub_ps(iy0,jy0);
1132             dz00             = _mm_sub_ps(iz0,jz0);
1133             dx10             = _mm_sub_ps(ix1,jx0);
1134             dy10             = _mm_sub_ps(iy1,jy0);
1135             dz10             = _mm_sub_ps(iz1,jz0);
1136             dx20             = _mm_sub_ps(ix2,jx0);
1137             dy20             = _mm_sub_ps(iy2,jy0);
1138             dz20             = _mm_sub_ps(iz2,jz0);
1139             dx30             = _mm_sub_ps(ix3,jx0);
1140             dy30             = _mm_sub_ps(iy3,jy0);
1141             dz30             = _mm_sub_ps(iz3,jz0);
1142
1143             /* Calculate squared distance and things based on it */
1144             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1145             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1146             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1147             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1148
1149             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1150             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1151             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1152
1153             rinvsq00         = gmx_mm_inv_ps(rsq00);
1154             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1155             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1156             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1157
1158             /* Load parameters for j particles */
1159             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1160                                                               charge+jnrC+0,charge+jnrD+0);
1161             vdwjidx0A        = 2*vdwtype[jnrA+0];
1162             vdwjidx0B        = 2*vdwtype[jnrB+0];
1163             vdwjidx0C        = 2*vdwtype[jnrC+0];
1164             vdwjidx0D        = 2*vdwtype[jnrD+0];
1165
1166             fjx0             = _mm_setzero_ps();
1167             fjy0             = _mm_setzero_ps();
1168             fjz0             = _mm_setzero_ps();
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             if (gmx_mm_any_lt(rsq00,rcutoff2))
1175             {
1176
1177             /* Compute parameters for interactions between i and j atoms */
1178             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1179                                          vdwparam+vdwioffset0+vdwjidx0B,
1180                                          vdwparam+vdwioffset0+vdwjidx0C,
1181                                          vdwparam+vdwioffset0+vdwjidx0D,
1182                                          &c6_00,&c12_00);
1183
1184             /* LENNARD-JONES DISPERSION/REPULSION */
1185
1186             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1187             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1188
1189             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1190
1191             fscal            = fvdw;
1192
1193             fscal            = _mm_and_ps(fscal,cutoff_mask);
1194
1195             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1196
1197              /* Update vectorial force */
1198             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1199             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1200             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1201
1202             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1203             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1204             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1205
1206             }
1207
1208             /**************************
1209              * CALCULATE INTERACTIONS *
1210              **************************/
1211
1212             if (gmx_mm_any_lt(rsq10,rcutoff2))
1213             {
1214
1215             r10              = _mm_mul_ps(rsq10,rinv10);
1216             r10              = _mm_andnot_ps(dummy_mask,r10);
1217
1218             /* Compute parameters for interactions between i and j atoms */
1219             qq10             = _mm_mul_ps(iq1,jq0);
1220
1221             /* EWALD ELECTROSTATICS */
1222
1223             /* Analytical PME correction */
1224             zeta2            = _mm_mul_ps(beta2,rsq10);
1225             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1226             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1227             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1228             felec            = _mm_mul_ps(qq10,felec);
1229
1230             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1231
1232             fscal            = felec;
1233
1234             fscal            = _mm_and_ps(fscal,cutoff_mask);
1235
1236             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1237
1238              /* Update vectorial force */
1239             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1240             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1241             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1242
1243             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1244             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1245             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1246
1247             }
1248
1249             /**************************
1250              * CALCULATE INTERACTIONS *
1251              **************************/
1252
1253             if (gmx_mm_any_lt(rsq20,rcutoff2))
1254             {
1255
1256             r20              = _mm_mul_ps(rsq20,rinv20);
1257             r20              = _mm_andnot_ps(dummy_mask,r20);
1258
1259             /* Compute parameters for interactions between i and j atoms */
1260             qq20             = _mm_mul_ps(iq2,jq0);
1261
1262             /* EWALD ELECTROSTATICS */
1263
1264             /* Analytical PME correction */
1265             zeta2            = _mm_mul_ps(beta2,rsq20);
1266             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1267             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1268             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1269             felec            = _mm_mul_ps(qq20,felec);
1270
1271             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1272
1273             fscal            = felec;
1274
1275             fscal            = _mm_and_ps(fscal,cutoff_mask);
1276
1277             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1278
1279              /* Update vectorial force */
1280             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1281             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1282             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1283
1284             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1285             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1286             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1287
1288             }
1289
1290             /**************************
1291              * CALCULATE INTERACTIONS *
1292              **************************/
1293
1294             if (gmx_mm_any_lt(rsq30,rcutoff2))
1295             {
1296
1297             r30              = _mm_mul_ps(rsq30,rinv30);
1298             r30              = _mm_andnot_ps(dummy_mask,r30);
1299
1300             /* Compute parameters for interactions between i and j atoms */
1301             qq30             = _mm_mul_ps(iq3,jq0);
1302
1303             /* EWALD ELECTROSTATICS */
1304
1305             /* Analytical PME correction */
1306             zeta2            = _mm_mul_ps(beta2,rsq30);
1307             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1308             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1309             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1310             felec            = _mm_mul_ps(qq30,felec);
1311
1312             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1313
1314             fscal            = felec;
1315
1316             fscal            = _mm_and_ps(fscal,cutoff_mask);
1317
1318             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1319
1320              /* Update vectorial force */
1321             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1322             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1323             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1324
1325             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1326             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1327             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1328
1329             }
1330
1331             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1332             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1333             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1334             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1335
1336             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1337
1338             /* Inner loop uses 129 flops */
1339         }
1340
1341         /* End of innermost loop */
1342
1343         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1344                                               f+i_coord_offset,fshift+i_shift_offset);
1345
1346         /* Increment number of inner iterations */
1347         inneriter                  += j_index_end - j_index_start;
1348
1349         /* Outer loop uses 24 flops */
1350     }
1351
1352     /* Increment number of outer iterations */
1353     outeriter        += nri;
1354
1355     /* Update outer/inner flops */
1356
1357     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*129);
1358 }