src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwNone_GeomW4W4_avx_128_fma_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4W4_VF_avx_128_fma_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwNone_GeomW4W4_VF_avx_128_fma_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset1;
  83     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  89     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  90     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  91     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  92     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  93     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  94     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  95     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  96     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  97     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  98     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  99     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 100     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 101     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 102     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 103     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 104     real             *charge;
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     krf              = _mm_set1_ps(fr->ic->k_rf);
 123     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 124     crf              = _mm_set1_ps(fr->ic->c_rf);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 129     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 130     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 131
 132     jq1              = _mm_set1_ps(charge[inr+1]);
 133     jq2              = _mm_set1_ps(charge[inr+2]);
 134     jq3              = _mm_set1_ps(charge[inr+3]);
 135     qq11             = _mm_mul_ps(iq1,jq1);
 136     qq12             = _mm_mul_ps(iq1,jq2);
 137     qq13             = _mm_mul_ps(iq1,jq3);
 138     qq21             = _mm_mul_ps(iq2,jq1);
 139     qq22             = _mm_mul_ps(iq2,jq2);
 140     qq23             = _mm_mul_ps(iq2,jq3);
 141     qq31             = _mm_mul_ps(iq3,jq1);
 142     qq32             = _mm_mul_ps(iq3,jq2);
 143     qq33             = _mm_mul_ps(iq3,jq3);
 144
 145     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 146     rcutoff_scalar   = fr->ic->rcoulomb;
 147     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 148     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = jnrC = jnrD = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154     j_coord_offsetC = 0;
 155     j_coord_offsetD = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 181                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix1             = _mm_setzero_ps();
 184         fiy1             = _mm_setzero_ps();
 185         fiz1             = _mm_setzero_ps();
 186         fix2             = _mm_setzero_ps();
 187         fiy2             = _mm_setzero_ps();
 188         fiz2             = _mm_setzero_ps();
 189         fix3             = _mm_setzero_ps();
 190         fiy3             = _mm_setzero_ps();
 191         fiz3             = _mm_setzero_ps();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm_setzero_ps();
 195
 196         /* Start inner kernel loop */
 197         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 198         {
 199
 200             /* Get j neighbor index, and coordinate index */
 201             jnrA             = jjnr[jidx];
 202             jnrB             = jjnr[jidx+1];
 203             jnrC             = jjnr[jidx+2];
 204             jnrD             = jjnr[jidx+3];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209
 210             /* load j atom coordinates */
 211             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 212                                               x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 213                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 214
 215             /* Calculate displacement vector */
 216             dx11             = _mm_sub_ps(ix1,jx1);
 217             dy11             = _mm_sub_ps(iy1,jy1);
 218             dz11             = _mm_sub_ps(iz1,jz1);
 219             dx12             = _mm_sub_ps(ix1,jx2);
 220             dy12             = _mm_sub_ps(iy1,jy2);
 221             dz12             = _mm_sub_ps(iz1,jz2);
 222             dx13             = _mm_sub_ps(ix1,jx3);
 223             dy13             = _mm_sub_ps(iy1,jy3);
 224             dz13             = _mm_sub_ps(iz1,jz3);
 225             dx21             = _mm_sub_ps(ix2,jx1);
 226             dy21             = _mm_sub_ps(iy2,jy1);
 227             dz21             = _mm_sub_ps(iz2,jz1);
 228             dx22             = _mm_sub_ps(ix2,jx2);
 229             dy22             = _mm_sub_ps(iy2,jy2);
 230             dz22             = _mm_sub_ps(iz2,jz2);
 231             dx23             = _mm_sub_ps(ix2,jx3);
 232             dy23             = _mm_sub_ps(iy2,jy3);
 233             dz23             = _mm_sub_ps(iz2,jz3);
 234             dx31             = _mm_sub_ps(ix3,jx1);
 235             dy31             = _mm_sub_ps(iy3,jy1);
 236             dz31             = _mm_sub_ps(iz3,jz1);
 237             dx32             = _mm_sub_ps(ix3,jx2);
 238             dy32             = _mm_sub_ps(iy3,jy2);
 239             dz32             = _mm_sub_ps(iz3,jz2);
 240             dx33             = _mm_sub_ps(ix3,jx3);
 241             dy33             = _mm_sub_ps(iy3,jy3);
 242             dz33             = _mm_sub_ps(iz3,jz3);
 243
 244             /* Calculate squared distance and things based on it */
 245             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 246             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 247             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 248             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 249             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 250             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 251             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 252             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 253             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 254
 255             rinv11           = avx128fma_invsqrt_f(rsq11);
 256             rinv12           = avx128fma_invsqrt_f(rsq12);
 257             rinv13           = avx128fma_invsqrt_f(rsq13);
 258             rinv21           = avx128fma_invsqrt_f(rsq21);
 259             rinv22           = avx128fma_invsqrt_f(rsq22);
 260             rinv23           = avx128fma_invsqrt_f(rsq23);
 261             rinv31           = avx128fma_invsqrt_f(rsq31);
 262             rinv32           = avx128fma_invsqrt_f(rsq32);
 263             rinv33           = avx128fma_invsqrt_f(rsq33);
 264
 265             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 266             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 267             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 268             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 269             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 270             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 271             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 272             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 273             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 274
 275             fjx1             = _mm_setzero_ps();
 276             fjy1             = _mm_setzero_ps();
 277             fjz1             = _mm_setzero_ps();
 278             fjx2             = _mm_setzero_ps();
 279             fjy2             = _mm_setzero_ps();
 280             fjz2             = _mm_setzero_ps();
 281             fjx3             = _mm_setzero_ps();
 282             fjy3             = _mm_setzero_ps();
 283             fjz3             = _mm_setzero_ps();
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             if (gmx_mm_any_lt(rsq11,rcutoff2))
 290             {
 291
 292             /* REACTION-FIELD ELECTROSTATICS */
 293             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 294             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 295
 296             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velec            = _mm_and_ps(velec,cutoff_mask);
 300             velecsum         = _mm_add_ps(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             fscal            = _mm_and_ps(fscal,cutoff_mask);
 305
 306              /* Update vectorial force */
 307             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 308             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 309             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 310
 311             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 312             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 313             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 314
 315             }
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             if (gmx_mm_any_lt(rsq12,rcutoff2))
 322             {
 323
 324             /* REACTION-FIELD ELECTROSTATICS */
 325             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 326             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 327
 328             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velec            = _mm_and_ps(velec,cutoff_mask);
 332             velecsum         = _mm_add_ps(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336             fscal            = _mm_and_ps(fscal,cutoff_mask);
 337
 338              /* Update vectorial force */
 339             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 340             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 341             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 342
 343             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 344             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 345             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 346
 347             }
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             if (gmx_mm_any_lt(rsq13,rcutoff2))
 354             {
 355
 356             /* REACTION-FIELD ELECTROSTATICS */
 357             velec            = _mm_mul_ps(qq13,_mm_sub_ps(_mm_macc_ps(krf,rsq13,rinv13),crf));
 358             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
 359
 360             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velec            = _mm_and_ps(velec,cutoff_mask);
 364             velecsum         = _mm_add_ps(velecsum,velec);
 365
 366             fscal            = felec;
 367
 368             fscal            = _mm_and_ps(fscal,cutoff_mask);
 369
 370              /* Update vectorial force */
 371             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 372             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 373             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 374
 375             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 376             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 377             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 378
 379             }
 380
 381             /**************************
 382              * CALCULATE INTERACTIONS *
 383              **************************/
 384
 385             if (gmx_mm_any_lt(rsq21,rcutoff2))
 386             {
 387
 388             /* REACTION-FIELD ELECTROSTATICS */
 389             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 390             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 391
 392             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velec            = _mm_and_ps(velec,cutoff_mask);
 396             velecsum         = _mm_add_ps(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             fscal            = _mm_and_ps(fscal,cutoff_mask);
 401
 402              /* Update vectorial force */
 403             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 404             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 405             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 406
 407             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 408             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 409             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 410
 411             }
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             if (gmx_mm_any_lt(rsq22,rcutoff2))
 418             {
 419
 420             /* REACTION-FIELD ELECTROSTATICS */
 421             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 422             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 423
 424             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velec            = _mm_and_ps(velec,cutoff_mask);
 428             velecsum         = _mm_add_ps(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             fscal            = _mm_and_ps(fscal,cutoff_mask);
 433
 434              /* Update vectorial force */
 435             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 436             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 437             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 438
 439             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 440             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 441             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 442
 443             }
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             if (gmx_mm_any_lt(rsq23,rcutoff2))
 450             {
 451
 452             /* REACTION-FIELD ELECTROSTATICS */
 453             velec            = _mm_mul_ps(qq23,_mm_sub_ps(_mm_macc_ps(krf,rsq23,rinv23),crf));
 454             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
 455
 456             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm_and_ps(velec,cutoff_mask);
 460             velecsum         = _mm_add_ps(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             fscal            = _mm_and_ps(fscal,cutoff_mask);
 465
 466              /* Update vectorial force */
 467             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 468             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 469             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 470
 471             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 472             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 473             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 474
 475             }
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             if (gmx_mm_any_lt(rsq31,rcutoff2))
 482             {
 483
 484             /* REACTION-FIELD ELECTROSTATICS */
 485             velec            = _mm_mul_ps(qq31,_mm_sub_ps(_mm_macc_ps(krf,rsq31,rinv31),crf));
 486             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
 487
 488             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velec            = _mm_and_ps(velec,cutoff_mask);
 492             velecsum         = _mm_add_ps(velecsum,velec);
 493
 494             fscal            = felec;
 495
 496             fscal            = _mm_and_ps(fscal,cutoff_mask);
 497
 498              /* Update vectorial force */
 499             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 500             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 501             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 502
 503             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 504             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 505             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 506
 507             }
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             if (gmx_mm_any_lt(rsq32,rcutoff2))
 514             {
 515
 516             /* REACTION-FIELD ELECTROSTATICS */
 517             velec            = _mm_mul_ps(qq32,_mm_sub_ps(_mm_macc_ps(krf,rsq32,rinv32),crf));
 518             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
 519
 520             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _mm_and_ps(velec,cutoff_mask);
 524             velecsum         = _mm_add_ps(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_and_ps(fscal,cutoff_mask);
 529
 530              /* Update vectorial force */
 531             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 532             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 533             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 534
 535             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 536             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 537             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 538
 539             }
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             if (gmx_mm_any_lt(rsq33,rcutoff2))
 546             {
 547
 548             /* REACTION-FIELD ELECTROSTATICS */
 549             velec            = _mm_mul_ps(qq33,_mm_sub_ps(_mm_macc_ps(krf,rsq33,rinv33),crf));
 550             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
 551
 552             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm_and_ps(velec,cutoff_mask);
 556             velecsum         = _mm_add_ps(velecsum,velec);
 557
 558             fscal            = felec;
 559
 560             fscal            = _mm_and_ps(fscal,cutoff_mask);
 561
 562              /* Update vectorial force */
 563             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 564             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 565             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 566
 567             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 568             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 569             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 570
 571             }
 572
 573             fjptrA             = f+j_coord_offsetA;
 574             fjptrB             = f+j_coord_offsetB;
 575             fjptrC             = f+j_coord_offsetC;
 576             fjptrD             = f+j_coord_offsetD;
 577
 578             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 579                                                    fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 580
 581             /* Inner loop uses 351 flops */
 582         }
 583
 584         if(jidx<j_index_end)
 585         {
 586
 587             /* Get j neighbor index, and coordinate index */
 588             jnrlistA         = jjnr[jidx];
 589             jnrlistB         = jjnr[jidx+1];
 590             jnrlistC         = jjnr[jidx+2];
 591             jnrlistD         = jjnr[jidx+3];
 592             /* Sign of each element will be negative for non-real atoms.
 593              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 594              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 595              */
 596             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 597             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 598             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 599             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 600             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 601             j_coord_offsetA  = DIM*jnrA;
 602             j_coord_offsetB  = DIM*jnrB;
 603             j_coord_offsetC  = DIM*jnrC;
 604             j_coord_offsetD  = DIM*jnrD;
 605
 606             /* load j atom coordinates */
 607             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 608                                               x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 609                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 610
 611             /* Calculate displacement vector */
 612             dx11             = _mm_sub_ps(ix1,jx1);
 613             dy11             = _mm_sub_ps(iy1,jy1);
 614             dz11             = _mm_sub_ps(iz1,jz1);
 615             dx12             = _mm_sub_ps(ix1,jx2);
 616             dy12             = _mm_sub_ps(iy1,jy2);
 617             dz12             = _mm_sub_ps(iz1,jz2);
 618             dx13             = _mm_sub_ps(ix1,jx3);
 619             dy13             = _mm_sub_ps(iy1,jy3);
 620             dz13             = _mm_sub_ps(iz1,jz3);
 621             dx21             = _mm_sub_ps(ix2,jx1);
 622             dy21             = _mm_sub_ps(iy2,jy1);
 623             dz21             = _mm_sub_ps(iz2,jz1);
 624             dx22             = _mm_sub_ps(ix2,jx2);
 625             dy22             = _mm_sub_ps(iy2,jy2);
 626             dz22             = _mm_sub_ps(iz2,jz2);
 627             dx23             = _mm_sub_ps(ix2,jx3);
 628             dy23             = _mm_sub_ps(iy2,jy3);
 629             dz23             = _mm_sub_ps(iz2,jz3);
 630             dx31             = _mm_sub_ps(ix3,jx1);
 631             dy31             = _mm_sub_ps(iy3,jy1);
 632             dz31             = _mm_sub_ps(iz3,jz1);
 633             dx32             = _mm_sub_ps(ix3,jx2);
 634             dy32             = _mm_sub_ps(iy3,jy2);
 635             dz32             = _mm_sub_ps(iz3,jz2);
 636             dx33             = _mm_sub_ps(ix3,jx3);
 637             dy33             = _mm_sub_ps(iy3,jy3);
 638             dz33             = _mm_sub_ps(iz3,jz3);
 639
 640             /* Calculate squared distance and things based on it */
 641             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 642             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 643             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 644             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 645             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 646             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 647             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 648             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 649             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 650
 651             rinv11           = avx128fma_invsqrt_f(rsq11);
 652             rinv12           = avx128fma_invsqrt_f(rsq12);
 653             rinv13           = avx128fma_invsqrt_f(rsq13);
 654             rinv21           = avx128fma_invsqrt_f(rsq21);
 655             rinv22           = avx128fma_invsqrt_f(rsq22);
 656             rinv23           = avx128fma_invsqrt_f(rsq23);
 657             rinv31           = avx128fma_invsqrt_f(rsq31);
 658             rinv32           = avx128fma_invsqrt_f(rsq32);
 659             rinv33           = avx128fma_invsqrt_f(rsq33);
 660
 661             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 662             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 663             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 664             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 665             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 666             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 667             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 668             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 669             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 670
 671             fjx1             = _mm_setzero_ps();
 672             fjy1             = _mm_setzero_ps();
 673             fjz1             = _mm_setzero_ps();
 674             fjx2             = _mm_setzero_ps();
 675             fjy2             = _mm_setzero_ps();
 676             fjz2             = _mm_setzero_ps();
 677             fjx3             = _mm_setzero_ps();
 678             fjy3             = _mm_setzero_ps();
 679             fjz3             = _mm_setzero_ps();
 680
 681             /**************************
 682              * CALCULATE INTERACTIONS *
 683              **************************/
 684
 685             if (gmx_mm_any_lt(rsq11,rcutoff2))
 686             {
 687
 688             /* REACTION-FIELD ELECTROSTATICS */
 689             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 690             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 691
 692             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 693
 694             /* Update potential sum for this i atom from the interaction with this j atom. */
 695             velec            = _mm_and_ps(velec,cutoff_mask);
 696             velec            = _mm_andnot_ps(dummy_mask,velec);
 697             velecsum         = _mm_add_ps(velecsum,velec);
 698
 699             fscal            = felec;
 700
 701             fscal            = _mm_and_ps(fscal,cutoff_mask);
 702
 703             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 704
 705              /* Update vectorial force */
 706             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 707             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 708             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 709
 710             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 711             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 712             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 713
 714             }
 715
 716             /**************************
 717              * CALCULATE INTERACTIONS *
 718              **************************/
 719
 720             if (gmx_mm_any_lt(rsq12,rcutoff2))
 721             {
 722
 723             /* REACTION-FIELD ELECTROSTATICS */
 724             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 725             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 726
 727             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
 728
 729             /* Update potential sum for this i atom from the interaction with this j atom. */
 730             velec            = _mm_and_ps(velec,cutoff_mask);
 731             velec            = _mm_andnot_ps(dummy_mask,velec);
 732             velecsum         = _mm_add_ps(velecsum,velec);
 733
 734             fscal            = felec;
 735
 736             fscal            = _mm_and_ps(fscal,cutoff_mask);
 737
 738             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 739
 740              /* Update vectorial force */
 741             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 742             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 743             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 744
 745             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 746             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 747             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 748
 749             }
 750
 751             /**************************
 752              * CALCULATE INTERACTIONS *
 753              **************************/
 754
 755             if (gmx_mm_any_lt(rsq13,rcutoff2))
 756             {
 757
 758             /* REACTION-FIELD ELECTROSTATICS */
 759             velec            = _mm_mul_ps(qq13,_mm_sub_ps(_mm_macc_ps(krf,rsq13,rinv13),crf));
 760             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
 761
 762             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
 763
 764             /* Update potential sum for this i atom from the interaction with this j atom. */
 765             velec            = _mm_and_ps(velec,cutoff_mask);
 766             velec            = _mm_andnot_ps(dummy_mask,velec);
 767             velecsum         = _mm_add_ps(velecsum,velec);
 768
 769             fscal            = felec;
 770
 771             fscal            = _mm_and_ps(fscal,cutoff_mask);
 772
 773             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 774
 775              /* Update vectorial force */
 776             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 777             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 778             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 779
 780             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 781             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 782             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 783
 784             }
 785
 786             /**************************
 787              * CALCULATE INTERACTIONS *
 788              **************************/
 789
 790             if (gmx_mm_any_lt(rsq21,rcutoff2))
 791             {
 792
 793             /* REACTION-FIELD ELECTROSTATICS */
 794             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 795             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 796
 797             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
 798
 799             /* Update potential sum for this i atom from the interaction with this j atom. */
 800             velec            = _mm_and_ps(velec,cutoff_mask);
 801             velec            = _mm_andnot_ps(dummy_mask,velec);
 802             velecsum         = _mm_add_ps(velecsum,velec);
 803
 804             fscal            = felec;
 805
 806             fscal            = _mm_and_ps(fscal,cutoff_mask);
 807
 808             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 809
 810              /* Update vectorial force */
 811             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 812             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 813             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 814
 815             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 816             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 817             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 818
 819             }
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             if (gmx_mm_any_lt(rsq22,rcutoff2))
 826             {
 827
 828             /* REACTION-FIELD ELECTROSTATICS */
 829             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 830             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 831
 832             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
 833
 834             /* Update potential sum for this i atom from the interaction with this j atom. */
 835             velec            = _mm_and_ps(velec,cutoff_mask);
 836             velec            = _mm_andnot_ps(dummy_mask,velec);
 837             velecsum         = _mm_add_ps(velecsum,velec);
 838
 839             fscal            = felec;
 840
 841             fscal            = _mm_and_ps(fscal,cutoff_mask);
 842
 843             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 844
 845              /* Update vectorial force */
 846             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 847             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 848             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 849
 850             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 851             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 852             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 853
 854             }
 855
 856             /**************************
 857              * CALCULATE INTERACTIONS *
 858              **************************/
 859
 860             if (gmx_mm_any_lt(rsq23,rcutoff2))
 861             {
 862
 863             /* REACTION-FIELD ELECTROSTATICS */
 864             velec            = _mm_mul_ps(qq23,_mm_sub_ps(_mm_macc_ps(krf,rsq23,rinv23),crf));
 865             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
 866
 867             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
 868
 869             /* Update potential sum for this i atom from the interaction with this j atom. */
 870             velec            = _mm_and_ps(velec,cutoff_mask);
 871             velec            = _mm_andnot_ps(dummy_mask,velec);
 872             velecsum         = _mm_add_ps(velecsum,velec);
 873
 874             fscal            = felec;
 875
 876             fscal            = _mm_and_ps(fscal,cutoff_mask);
 877
 878             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 879
 880              /* Update vectorial force */
 881             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 882             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 883             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 884
 885             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 886             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 887             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 888
 889             }
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             if (gmx_mm_any_lt(rsq31,rcutoff2))
 896             {
 897
 898             /* REACTION-FIELD ELECTROSTATICS */
 899             velec            = _mm_mul_ps(qq31,_mm_sub_ps(_mm_macc_ps(krf,rsq31,rinv31),crf));
 900             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
 901
 902             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
 903
 904             /* Update potential sum for this i atom from the interaction with this j atom. */
 905             velec            = _mm_and_ps(velec,cutoff_mask);
 906             velec            = _mm_andnot_ps(dummy_mask,velec);
 907             velecsum         = _mm_add_ps(velecsum,velec);
 908
 909             fscal            = felec;
 910
 911             fscal            = _mm_and_ps(fscal,cutoff_mask);
 912
 913             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 914
 915              /* Update vectorial force */
 916             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 917             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 918             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 919
 920             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 921             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 922             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 923
 924             }
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             if (gmx_mm_any_lt(rsq32,rcutoff2))
 931             {
 932
 933             /* REACTION-FIELD ELECTROSTATICS */
 934             velec            = _mm_mul_ps(qq32,_mm_sub_ps(_mm_macc_ps(krf,rsq32,rinv32),crf));
 935             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
 936
 937             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
 938
 939             /* Update potential sum for this i atom from the interaction with this j atom. */
 940             velec            = _mm_and_ps(velec,cutoff_mask);
 941             velec            = _mm_andnot_ps(dummy_mask,velec);
 942             velecsum         = _mm_add_ps(velecsum,velec);
 943
 944             fscal            = felec;
 945
 946             fscal            = _mm_and_ps(fscal,cutoff_mask);
 947
 948             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 949
 950              /* Update vectorial force */
 951             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 952             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 953             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 954
 955             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 956             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 957             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 958
 959             }
 960
 961             /**************************
 962              * CALCULATE INTERACTIONS *
 963              **************************/
 964
 965             if (gmx_mm_any_lt(rsq33,rcutoff2))
 966             {
 967
 968             /* REACTION-FIELD ELECTROSTATICS */
 969             velec            = _mm_mul_ps(qq33,_mm_sub_ps(_mm_macc_ps(krf,rsq33,rinv33),crf));
 970             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
 971
 972             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
 973
 974             /* Update potential sum for this i atom from the interaction with this j atom. */
 975             velec            = _mm_and_ps(velec,cutoff_mask);
 976             velec            = _mm_andnot_ps(dummy_mask,velec);
 977             velecsum         = _mm_add_ps(velecsum,velec);
 978
 979             fscal            = felec;
 980
 981             fscal            = _mm_and_ps(fscal,cutoff_mask);
 982
 983             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 984
 985              /* Update vectorial force */
 986             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 987             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 988             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 989
 990             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 991             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 992             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 993
 994             }
 995
 996             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 997             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 998             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 999             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1000
1001             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1002                                                    fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1003
1004             /* Inner loop uses 351 flops */
1005         }
1006
1007         /* End of innermost loop */
1008
1009         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1010                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1011
1012         ggid                        = gid[iidx];
1013         /* Update potential energies */
1014         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1015
1016         /* Increment number of inner iterations */
1017         inneriter                  += j_index_end - j_index_start;
1018
1019         /* Outer loop uses 19 flops */
1020     }
1021
1022     /* Increment number of outer iterations */
1023     outeriter        += nri;
1024
1025     /* Update outer/inner flops */
1026
1027     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*351);
1028 }
1029 /*
1030  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4W4_F_avx_128_fma_single
1031  * Electrostatics interaction: ReactionField
1032  * VdW interaction:            None
1033  * Geometry:                   Water4-Water4
1034  * Calculate force/pot:        Force
1035  */
1036 void
1037 nb_kernel_ElecRFCut_VdwNone_GeomW4W4_F_avx_128_fma_single
1038                     (t_nblist                    * gmx_restrict       nlist,
1039                      rvec                        * gmx_restrict          xx,
1040                      rvec                        * gmx_restrict          ff,
1041                      struct t_forcerec           * gmx_restrict          fr,
1042                      t_mdatoms                   * gmx_restrict     mdatoms,
1043                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1044                      t_nrnb                      * gmx_restrict        nrnb)
1045 {
1046     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1047      * just 0 for non-waters.
1048      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1049      * jnr indices corresponding to data put in the four positions in the SIMD register.
1050      */
1051     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1052     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1053     int              jnrA,jnrB,jnrC,jnrD;
1054     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1055     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1056     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1057     real             rcutoff_scalar;
1058     real             *shiftvec,*fshift,*x,*f;
1059     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1060     real             scratch[4*DIM];
1061     __m128           fscal,rcutoff,rcutoff2,jidxall;
1062     int              vdwioffset1;
1063     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1064     int              vdwioffset2;
1065     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1066     int              vdwioffset3;
1067     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1068     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1069     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1070     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1071     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1072     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1073     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1074     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1075     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1076     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1077     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1078     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1079     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1080     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1081     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1082     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1083     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1084     real             *charge;
1085     __m128           dummy_mask,cutoff_mask;
1086     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1087     __m128           one     = _mm_set1_ps(1.0);
1088     __m128           two     = _mm_set1_ps(2.0);
1089     x                = xx[0];
1090     f                = ff[0];
1091
1092     nri              = nlist->nri;
1093     iinr             = nlist->iinr;
1094     jindex           = nlist->jindex;
1095     jjnr             = nlist->jjnr;
1096     shiftidx         = nlist->shift;
1097     gid              = nlist->gid;
1098     shiftvec         = fr->shift_vec[0];
1099     fshift           = fr->fshift[0];
1100     facel            = _mm_set1_ps(fr->ic->epsfac);
1101     charge           = mdatoms->chargeA;
1102     krf              = _mm_set1_ps(fr->ic->k_rf);
1103     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
1104     crf              = _mm_set1_ps(fr->ic->c_rf);
1105
1106     /* Setup water-specific parameters */
1107     inr              = nlist->iinr[0];
1108     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1109     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1110     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1111
1112     jq1              = _mm_set1_ps(charge[inr+1]);
1113     jq2              = _mm_set1_ps(charge[inr+2]);
1114     jq3              = _mm_set1_ps(charge[inr+3]);
1115     qq11             = _mm_mul_ps(iq1,jq1);
1116     qq12             = _mm_mul_ps(iq1,jq2);
1117     qq13             = _mm_mul_ps(iq1,jq3);
1118     qq21             = _mm_mul_ps(iq2,jq1);
1119     qq22             = _mm_mul_ps(iq2,jq2);
1120     qq23             = _mm_mul_ps(iq2,jq3);
1121     qq31             = _mm_mul_ps(iq3,jq1);
1122     qq32             = _mm_mul_ps(iq3,jq2);
1123     qq33             = _mm_mul_ps(iq3,jq3);
1124
1125     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1126     rcutoff_scalar   = fr->ic->rcoulomb;
1127     rcutoff          = _mm_set1_ps(rcutoff_scalar);
1128     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
1129
1130     /* Avoid stupid compiler warnings */
1131     jnrA = jnrB = jnrC = jnrD = 0;
1132     j_coord_offsetA = 0;
1133     j_coord_offsetB = 0;
1134     j_coord_offsetC = 0;
1135     j_coord_offsetD = 0;
1136
1137     outeriter        = 0;
1138     inneriter        = 0;
1139
1140     for(iidx=0;iidx<4*DIM;iidx++)
1141     {
1142         scratch[iidx] = 0.0;
1143     }
1144
1145     /* Start outer loop over neighborlists */
1146     for(iidx=0; iidx<nri; iidx++)
1147     {
1148         /* Load shift vector for this list */
1149         i_shift_offset   = DIM*shiftidx[iidx];
1150
1151         /* Load limits for loop over neighbors */
1152         j_index_start    = jindex[iidx];
1153         j_index_end      = jindex[iidx+1];
1154
1155         /* Get outer coordinate index */
1156         inr              = iinr[iidx];
1157         i_coord_offset   = DIM*inr;
1158
1159         /* Load i particle coords and add shift vector */
1160         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1161                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1162
1163         fix1             = _mm_setzero_ps();
1164         fiy1             = _mm_setzero_ps();
1165         fiz1             = _mm_setzero_ps();
1166         fix2             = _mm_setzero_ps();
1167         fiy2             = _mm_setzero_ps();
1168         fiz2             = _mm_setzero_ps();
1169         fix3             = _mm_setzero_ps();
1170         fiy3             = _mm_setzero_ps();
1171         fiz3             = _mm_setzero_ps();
1172
1173         /* Start inner kernel loop */
1174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1175         {
1176
1177             /* Get j neighbor index, and coordinate index */
1178             jnrA             = jjnr[jidx];
1179             jnrB             = jjnr[jidx+1];
1180             jnrC             = jjnr[jidx+2];
1181             jnrD             = jjnr[jidx+3];
1182             j_coord_offsetA  = DIM*jnrA;
1183             j_coord_offsetB  = DIM*jnrB;
1184             j_coord_offsetC  = DIM*jnrC;
1185             j_coord_offsetD  = DIM*jnrD;
1186
1187             /* load j atom coordinates */
1188             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1189                                               x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1190                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1191
1192             /* Calculate displacement vector */
1193             dx11             = _mm_sub_ps(ix1,jx1);
1194             dy11             = _mm_sub_ps(iy1,jy1);
1195             dz11             = _mm_sub_ps(iz1,jz1);
1196             dx12             = _mm_sub_ps(ix1,jx2);
1197             dy12             = _mm_sub_ps(iy1,jy2);
1198             dz12             = _mm_sub_ps(iz1,jz2);
1199             dx13             = _mm_sub_ps(ix1,jx3);
1200             dy13             = _mm_sub_ps(iy1,jy3);
1201             dz13             = _mm_sub_ps(iz1,jz3);
1202             dx21             = _mm_sub_ps(ix2,jx1);
1203             dy21             = _mm_sub_ps(iy2,jy1);
1204             dz21             = _mm_sub_ps(iz2,jz1);
1205             dx22             = _mm_sub_ps(ix2,jx2);
1206             dy22             = _mm_sub_ps(iy2,jy2);
1207             dz22             = _mm_sub_ps(iz2,jz2);
1208             dx23             = _mm_sub_ps(ix2,jx3);
1209             dy23             = _mm_sub_ps(iy2,jy3);
1210             dz23             = _mm_sub_ps(iz2,jz3);
1211             dx31             = _mm_sub_ps(ix3,jx1);
1212             dy31             = _mm_sub_ps(iy3,jy1);
1213             dz31             = _mm_sub_ps(iz3,jz1);
1214             dx32             = _mm_sub_ps(ix3,jx2);
1215             dy32             = _mm_sub_ps(iy3,jy2);
1216             dz32             = _mm_sub_ps(iz3,jz2);
1217             dx33             = _mm_sub_ps(ix3,jx3);
1218             dy33             = _mm_sub_ps(iy3,jy3);
1219             dz33             = _mm_sub_ps(iz3,jz3);
1220
1221             /* Calculate squared distance and things based on it */
1222             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1223             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1224             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1225             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1226             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1227             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1228             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1229             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1230             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1231
1232             rinv11           = avx128fma_invsqrt_f(rsq11);
1233             rinv12           = avx128fma_invsqrt_f(rsq12);
1234             rinv13           = avx128fma_invsqrt_f(rsq13);
1235             rinv21           = avx128fma_invsqrt_f(rsq21);
1236             rinv22           = avx128fma_invsqrt_f(rsq22);
1237             rinv23           = avx128fma_invsqrt_f(rsq23);
1238             rinv31           = avx128fma_invsqrt_f(rsq31);
1239             rinv32           = avx128fma_invsqrt_f(rsq32);
1240             rinv33           = avx128fma_invsqrt_f(rsq33);
1241
1242             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1243             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1244             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1245             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1246             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1247             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1248             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1249             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1250             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1251
1252             fjx1             = _mm_setzero_ps();
1253             fjy1             = _mm_setzero_ps();
1254             fjz1             = _mm_setzero_ps();
1255             fjx2             = _mm_setzero_ps();
1256             fjy2             = _mm_setzero_ps();
1257             fjz2             = _mm_setzero_ps();
1258             fjx3             = _mm_setzero_ps();
1259             fjy3             = _mm_setzero_ps();
1260             fjz3             = _mm_setzero_ps();
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm_any_lt(rsq11,rcutoff2))
1267             {
1268
1269             /* REACTION-FIELD ELECTROSTATICS */
1270             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1271
1272             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1273
1274             fscal            = felec;
1275
1276             fscal            = _mm_and_ps(fscal,cutoff_mask);
1277
1278              /* Update vectorial force */
1279             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1280             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1281             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1282
1283             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1284             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1285             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1286
1287             }
1288
1289             /**************************
1290              * CALCULATE INTERACTIONS *
1291              **************************/
1292
1293             if (gmx_mm_any_lt(rsq12,rcutoff2))
1294             {
1295
1296             /* REACTION-FIELD ELECTROSTATICS */
1297             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1298
1299             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1300
1301             fscal            = felec;
1302
1303             fscal            = _mm_and_ps(fscal,cutoff_mask);
1304
1305              /* Update vectorial force */
1306             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1307             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1308             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1309
1310             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1311             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1312             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1313
1314             }
1315
1316             /**************************
1317              * CALCULATE INTERACTIONS *
1318              **************************/
1319
1320             if (gmx_mm_any_lt(rsq13,rcutoff2))
1321             {
1322
1323             /* REACTION-FIELD ELECTROSTATICS */
1324             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
1325
1326             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
1327
1328             fscal            = felec;
1329
1330             fscal            = _mm_and_ps(fscal,cutoff_mask);
1331
1332              /* Update vectorial force */
1333             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1334             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1335             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1336
1337             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1338             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1339             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1340
1341             }
1342
1343             /**************************
1344              * CALCULATE INTERACTIONS *
1345              **************************/
1346
1347             if (gmx_mm_any_lt(rsq21,rcutoff2))
1348             {
1349
1350             /* REACTION-FIELD ELECTROSTATICS */
1351             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1352
1353             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1354
1355             fscal            = felec;
1356
1357             fscal            = _mm_and_ps(fscal,cutoff_mask);
1358
1359              /* Update vectorial force */
1360             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1361             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1362             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1363
1364             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1365             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1366             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1367
1368             }
1369
1370             /**************************
1371              * CALCULATE INTERACTIONS *
1372              **************************/
1373
1374             if (gmx_mm_any_lt(rsq22,rcutoff2))
1375             {
1376
1377             /* REACTION-FIELD ELECTROSTATICS */
1378             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1379
1380             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1381
1382             fscal            = felec;
1383
1384             fscal            = _mm_and_ps(fscal,cutoff_mask);
1385
1386              /* Update vectorial force */
1387             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1388             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1389             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1390
1391             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1392             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1393             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1394
1395             }
1396
1397             /**************************
1398              * CALCULATE INTERACTIONS *
1399              **************************/
1400
1401             if (gmx_mm_any_lt(rsq23,rcutoff2))
1402             {
1403
1404             /* REACTION-FIELD ELECTROSTATICS */
1405             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
1406
1407             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
1408
1409             fscal            = felec;
1410
1411             fscal            = _mm_and_ps(fscal,cutoff_mask);
1412
1413              /* Update vectorial force */
1414             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1415             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1416             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1417
1418             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1419             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1420             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1421
1422             }
1423
1424             /**************************
1425              * CALCULATE INTERACTIONS *
1426              **************************/
1427
1428             if (gmx_mm_any_lt(rsq31,rcutoff2))
1429             {
1430
1431             /* REACTION-FIELD ELECTROSTATICS */
1432             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
1433
1434             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
1435
1436             fscal            = felec;
1437
1438             fscal            = _mm_and_ps(fscal,cutoff_mask);
1439
1440              /* Update vectorial force */
1441             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1442             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1443             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1444
1445             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1446             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1447             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1448
1449             }
1450
1451             /**************************
1452              * CALCULATE INTERACTIONS *
1453              **************************/
1454
1455             if (gmx_mm_any_lt(rsq32,rcutoff2))
1456             {
1457
1458             /* REACTION-FIELD ELECTROSTATICS */
1459             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
1460
1461             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
1462
1463             fscal            = felec;
1464
1465             fscal            = _mm_and_ps(fscal,cutoff_mask);
1466
1467              /* Update vectorial force */
1468             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1469             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1470             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1471
1472             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1473             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1474             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1475
1476             }
1477
1478             /**************************
1479              * CALCULATE INTERACTIONS *
1480              **************************/
1481
1482             if (gmx_mm_any_lt(rsq33,rcutoff2))
1483             {
1484
1485             /* REACTION-FIELD ELECTROSTATICS */
1486             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
1487
1488             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
1489
1490             fscal            = felec;
1491
1492             fscal            = _mm_and_ps(fscal,cutoff_mask);
1493
1494              /* Update vectorial force */
1495             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1496             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1497             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1498
1499             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1500             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1501             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1502
1503             }
1504
1505             fjptrA             = f+j_coord_offsetA;
1506             fjptrB             = f+j_coord_offsetB;
1507             fjptrC             = f+j_coord_offsetC;
1508             fjptrD             = f+j_coord_offsetD;
1509
1510             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1511                                                    fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1512
1513             /* Inner loop uses 297 flops */
1514         }
1515
1516         if(jidx<j_index_end)
1517         {
1518
1519             /* Get j neighbor index, and coordinate index */
1520             jnrlistA         = jjnr[jidx];
1521             jnrlistB         = jjnr[jidx+1];
1522             jnrlistC         = jjnr[jidx+2];
1523             jnrlistD         = jjnr[jidx+3];
1524             /* Sign of each element will be negative for non-real atoms.
1525              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1526              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1527              */
1528             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1529             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1530             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1531             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1532             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1533             j_coord_offsetA  = DIM*jnrA;
1534             j_coord_offsetB  = DIM*jnrB;
1535             j_coord_offsetC  = DIM*jnrC;
1536             j_coord_offsetD  = DIM*jnrD;
1537
1538             /* load j atom coordinates */
1539             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1540                                               x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1541                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1542
1543             /* Calculate displacement vector */
1544             dx11             = _mm_sub_ps(ix1,jx1);
1545             dy11             = _mm_sub_ps(iy1,jy1);
1546             dz11             = _mm_sub_ps(iz1,jz1);
1547             dx12             = _mm_sub_ps(ix1,jx2);
1548             dy12             = _mm_sub_ps(iy1,jy2);
1549             dz12             = _mm_sub_ps(iz1,jz2);
1550             dx13             = _mm_sub_ps(ix1,jx3);
1551             dy13             = _mm_sub_ps(iy1,jy3);
1552             dz13             = _mm_sub_ps(iz1,jz3);
1553             dx21             = _mm_sub_ps(ix2,jx1);
1554             dy21             = _mm_sub_ps(iy2,jy1);
1555             dz21             = _mm_sub_ps(iz2,jz1);
1556             dx22             = _mm_sub_ps(ix2,jx2);
1557             dy22             = _mm_sub_ps(iy2,jy2);
1558             dz22             = _mm_sub_ps(iz2,jz2);
1559             dx23             = _mm_sub_ps(ix2,jx3);
1560             dy23             = _mm_sub_ps(iy2,jy3);
1561             dz23             = _mm_sub_ps(iz2,jz3);
1562             dx31             = _mm_sub_ps(ix3,jx1);
1563             dy31             = _mm_sub_ps(iy3,jy1);
1564             dz31             = _mm_sub_ps(iz3,jz1);
1565             dx32             = _mm_sub_ps(ix3,jx2);
1566             dy32             = _mm_sub_ps(iy3,jy2);
1567             dz32             = _mm_sub_ps(iz3,jz2);
1568             dx33             = _mm_sub_ps(ix3,jx3);
1569             dy33             = _mm_sub_ps(iy3,jy3);
1570             dz33             = _mm_sub_ps(iz3,jz3);
1571
1572             /* Calculate squared distance and things based on it */
1573             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1574             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1575             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1576             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1577             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1578             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1579             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1580             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1581             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1582
1583             rinv11           = avx128fma_invsqrt_f(rsq11);
1584             rinv12           = avx128fma_invsqrt_f(rsq12);
1585             rinv13           = avx128fma_invsqrt_f(rsq13);
1586             rinv21           = avx128fma_invsqrt_f(rsq21);
1587             rinv22           = avx128fma_invsqrt_f(rsq22);
1588             rinv23           = avx128fma_invsqrt_f(rsq23);
1589             rinv31           = avx128fma_invsqrt_f(rsq31);
1590             rinv32           = avx128fma_invsqrt_f(rsq32);
1591             rinv33           = avx128fma_invsqrt_f(rsq33);
1592
1593             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1594             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1595             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1596             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1597             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1598             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1599             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1600             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1601             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1602
1603             fjx1             = _mm_setzero_ps();
1604             fjy1             = _mm_setzero_ps();
1605             fjz1             = _mm_setzero_ps();
1606             fjx2             = _mm_setzero_ps();
1607             fjy2             = _mm_setzero_ps();
1608             fjz2             = _mm_setzero_ps();
1609             fjx3             = _mm_setzero_ps();
1610             fjy3             = _mm_setzero_ps();
1611             fjz3             = _mm_setzero_ps();
1612
1613             /**************************
1614              * CALCULATE INTERACTIONS *
1615              **************************/
1616
1617             if (gmx_mm_any_lt(rsq11,rcutoff2))
1618             {
1619
1620             /* REACTION-FIELD ELECTROSTATICS */
1621             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1622
1623             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1624
1625             fscal            = felec;
1626
1627             fscal            = _mm_and_ps(fscal,cutoff_mask);
1628
1629             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1630
1631              /* Update vectorial force */
1632             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1633             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1634             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1635
1636             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1637             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1638             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1639
1640             }
1641
1642             /**************************
1643              * CALCULATE INTERACTIONS *
1644              **************************/
1645
1646             if (gmx_mm_any_lt(rsq12,rcutoff2))
1647             {
1648
1649             /* REACTION-FIELD ELECTROSTATICS */
1650             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1651
1652             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1653
1654             fscal            = felec;
1655
1656             fscal            = _mm_and_ps(fscal,cutoff_mask);
1657
1658             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1659
1660              /* Update vectorial force */
1661             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1662             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1663             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1664
1665             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1666             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1667             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1668
1669             }
1670
1671             /**************************
1672              * CALCULATE INTERACTIONS *
1673              **************************/
1674
1675             if (gmx_mm_any_lt(rsq13,rcutoff2))
1676             {
1677
1678             /* REACTION-FIELD ELECTROSTATICS */
1679             felec            = _mm_mul_ps(qq13,_mm_msub_ps(rinv13,rinvsq13,krf2));
1680
1681             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
1682
1683             fscal            = felec;
1684
1685             fscal            = _mm_and_ps(fscal,cutoff_mask);
1686
1687             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1688
1689              /* Update vectorial force */
1690             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1691             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1692             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1693
1694             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1695             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1696             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1697
1698             }
1699
1700             /**************************
1701              * CALCULATE INTERACTIONS *
1702              **************************/
1703
1704             if (gmx_mm_any_lt(rsq21,rcutoff2))
1705             {
1706
1707             /* REACTION-FIELD ELECTROSTATICS */
1708             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1709
1710             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1711
1712             fscal            = felec;
1713
1714             fscal            = _mm_and_ps(fscal,cutoff_mask);
1715
1716             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1717
1718              /* Update vectorial force */
1719             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1720             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1721             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1722
1723             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1724             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1725             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1726
1727             }
1728
1729             /**************************
1730              * CALCULATE INTERACTIONS *
1731              **************************/
1732
1733             if (gmx_mm_any_lt(rsq22,rcutoff2))
1734             {
1735
1736             /* REACTION-FIELD ELECTROSTATICS */
1737             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1738
1739             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1740
1741             fscal            = felec;
1742
1743             fscal            = _mm_and_ps(fscal,cutoff_mask);
1744
1745             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1746
1747              /* Update vectorial force */
1748             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1749             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1750             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1751
1752             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1753             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1754             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1755
1756             }
1757
1758             /**************************
1759              * CALCULATE INTERACTIONS *
1760              **************************/
1761
1762             if (gmx_mm_any_lt(rsq23,rcutoff2))
1763             {
1764
1765             /* REACTION-FIELD ELECTROSTATICS */
1766             felec            = _mm_mul_ps(qq23,_mm_msub_ps(rinv23,rinvsq23,krf2));
1767
1768             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
1769
1770             fscal            = felec;
1771
1772             fscal            = _mm_and_ps(fscal,cutoff_mask);
1773
1774             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1775
1776              /* Update vectorial force */
1777             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1778             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1779             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1780
1781             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1782             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1783             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1784
1785             }
1786
1787             /**************************
1788              * CALCULATE INTERACTIONS *
1789              **************************/
1790
1791             if (gmx_mm_any_lt(rsq31,rcutoff2))
1792             {
1793
1794             /* REACTION-FIELD ELECTROSTATICS */
1795             felec            = _mm_mul_ps(qq31,_mm_msub_ps(rinv31,rinvsq31,krf2));
1796
1797             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
1798
1799             fscal            = felec;
1800
1801             fscal            = _mm_and_ps(fscal,cutoff_mask);
1802
1803             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1804
1805              /* Update vectorial force */
1806             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1807             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1808             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1809
1810             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1811             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1812             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1813
1814             }
1815
1816             /**************************
1817              * CALCULATE INTERACTIONS *
1818              **************************/
1819
1820             if (gmx_mm_any_lt(rsq32,rcutoff2))
1821             {
1822
1823             /* REACTION-FIELD ELECTROSTATICS */
1824             felec            = _mm_mul_ps(qq32,_mm_msub_ps(rinv32,rinvsq32,krf2));
1825
1826             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
1827
1828             fscal            = felec;
1829
1830             fscal            = _mm_and_ps(fscal,cutoff_mask);
1831
1832             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1833
1834              /* Update vectorial force */
1835             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1836             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1837             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1838
1839             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1840             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1841             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1842
1843             }
1844
1845             /**************************
1846              * CALCULATE INTERACTIONS *
1847              **************************/
1848
1849             if (gmx_mm_any_lt(rsq33,rcutoff2))
1850             {
1851
1852             /* REACTION-FIELD ELECTROSTATICS */
1853             felec            = _mm_mul_ps(qq33,_mm_msub_ps(rinv33,rinvsq33,krf2));
1854
1855             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
1856
1857             fscal            = felec;
1858
1859             fscal            = _mm_and_ps(fscal,cutoff_mask);
1860
1861             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1862
1863              /* Update vectorial force */
1864             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1865             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1866             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1867
1868             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1869             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1870             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1871
1872             }
1873
1874             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1875             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1876             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1877             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1878
1879             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1880                                                    fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1881
1882             /* Inner loop uses 297 flops */
1883         }
1884
1885         /* End of innermost loop */
1886
1887         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1888                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1889
1890         /* Increment number of inner iterations */
1891         inneriter                  += j_index_end - j_index_start;
1892
1893         /* Outer loop uses 18 flops */
1894     }
1895
1896     /* Increment number of outer iterations */
1897     outeriter        += nri;
1898
1899     /* Update outer/inner flops */
1900
1901     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*297);
1902 }