src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_VF_avx_128_fma_single
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  92     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  93     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  94     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  97     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  98     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  99     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     int              nvdwtype;
 107     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 108     int              *vdwtype;
 109     real             *vdwparam;
 110     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 111     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 112     __m128i          vfitab;
 113     __m128i          ifour       = _mm_set1_epi32(4);
 114     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 115     real             *vftab;
 116     __m128           dummy_mask,cutoff_mask;
 117     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 118     __m128           one     = _mm_set1_ps(1.0);
 119     __m128           two     = _mm_set1_ps(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm_set1_ps(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133     nvdwtype         = fr->ntype;
 134     vdwparam         = fr->nbfp;
 135     vdwtype          = mdatoms->typeA;
 136
 137     vftab            = kernel_data->table_vdw->data;
 138     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 143     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 144     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 145     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 146
 147     jq0              = _mm_set1_ps(charge[inr+0]);
 148     jq1              = _mm_set1_ps(charge[inr+1]);
 149     jq2              = _mm_set1_ps(charge[inr+2]);
 150     vdwjidx0A        = 2*vdwtype[inr+0];
 151     qq00             = _mm_mul_ps(iq0,jq0);
 152     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 153     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 154     qq01             = _mm_mul_ps(iq0,jq1);
 155     qq02             = _mm_mul_ps(iq0,jq2);
 156     qq10             = _mm_mul_ps(iq1,jq0);
 157     qq11             = _mm_mul_ps(iq1,jq1);
 158     qq12             = _mm_mul_ps(iq1,jq2);
 159     qq20             = _mm_mul_ps(iq2,jq0);
 160     qq21             = _mm_mul_ps(iq2,jq1);
 161     qq22             = _mm_mul_ps(iq2,jq2);
 162
 163     /* Avoid stupid compiler warnings */
 164     jnrA = jnrB = jnrC = jnrD = 0;
 165     j_coord_offsetA = 0;
 166     j_coord_offsetB = 0;
 167     j_coord_offsetC = 0;
 168     j_coord_offsetD = 0;
 169
 170     outeriter        = 0;
 171     inneriter        = 0;
 172
 173     for(iidx=0;iidx<4*DIM;iidx++)
 174     {
 175         scratch[iidx] = 0.0;
 176     }
 177
 178     /* Start outer loop over neighborlists */
 179     for(iidx=0; iidx<nri; iidx++)
 180     {
 181         /* Load shift vector for this list */
 182         i_shift_offset   = DIM*shiftidx[iidx];
 183
 184         /* Load limits for loop over neighbors */
 185         j_index_start    = jindex[iidx];
 186         j_index_end      = jindex[iidx+1];
 187
 188         /* Get outer coordinate index */
 189         inr              = iinr[iidx];
 190         i_coord_offset   = DIM*inr;
 191
 192         /* Load i particle coords and add shift vector */
 193         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 194                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 195
 196         fix0             = _mm_setzero_ps();
 197         fiy0             = _mm_setzero_ps();
 198         fiz0             = _mm_setzero_ps();
 199         fix1             = _mm_setzero_ps();
 200         fiy1             = _mm_setzero_ps();
 201         fiz1             = _mm_setzero_ps();
 202         fix2             = _mm_setzero_ps();
 203         fiy2             = _mm_setzero_ps();
 204         fiz2             = _mm_setzero_ps();
 205
 206         /* Reset potential sums */
 207         velecsum         = _mm_setzero_ps();
 208         vvdwsum          = _mm_setzero_ps();
 209
 210         /* Start inner kernel loop */
 211         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 212         {
 213
 214             /* Get j neighbor index, and coordinate index */
 215             jnrA             = jjnr[jidx];
 216             jnrB             = jjnr[jidx+1];
 217             jnrC             = jjnr[jidx+2];
 218             jnrD             = jjnr[jidx+3];
 219             j_coord_offsetA  = DIM*jnrA;
 220             j_coord_offsetB  = DIM*jnrB;
 221             j_coord_offsetC  = DIM*jnrC;
 222             j_coord_offsetD  = DIM*jnrD;
 223
 224             /* load j atom coordinates */
 225             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 226                                               x+j_coord_offsetC,x+j_coord_offsetD,
 227                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 228
 229             /* Calculate displacement vector */
 230             dx00             = _mm_sub_ps(ix0,jx0);
 231             dy00             = _mm_sub_ps(iy0,jy0);
 232             dz00             = _mm_sub_ps(iz0,jz0);
 233             dx01             = _mm_sub_ps(ix0,jx1);
 234             dy01             = _mm_sub_ps(iy0,jy1);
 235             dz01             = _mm_sub_ps(iz0,jz1);
 236             dx02             = _mm_sub_ps(ix0,jx2);
 237             dy02             = _mm_sub_ps(iy0,jy2);
 238             dz02             = _mm_sub_ps(iz0,jz2);
 239             dx10             = _mm_sub_ps(ix1,jx0);
 240             dy10             = _mm_sub_ps(iy1,jy0);
 241             dz10             = _mm_sub_ps(iz1,jz0);
 242             dx11             = _mm_sub_ps(ix1,jx1);
 243             dy11             = _mm_sub_ps(iy1,jy1);
 244             dz11             = _mm_sub_ps(iz1,jz1);
 245             dx12             = _mm_sub_ps(ix1,jx2);
 246             dy12             = _mm_sub_ps(iy1,jy2);
 247             dz12             = _mm_sub_ps(iz1,jz2);
 248             dx20             = _mm_sub_ps(ix2,jx0);
 249             dy20             = _mm_sub_ps(iy2,jy0);
 250             dz20             = _mm_sub_ps(iz2,jz0);
 251             dx21             = _mm_sub_ps(ix2,jx1);
 252             dy21             = _mm_sub_ps(iy2,jy1);
 253             dz21             = _mm_sub_ps(iz2,jz1);
 254             dx22             = _mm_sub_ps(ix2,jx2);
 255             dy22             = _mm_sub_ps(iy2,jy2);
 256             dz22             = _mm_sub_ps(iz2,jz2);
 257
 258             /* Calculate squared distance and things based on it */
 259             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 260             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 261             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 262             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 263             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 264             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 265             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 266             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 267             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 268
 269             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 270             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 271             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 272             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 273             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 274             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 275             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 276             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 277             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 278
 279             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 280             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 281             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 282             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 283             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 284             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 285             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 286             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 287             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 288
 289             fjx0             = _mm_setzero_ps();
 290             fjy0             = _mm_setzero_ps();
 291             fjz0             = _mm_setzero_ps();
 292             fjx1             = _mm_setzero_ps();
 293             fjy1             = _mm_setzero_ps();
 294             fjz1             = _mm_setzero_ps();
 295             fjx2             = _mm_setzero_ps();
 296             fjy2             = _mm_setzero_ps();
 297             fjz2             = _mm_setzero_ps();
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r00              = _mm_mul_ps(rsq00,rinv00);
 304
 305             /* Calculate table index by multiplying r with table scale and truncate to integer */
 306             rt               = _mm_mul_ps(r00,vftabscale);
 307             vfitab           = _mm_cvttps_epi32(rt);
 308 #ifdef __XOP__
 309             vfeps            = _mm_frcz_ps(rt);
 310 #else
 311             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 312 #endif
 313             twovfeps         = _mm_add_ps(vfeps,vfeps);
 314             vfitab           = _mm_slli_epi32(vfitab,3);
 315
 316             /* COULOMB ELECTROSTATICS */
 317             velec            = _mm_mul_ps(qq00,rinv00);
 318             felec            = _mm_mul_ps(velec,rinvsq00);
 319
 320             /* CUBIC SPLINE TABLE DISPERSION */
 321             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 322             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 323             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 324             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 325             _MM_TRANSPOSE4_PS(Y,F,G,H);
 326             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 327             VV               = _mm_macc_ps(vfeps,Fp,Y);
 328             vvdw6            = _mm_mul_ps(c6_00,VV);
 329             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 330             fvdw6            = _mm_mul_ps(c6_00,FF);
 331
 332             /* CUBIC SPLINE TABLE REPULSION */
 333             vfitab           = _mm_add_epi32(vfitab,ifour);
 334             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 335             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 336             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 337             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 338             _MM_TRANSPOSE4_PS(Y,F,G,H);
 339             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 340             VV               = _mm_macc_ps(vfeps,Fp,Y);
 341             vvdw12           = _mm_mul_ps(c12_00,VV);
 342             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 343             fvdw12           = _mm_mul_ps(c12_00,FF);
 344             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 345             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm_add_ps(velecsum,velec);
 349             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 350
 351             fscal            = _mm_add_ps(felec,fvdw);
 352
 353              /* Update vectorial force */
 354             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 355             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 356             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 357
 358             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 359             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 360             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             /* COULOMB ELECTROSTATICS */
 367             velec            = _mm_mul_ps(qq01,rinv01);
 368             felec            = _mm_mul_ps(velec,rinvsq01);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velecsum         = _mm_add_ps(velecsum,velec);
 372
 373             fscal            = felec;
 374
 375              /* Update vectorial force */
 376             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 377             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 378             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 379
 380             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 381             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 382             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             /* COULOMB ELECTROSTATICS */
 389             velec            = _mm_mul_ps(qq02,rinv02);
 390             felec            = _mm_mul_ps(velec,rinvsq02);
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397              /* Update vectorial force */
 398             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 399             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 400             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 401
 402             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 403             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 404             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             /* COULOMB ELECTROSTATICS */
 411             velec            = _mm_mul_ps(qq10,rinv10);
 412             felec            = _mm_mul_ps(velec,rinvsq10);
 413
 414             /* Update potential sum for this i atom from the interaction with this j atom. */
 415             velecsum         = _mm_add_ps(velecsum,velec);
 416
 417             fscal            = felec;
 418
 419              /* Update vectorial force */
 420             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 421             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 422             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 423
 424             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 425             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 426             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             /* COULOMB ELECTROSTATICS */
 433             velec            = _mm_mul_ps(qq11,rinv11);
 434             felec            = _mm_mul_ps(velec,rinvsq11);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velecsum         = _mm_add_ps(velecsum,velec);
 438
 439             fscal            = felec;
 440
 441              /* Update vectorial force */
 442             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 443             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 444             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 445
 446             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 447             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 448             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             /* COULOMB ELECTROSTATICS */
 455             velec            = _mm_mul_ps(qq12,rinv12);
 456             felec            = _mm_mul_ps(velec,rinvsq12);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velecsum         = _mm_add_ps(velecsum,velec);
 460
 461             fscal            = felec;
 462
 463              /* Update vectorial force */
 464             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 465             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 466             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 467
 468             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 469             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 470             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             /* COULOMB ELECTROSTATICS */
 477             velec            = _mm_mul_ps(qq20,rinv20);
 478             felec            = _mm_mul_ps(velec,rinvsq20);
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velecsum         = _mm_add_ps(velecsum,velec);
 482
 483             fscal            = felec;
 484
 485              /* Update vectorial force */
 486             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 487             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 488             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 489
 490             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 491             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 492             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             /* COULOMB ELECTROSTATICS */
 499             velec            = _mm_mul_ps(qq21,rinv21);
 500             felec            = _mm_mul_ps(velec,rinvsq21);
 501
 502             /* Update potential sum for this i atom from the interaction with this j atom. */
 503             velecsum         = _mm_add_ps(velecsum,velec);
 504
 505             fscal            = felec;
 506
 507              /* Update vectorial force */
 508             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 509             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 510             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 511
 512             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 513             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 514             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             /* COULOMB ELECTROSTATICS */
 521             velec            = _mm_mul_ps(qq22,rinv22);
 522             felec            = _mm_mul_ps(velec,rinvsq22);
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velecsum         = _mm_add_ps(velecsum,velec);
 526
 527             fscal            = felec;
 528
 529              /* Update vectorial force */
 530             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 531             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 532             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 533
 534             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 535             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 536             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 537
 538             fjptrA             = f+j_coord_offsetA;
 539             fjptrB             = f+j_coord_offsetB;
 540             fjptrC             = f+j_coord_offsetC;
 541             fjptrD             = f+j_coord_offsetD;
 542
 543             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 544                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 545
 546             /* Inner loop uses 314 flops */
 547         }
 548
 549         if(jidx<j_index_end)
 550         {
 551
 552             /* Get j neighbor index, and coordinate index */
 553             jnrlistA         = jjnr[jidx];
 554             jnrlistB         = jjnr[jidx+1];
 555             jnrlistC         = jjnr[jidx+2];
 556             jnrlistD         = jjnr[jidx+3];
 557             /* Sign of each element will be negative for non-real atoms.
 558              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 559              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 560              */
 561             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 562             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 563             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 564             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 565             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 566             j_coord_offsetA  = DIM*jnrA;
 567             j_coord_offsetB  = DIM*jnrB;
 568             j_coord_offsetC  = DIM*jnrC;
 569             j_coord_offsetD  = DIM*jnrD;
 570
 571             /* load j atom coordinates */
 572             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 573                                               x+j_coord_offsetC,x+j_coord_offsetD,
 574                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 575
 576             /* Calculate displacement vector */
 577             dx00             = _mm_sub_ps(ix0,jx0);
 578             dy00             = _mm_sub_ps(iy0,jy0);
 579             dz00             = _mm_sub_ps(iz0,jz0);
 580             dx01             = _mm_sub_ps(ix0,jx1);
 581             dy01             = _mm_sub_ps(iy0,jy1);
 582             dz01             = _mm_sub_ps(iz0,jz1);
 583             dx02             = _mm_sub_ps(ix0,jx2);
 584             dy02             = _mm_sub_ps(iy0,jy2);
 585             dz02             = _mm_sub_ps(iz0,jz2);
 586             dx10             = _mm_sub_ps(ix1,jx0);
 587             dy10             = _mm_sub_ps(iy1,jy0);
 588             dz10             = _mm_sub_ps(iz1,jz0);
 589             dx11             = _mm_sub_ps(ix1,jx1);
 590             dy11             = _mm_sub_ps(iy1,jy1);
 591             dz11             = _mm_sub_ps(iz1,jz1);
 592             dx12             = _mm_sub_ps(ix1,jx2);
 593             dy12             = _mm_sub_ps(iy1,jy2);
 594             dz12             = _mm_sub_ps(iz1,jz2);
 595             dx20             = _mm_sub_ps(ix2,jx0);
 596             dy20             = _mm_sub_ps(iy2,jy0);
 597             dz20             = _mm_sub_ps(iz2,jz0);
 598             dx21             = _mm_sub_ps(ix2,jx1);
 599             dy21             = _mm_sub_ps(iy2,jy1);
 600             dz21             = _mm_sub_ps(iz2,jz1);
 601             dx22             = _mm_sub_ps(ix2,jx2);
 602             dy22             = _mm_sub_ps(iy2,jy2);
 603             dz22             = _mm_sub_ps(iz2,jz2);
 604
 605             /* Calculate squared distance and things based on it */
 606             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 607             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 608             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 609             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 610             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 611             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 612             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 613             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 614             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 615
 616             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 617             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 618             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 619             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 620             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 621             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 622             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 623             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 624             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 625
 626             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 627             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 628             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 629             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 630             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 631             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 632             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 633             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 634             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 635
 636             fjx0             = _mm_setzero_ps();
 637             fjy0             = _mm_setzero_ps();
 638             fjz0             = _mm_setzero_ps();
 639             fjx1             = _mm_setzero_ps();
 640             fjy1             = _mm_setzero_ps();
 641             fjz1             = _mm_setzero_ps();
 642             fjx2             = _mm_setzero_ps();
 643             fjy2             = _mm_setzero_ps();
 644             fjz2             = _mm_setzero_ps();
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             r00              = _mm_mul_ps(rsq00,rinv00);
 651             r00              = _mm_andnot_ps(dummy_mask,r00);
 652
 653             /* Calculate table index by multiplying r with table scale and truncate to integer */
 654             rt               = _mm_mul_ps(r00,vftabscale);
 655             vfitab           = _mm_cvttps_epi32(rt);
 656 #ifdef __XOP__
 657             vfeps            = _mm_frcz_ps(rt);
 658 #else
 659             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 660 #endif
 661             twovfeps         = _mm_add_ps(vfeps,vfeps);
 662             vfitab           = _mm_slli_epi32(vfitab,3);
 663
 664             /* COULOMB ELECTROSTATICS */
 665             velec            = _mm_mul_ps(qq00,rinv00);
 666             felec            = _mm_mul_ps(velec,rinvsq00);
 667
 668             /* CUBIC SPLINE TABLE DISPERSION */
 669             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 670             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 671             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 672             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 673             _MM_TRANSPOSE4_PS(Y,F,G,H);
 674             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 675             VV               = _mm_macc_ps(vfeps,Fp,Y);
 676             vvdw6            = _mm_mul_ps(c6_00,VV);
 677             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 678             fvdw6            = _mm_mul_ps(c6_00,FF);
 679
 680             /* CUBIC SPLINE TABLE REPULSION */
 681             vfitab           = _mm_add_epi32(vfitab,ifour);
 682             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 683             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 684             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 685             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 686             _MM_TRANSPOSE4_PS(Y,F,G,H);
 687             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 688             VV               = _mm_macc_ps(vfeps,Fp,Y);
 689             vvdw12           = _mm_mul_ps(c12_00,VV);
 690             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 691             fvdw12           = _mm_mul_ps(c12_00,FF);
 692             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 693             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 694
 695             /* Update potential sum for this i atom from the interaction with this j atom. */
 696             velec            = _mm_andnot_ps(dummy_mask,velec);
 697             velecsum         = _mm_add_ps(velecsum,velec);
 698             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 699             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 700
 701             fscal            = _mm_add_ps(felec,fvdw);
 702
 703             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 704
 705              /* Update vectorial force */
 706             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 707             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 708             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 709
 710             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 711             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 712             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 713
 714             /**************************
 715              * CALCULATE INTERACTIONS *
 716              **************************/
 717
 718             /* COULOMB ELECTROSTATICS */
 719             velec            = _mm_mul_ps(qq01,rinv01);
 720             felec            = _mm_mul_ps(velec,rinvsq01);
 721
 722             /* Update potential sum for this i atom from the interaction with this j atom. */
 723             velec            = _mm_andnot_ps(dummy_mask,velec);
 724             velecsum         = _mm_add_ps(velecsum,velec);
 725
 726             fscal            = felec;
 727
 728             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 729
 730              /* Update vectorial force */
 731             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 732             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 733             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 734
 735             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 736             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 737             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 738
 739             /**************************
 740              * CALCULATE INTERACTIONS *
 741              **************************/
 742
 743             /* COULOMB ELECTROSTATICS */
 744             velec            = _mm_mul_ps(qq02,rinv02);
 745             felec            = _mm_mul_ps(velec,rinvsq02);
 746
 747             /* Update potential sum for this i atom from the interaction with this j atom. */
 748             velec            = _mm_andnot_ps(dummy_mask,velec);
 749             velecsum         = _mm_add_ps(velecsum,velec);
 750
 751             fscal            = felec;
 752
 753             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 754
 755              /* Update vectorial force */
 756             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 757             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 758             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 759
 760             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 761             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 762             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             /* COULOMB ELECTROSTATICS */
 769             velec            = _mm_mul_ps(qq10,rinv10);
 770             felec            = _mm_mul_ps(velec,rinvsq10);
 771
 772             /* Update potential sum for this i atom from the interaction with this j atom. */
 773             velec            = _mm_andnot_ps(dummy_mask,velec);
 774             velecsum         = _mm_add_ps(velecsum,velec);
 775
 776             fscal            = felec;
 777
 778             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 779
 780              /* Update vectorial force */
 781             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 782             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 783             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 784
 785             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 786             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 787             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 788
 789             /**************************
 790              * CALCULATE INTERACTIONS *
 791              **************************/
 792
 793             /* COULOMB ELECTROSTATICS */
 794             velec            = _mm_mul_ps(qq11,rinv11);
 795             felec            = _mm_mul_ps(velec,rinvsq11);
 796
 797             /* Update potential sum for this i atom from the interaction with this j atom. */
 798             velec            = _mm_andnot_ps(dummy_mask,velec);
 799             velecsum         = _mm_add_ps(velecsum,velec);
 800
 801             fscal            = felec;
 802
 803             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 804
 805              /* Update vectorial force */
 806             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 807             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 808             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 809
 810             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 811             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 812             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 813
 814             /**************************
 815              * CALCULATE INTERACTIONS *
 816              **************************/
 817
 818             /* COULOMB ELECTROSTATICS */
 819             velec            = _mm_mul_ps(qq12,rinv12);
 820             felec            = _mm_mul_ps(velec,rinvsq12);
 821
 822             /* Update potential sum for this i atom from the interaction with this j atom. */
 823             velec            = _mm_andnot_ps(dummy_mask,velec);
 824             velecsum         = _mm_add_ps(velecsum,velec);
 825
 826             fscal            = felec;
 827
 828             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 829
 830              /* Update vectorial force */
 831             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 832             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 833             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 834
 835             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 836             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 837             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 838
 839             /**************************
 840              * CALCULATE INTERACTIONS *
 841              **************************/
 842
 843             /* COULOMB ELECTROSTATICS */
 844             velec            = _mm_mul_ps(qq20,rinv20);
 845             felec            = _mm_mul_ps(velec,rinvsq20);
 846
 847             /* Update potential sum for this i atom from the interaction with this j atom. */
 848             velec            = _mm_andnot_ps(dummy_mask,velec);
 849             velecsum         = _mm_add_ps(velecsum,velec);
 850
 851             fscal            = felec;
 852
 853             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 854
 855              /* Update vectorial force */
 856             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 857             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 858             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 859
 860             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 861             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 862             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             /* COULOMB ELECTROSTATICS */
 869             velec            = _mm_mul_ps(qq21,rinv21);
 870             felec            = _mm_mul_ps(velec,rinvsq21);
 871
 872             /* Update potential sum for this i atom from the interaction with this j atom. */
 873             velec            = _mm_andnot_ps(dummy_mask,velec);
 874             velecsum         = _mm_add_ps(velecsum,velec);
 875
 876             fscal            = felec;
 877
 878             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 879
 880              /* Update vectorial force */
 881             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 882             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 883             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 884
 885             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 886             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 887             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 888
 889             /**************************
 890              * CALCULATE INTERACTIONS *
 891              **************************/
 892
 893             /* COULOMB ELECTROSTATICS */
 894             velec            = _mm_mul_ps(qq22,rinv22);
 895             felec            = _mm_mul_ps(velec,rinvsq22);
 896
 897             /* Update potential sum for this i atom from the interaction with this j atom. */
 898             velec            = _mm_andnot_ps(dummy_mask,velec);
 899             velecsum         = _mm_add_ps(velecsum,velec);
 900
 901             fscal            = felec;
 902
 903             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 904
 905              /* Update vectorial force */
 906             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 907             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 908             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 909
 910             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 911             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 912             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 913
 914             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 915             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 916             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 917             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 918
 919             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 920                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 921
 922             /* Inner loop uses 315 flops */
 923         }
 924
 925         /* End of innermost loop */
 926
 927         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 928                                               f+i_coord_offset,fshift+i_shift_offset);
 929
 930         ggid                        = gid[iidx];
 931         /* Update potential energies */
 932         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 933         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 934
 935         /* Increment number of inner iterations */
 936         inneriter                  += j_index_end - j_index_start;
 937
 938         /* Outer loop uses 20 flops */
 939     }
 940
 941     /* Increment number of outer iterations */
 942     outeriter        += nri;
 943
 944     /* Update outer/inner flops */
 945
 946     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*315);
 947 }
 948 /*
 949  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_128_fma_single
 950  * Electrostatics interaction: Coulomb
 951  * VdW interaction:            CubicSplineTable
 952  * Geometry:                   Water3-Water3
 953  * Calculate force/pot:        Force
 954  */
 955 void
 956 nb_kernel_ElecCoul_VdwCSTab_GeomW3W3_F_avx_128_fma_single
 957                     (t_nblist                    * gmx_restrict       nlist,
 958                      rvec                        * gmx_restrict          xx,
 959                      rvec                        * gmx_restrict          ff,
 960                      t_forcerec                  * gmx_restrict          fr,
 961                      t_mdatoms                   * gmx_restrict     mdatoms,
 962                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 963                      t_nrnb                      * gmx_restrict        nrnb)
 964 {
 965     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 966      * just 0 for non-waters.
 967      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 968      * jnr indices corresponding to data put in the four positions in the SIMD register.
 969      */
 970     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 971     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 972     int              jnrA,jnrB,jnrC,jnrD;
 973     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 974     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 975     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 976     real             rcutoff_scalar;
 977     real             *shiftvec,*fshift,*x,*f;
 978     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 979     real             scratch[4*DIM];
 980     __m128           fscal,rcutoff,rcutoff2,jidxall;
 981     int              vdwioffset0;
 982     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 983     int              vdwioffset1;
 984     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 985     int              vdwioffset2;
 986     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 987     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 988     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 989     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 990     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 991     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 992     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 993     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 994     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 995     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 996     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 997     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 998     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 999     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1000     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1001     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1002     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1003     real             *charge;
1004     int              nvdwtype;
1005     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1006     int              *vdwtype;
1007     real             *vdwparam;
1008     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1009     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1010     __m128i          vfitab;
1011     __m128i          ifour       = _mm_set1_epi32(4);
1012     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
1013     real             *vftab;
1014     __m128           dummy_mask,cutoff_mask;
1015     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1016     __m128           one     = _mm_set1_ps(1.0);
1017     __m128           two     = _mm_set1_ps(2.0);
1018     x                = xx[0];
1019     f                = ff[0];
1020
1021     nri              = nlist->nri;
1022     iinr             = nlist->iinr;
1023     jindex           = nlist->jindex;
1024     jjnr             = nlist->jjnr;
1025     shiftidx         = nlist->shift;
1026     gid              = nlist->gid;
1027     shiftvec         = fr->shift_vec[0];
1028     fshift           = fr->fshift[0];
1029     facel            = _mm_set1_ps(fr->epsfac);
1030     charge           = mdatoms->chargeA;
1031     nvdwtype         = fr->ntype;
1032     vdwparam         = fr->nbfp;
1033     vdwtype          = mdatoms->typeA;
1034
1035     vftab            = kernel_data->table_vdw->data;
1036     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
1037
1038     /* Setup water-specific parameters */
1039     inr              = nlist->iinr[0];
1040     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1041     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1042     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1043     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1044
1045     jq0              = _mm_set1_ps(charge[inr+0]);
1046     jq1              = _mm_set1_ps(charge[inr+1]);
1047     jq2              = _mm_set1_ps(charge[inr+2]);
1048     vdwjidx0A        = 2*vdwtype[inr+0];
1049     qq00             = _mm_mul_ps(iq0,jq0);
1050     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1051     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1052     qq01             = _mm_mul_ps(iq0,jq1);
1053     qq02             = _mm_mul_ps(iq0,jq2);
1054     qq10             = _mm_mul_ps(iq1,jq0);
1055     qq11             = _mm_mul_ps(iq1,jq1);
1056     qq12             = _mm_mul_ps(iq1,jq2);
1057     qq20             = _mm_mul_ps(iq2,jq0);
1058     qq21             = _mm_mul_ps(iq2,jq1);
1059     qq22             = _mm_mul_ps(iq2,jq2);
1060
1061     /* Avoid stupid compiler warnings */
1062     jnrA = jnrB = jnrC = jnrD = 0;
1063     j_coord_offsetA = 0;
1064     j_coord_offsetB = 0;
1065     j_coord_offsetC = 0;
1066     j_coord_offsetD = 0;
1067
1068     outeriter        = 0;
1069     inneriter        = 0;
1070
1071     for(iidx=0;iidx<4*DIM;iidx++)
1072     {
1073         scratch[iidx] = 0.0;
1074     }
1075
1076     /* Start outer loop over neighborlists */
1077     for(iidx=0; iidx<nri; iidx++)
1078     {
1079         /* Load shift vector for this list */
1080         i_shift_offset   = DIM*shiftidx[iidx];
1081
1082         /* Load limits for loop over neighbors */
1083         j_index_start    = jindex[iidx];
1084         j_index_end      = jindex[iidx+1];
1085
1086         /* Get outer coordinate index */
1087         inr              = iinr[iidx];
1088         i_coord_offset   = DIM*inr;
1089
1090         /* Load i particle coords and add shift vector */
1091         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1092                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1093
1094         fix0             = _mm_setzero_ps();
1095         fiy0             = _mm_setzero_ps();
1096         fiz0             = _mm_setzero_ps();
1097         fix1             = _mm_setzero_ps();
1098         fiy1             = _mm_setzero_ps();
1099         fiz1             = _mm_setzero_ps();
1100         fix2             = _mm_setzero_ps();
1101         fiy2             = _mm_setzero_ps();
1102         fiz2             = _mm_setzero_ps();
1103
1104         /* Start inner kernel loop */
1105         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1106         {
1107
1108             /* Get j neighbor index, and coordinate index */
1109             jnrA             = jjnr[jidx];
1110             jnrB             = jjnr[jidx+1];
1111             jnrC             = jjnr[jidx+2];
1112             jnrD             = jjnr[jidx+3];
1113             j_coord_offsetA  = DIM*jnrA;
1114             j_coord_offsetB  = DIM*jnrB;
1115             j_coord_offsetC  = DIM*jnrC;
1116             j_coord_offsetD  = DIM*jnrD;
1117
1118             /* load j atom coordinates */
1119             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1120                                               x+j_coord_offsetC,x+j_coord_offsetD,
1121                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1122
1123             /* Calculate displacement vector */
1124             dx00             = _mm_sub_ps(ix0,jx0);
1125             dy00             = _mm_sub_ps(iy0,jy0);
1126             dz00             = _mm_sub_ps(iz0,jz0);
1127             dx01             = _mm_sub_ps(ix0,jx1);
1128             dy01             = _mm_sub_ps(iy0,jy1);
1129             dz01             = _mm_sub_ps(iz0,jz1);
1130             dx02             = _mm_sub_ps(ix0,jx2);
1131             dy02             = _mm_sub_ps(iy0,jy2);
1132             dz02             = _mm_sub_ps(iz0,jz2);
1133             dx10             = _mm_sub_ps(ix1,jx0);
1134             dy10             = _mm_sub_ps(iy1,jy0);
1135             dz10             = _mm_sub_ps(iz1,jz0);
1136             dx11             = _mm_sub_ps(ix1,jx1);
1137             dy11             = _mm_sub_ps(iy1,jy1);
1138             dz11             = _mm_sub_ps(iz1,jz1);
1139             dx12             = _mm_sub_ps(ix1,jx2);
1140             dy12             = _mm_sub_ps(iy1,jy2);
1141             dz12             = _mm_sub_ps(iz1,jz2);
1142             dx20             = _mm_sub_ps(ix2,jx0);
1143             dy20             = _mm_sub_ps(iy2,jy0);
1144             dz20             = _mm_sub_ps(iz2,jz0);
1145             dx21             = _mm_sub_ps(ix2,jx1);
1146             dy21             = _mm_sub_ps(iy2,jy1);
1147             dz21             = _mm_sub_ps(iz2,jz1);
1148             dx22             = _mm_sub_ps(ix2,jx2);
1149             dy22             = _mm_sub_ps(iy2,jy2);
1150             dz22             = _mm_sub_ps(iz2,jz2);
1151
1152             /* Calculate squared distance and things based on it */
1153             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1154             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1155             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1156             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1157             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1158             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1159             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1160             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1161             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1162
1163             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1164             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1165             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1166             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1167             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1168             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1169             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1170             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1171             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1172
1173             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1174             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1175             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1176             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1177             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1178             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1179             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1180             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1181             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1182
1183             fjx0             = _mm_setzero_ps();
1184             fjy0             = _mm_setzero_ps();
1185             fjz0             = _mm_setzero_ps();
1186             fjx1             = _mm_setzero_ps();
1187             fjy1             = _mm_setzero_ps();
1188             fjz1             = _mm_setzero_ps();
1189             fjx2             = _mm_setzero_ps();
1190             fjy2             = _mm_setzero_ps();
1191             fjz2             = _mm_setzero_ps();
1192
1193             /**************************
1194              * CALCULATE INTERACTIONS *
1195              **************************/
1196
1197             r00              = _mm_mul_ps(rsq00,rinv00);
1198
1199             /* Calculate table index by multiplying r with table scale and truncate to integer */
1200             rt               = _mm_mul_ps(r00,vftabscale);
1201             vfitab           = _mm_cvttps_epi32(rt);
1202 #ifdef __XOP__
1203             vfeps            = _mm_frcz_ps(rt);
1204 #else
1205             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1206 #endif
1207             twovfeps         = _mm_add_ps(vfeps,vfeps);
1208             vfitab           = _mm_slli_epi32(vfitab,3);
1209
1210             /* COULOMB ELECTROSTATICS */
1211             velec            = _mm_mul_ps(qq00,rinv00);
1212             felec            = _mm_mul_ps(velec,rinvsq00);
1213
1214             /* CUBIC SPLINE TABLE DISPERSION */
1215             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1216             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1217             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1218             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1219             _MM_TRANSPOSE4_PS(Y,F,G,H);
1220             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1221             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1222             fvdw6            = _mm_mul_ps(c6_00,FF);
1223
1224             /* CUBIC SPLINE TABLE REPULSION */
1225             vfitab           = _mm_add_epi32(vfitab,ifour);
1226             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1227             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1228             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1229             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1230             _MM_TRANSPOSE4_PS(Y,F,G,H);
1231             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1232             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1233             fvdw12           = _mm_mul_ps(c12_00,FF);
1234             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1235
1236             fscal            = _mm_add_ps(felec,fvdw);
1237
1238              /* Update vectorial force */
1239             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1240             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1241             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1242
1243             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1244             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1245             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1246
1247             /**************************
1248              * CALCULATE INTERACTIONS *
1249              **************************/
1250
1251             /* COULOMB ELECTROSTATICS */
1252             velec            = _mm_mul_ps(qq01,rinv01);
1253             felec            = _mm_mul_ps(velec,rinvsq01);
1254
1255             fscal            = felec;
1256
1257              /* Update vectorial force */
1258             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1259             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1260             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1261
1262             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1263             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1264             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1265
1266             /**************************
1267              * CALCULATE INTERACTIONS *
1268              **************************/
1269
1270             /* COULOMB ELECTROSTATICS */
1271             velec            = _mm_mul_ps(qq02,rinv02);
1272             felec            = _mm_mul_ps(velec,rinvsq02);
1273
1274             fscal            = felec;
1275
1276              /* Update vectorial force */
1277             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1278             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1279             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1280
1281             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1282             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1283             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1284
1285             /**************************
1286              * CALCULATE INTERACTIONS *
1287              **************************/
1288
1289             /* COULOMB ELECTROSTATICS */
1290             velec            = _mm_mul_ps(qq10,rinv10);
1291             felec            = _mm_mul_ps(velec,rinvsq10);
1292
1293             fscal            = felec;
1294
1295              /* Update vectorial force */
1296             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1297             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1298             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1299
1300             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1301             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1302             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1303
1304             /**************************
1305              * CALCULATE INTERACTIONS *
1306              **************************/
1307
1308             /* COULOMB ELECTROSTATICS */
1309             velec            = _mm_mul_ps(qq11,rinv11);
1310             felec            = _mm_mul_ps(velec,rinvsq11);
1311
1312             fscal            = felec;
1313
1314              /* Update vectorial force */
1315             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1316             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1317             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1318
1319             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1320             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1321             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1322
1323             /**************************
1324              * CALCULATE INTERACTIONS *
1325              **************************/
1326
1327             /* COULOMB ELECTROSTATICS */
1328             velec            = _mm_mul_ps(qq12,rinv12);
1329             felec            = _mm_mul_ps(velec,rinvsq12);
1330
1331             fscal            = felec;
1332
1333              /* Update vectorial force */
1334             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1335             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1336             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1337
1338             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1339             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1340             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1341
1342             /**************************
1343              * CALCULATE INTERACTIONS *
1344              **************************/
1345
1346             /* COULOMB ELECTROSTATICS */
1347             velec            = _mm_mul_ps(qq20,rinv20);
1348             felec            = _mm_mul_ps(velec,rinvsq20);
1349
1350             fscal            = felec;
1351
1352              /* Update vectorial force */
1353             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1354             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1355             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1356
1357             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1358             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1359             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1360
1361             /**************************
1362              * CALCULATE INTERACTIONS *
1363              **************************/
1364
1365             /* COULOMB ELECTROSTATICS */
1366             velec            = _mm_mul_ps(qq21,rinv21);
1367             felec            = _mm_mul_ps(velec,rinvsq21);
1368
1369             fscal            = felec;
1370
1371              /* Update vectorial force */
1372             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1373             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1374             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1375
1376             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1377             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1378             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1379
1380             /**************************
1381              * CALCULATE INTERACTIONS *
1382              **************************/
1383
1384             /* COULOMB ELECTROSTATICS */
1385             velec            = _mm_mul_ps(qq22,rinv22);
1386             felec            = _mm_mul_ps(velec,rinvsq22);
1387
1388             fscal            = felec;
1389
1390              /* Update vectorial force */
1391             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1392             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1393             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1394
1395             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1396             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1397             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1398
1399             fjptrA             = f+j_coord_offsetA;
1400             fjptrB             = f+j_coord_offsetB;
1401             fjptrC             = f+j_coord_offsetC;
1402             fjptrD             = f+j_coord_offsetD;
1403
1404             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1405                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1406
1407             /* Inner loop uses 297 flops */
1408         }
1409
1410         if(jidx<j_index_end)
1411         {
1412
1413             /* Get j neighbor index, and coordinate index */
1414             jnrlistA         = jjnr[jidx];
1415             jnrlistB         = jjnr[jidx+1];
1416             jnrlistC         = jjnr[jidx+2];
1417             jnrlistD         = jjnr[jidx+3];
1418             /* Sign of each element will be negative for non-real atoms.
1419              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1420              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1421              */
1422             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1423             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1424             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1425             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1426             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1427             j_coord_offsetA  = DIM*jnrA;
1428             j_coord_offsetB  = DIM*jnrB;
1429             j_coord_offsetC  = DIM*jnrC;
1430             j_coord_offsetD  = DIM*jnrD;
1431
1432             /* load j atom coordinates */
1433             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1434                                               x+j_coord_offsetC,x+j_coord_offsetD,
1435                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1436
1437             /* Calculate displacement vector */
1438             dx00             = _mm_sub_ps(ix0,jx0);
1439             dy00             = _mm_sub_ps(iy0,jy0);
1440             dz00             = _mm_sub_ps(iz0,jz0);
1441             dx01             = _mm_sub_ps(ix0,jx1);
1442             dy01             = _mm_sub_ps(iy0,jy1);
1443             dz01             = _mm_sub_ps(iz0,jz1);
1444             dx02             = _mm_sub_ps(ix0,jx2);
1445             dy02             = _mm_sub_ps(iy0,jy2);
1446             dz02             = _mm_sub_ps(iz0,jz2);
1447             dx10             = _mm_sub_ps(ix1,jx0);
1448             dy10             = _mm_sub_ps(iy1,jy0);
1449             dz10             = _mm_sub_ps(iz1,jz0);
1450             dx11             = _mm_sub_ps(ix1,jx1);
1451             dy11             = _mm_sub_ps(iy1,jy1);
1452             dz11             = _mm_sub_ps(iz1,jz1);
1453             dx12             = _mm_sub_ps(ix1,jx2);
1454             dy12             = _mm_sub_ps(iy1,jy2);
1455             dz12             = _mm_sub_ps(iz1,jz2);
1456             dx20             = _mm_sub_ps(ix2,jx0);
1457             dy20             = _mm_sub_ps(iy2,jy0);
1458             dz20             = _mm_sub_ps(iz2,jz0);
1459             dx21             = _mm_sub_ps(ix2,jx1);
1460             dy21             = _mm_sub_ps(iy2,jy1);
1461             dz21             = _mm_sub_ps(iz2,jz1);
1462             dx22             = _mm_sub_ps(ix2,jx2);
1463             dy22             = _mm_sub_ps(iy2,jy2);
1464             dz22             = _mm_sub_ps(iz2,jz2);
1465
1466             /* Calculate squared distance and things based on it */
1467             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1468             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1469             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1470             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1471             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1472             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1473             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1474             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1475             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1476
1477             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1478             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1479             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1480             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1481             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1482             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1483             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1484             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1485             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1486
1487             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1488             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1489             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1490             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1491             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1492             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1493             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1494             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1495             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1496
1497             fjx0             = _mm_setzero_ps();
1498             fjy0             = _mm_setzero_ps();
1499             fjz0             = _mm_setzero_ps();
1500             fjx1             = _mm_setzero_ps();
1501             fjy1             = _mm_setzero_ps();
1502             fjz1             = _mm_setzero_ps();
1503             fjx2             = _mm_setzero_ps();
1504             fjy2             = _mm_setzero_ps();
1505             fjz2             = _mm_setzero_ps();
1506
1507             /**************************
1508              * CALCULATE INTERACTIONS *
1509              **************************/
1510
1511             r00              = _mm_mul_ps(rsq00,rinv00);
1512             r00              = _mm_andnot_ps(dummy_mask,r00);
1513
1514             /* Calculate table index by multiplying r with table scale and truncate to integer */
1515             rt               = _mm_mul_ps(r00,vftabscale);
1516             vfitab           = _mm_cvttps_epi32(rt);
1517 #ifdef __XOP__
1518             vfeps            = _mm_frcz_ps(rt);
1519 #else
1520             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1521 #endif
1522             twovfeps         = _mm_add_ps(vfeps,vfeps);
1523             vfitab           = _mm_slli_epi32(vfitab,3);
1524
1525             /* COULOMB ELECTROSTATICS */
1526             velec            = _mm_mul_ps(qq00,rinv00);
1527             felec            = _mm_mul_ps(velec,rinvsq00);
1528
1529             /* CUBIC SPLINE TABLE DISPERSION */
1530             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1531             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1532             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1533             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1534             _MM_TRANSPOSE4_PS(Y,F,G,H);
1535             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1536             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1537             fvdw6            = _mm_mul_ps(c6_00,FF);
1538
1539             /* CUBIC SPLINE TABLE REPULSION */
1540             vfitab           = _mm_add_epi32(vfitab,ifour);
1541             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1542             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1543             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1544             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1545             _MM_TRANSPOSE4_PS(Y,F,G,H);
1546             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1547             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1548             fvdw12           = _mm_mul_ps(c12_00,FF);
1549             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1550
1551             fscal            = _mm_add_ps(felec,fvdw);
1552
1553             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1554
1555              /* Update vectorial force */
1556             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1557             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1558             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1559
1560             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1561             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1562             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1563
1564             /**************************
1565              * CALCULATE INTERACTIONS *
1566              **************************/
1567
1568             /* COULOMB ELECTROSTATICS */
1569             velec            = _mm_mul_ps(qq01,rinv01);
1570             felec            = _mm_mul_ps(velec,rinvsq01);
1571
1572             fscal            = felec;
1573
1574             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1575
1576              /* Update vectorial force */
1577             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1578             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1579             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1580
1581             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1582             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1583             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1584
1585             /**************************
1586              * CALCULATE INTERACTIONS *
1587              **************************/
1588
1589             /* COULOMB ELECTROSTATICS */
1590             velec            = _mm_mul_ps(qq02,rinv02);
1591             felec            = _mm_mul_ps(velec,rinvsq02);
1592
1593             fscal            = felec;
1594
1595             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1596
1597              /* Update vectorial force */
1598             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1599             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1600             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1601
1602             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1603             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1604             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1605
1606             /**************************
1607              * CALCULATE INTERACTIONS *
1608              **************************/
1609
1610             /* COULOMB ELECTROSTATICS */
1611             velec            = _mm_mul_ps(qq10,rinv10);
1612             felec            = _mm_mul_ps(velec,rinvsq10);
1613
1614             fscal            = felec;
1615
1616             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1617
1618              /* Update vectorial force */
1619             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1620             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1621             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1622
1623             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1624             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1625             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1626
1627             /**************************
1628              * CALCULATE INTERACTIONS *
1629              **************************/
1630
1631             /* COULOMB ELECTROSTATICS */
1632             velec            = _mm_mul_ps(qq11,rinv11);
1633             felec            = _mm_mul_ps(velec,rinvsq11);
1634
1635             fscal            = felec;
1636
1637             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1638
1639              /* Update vectorial force */
1640             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1641             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1642             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1643
1644             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1645             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1646             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1647
1648             /**************************
1649              * CALCULATE INTERACTIONS *
1650              **************************/
1651
1652             /* COULOMB ELECTROSTATICS */
1653             velec            = _mm_mul_ps(qq12,rinv12);
1654             felec            = _mm_mul_ps(velec,rinvsq12);
1655
1656             fscal            = felec;
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              * CALCULATE INTERACTIONS *
1671              **************************/
1672
1673             /* COULOMB ELECTROSTATICS */
1674             velec            = _mm_mul_ps(qq20,rinv20);
1675             felec            = _mm_mul_ps(velec,rinvsq20);
1676
1677             fscal            = felec;
1678
1679             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1680
1681              /* Update vectorial force */
1682             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1683             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1684             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1685
1686             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1687             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1688             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1689
1690             /**************************
1691              * CALCULATE INTERACTIONS *
1692              **************************/
1693
1694             /* COULOMB ELECTROSTATICS */
1695             velec            = _mm_mul_ps(qq21,rinv21);
1696             felec            = _mm_mul_ps(velec,rinvsq21);
1697
1698             fscal            = felec;
1699
1700             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1701
1702              /* Update vectorial force */
1703             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1704             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1705             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1706
1707             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1708             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1709             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1710
1711             /**************************
1712              * CALCULATE INTERACTIONS *
1713              **************************/
1714
1715             /* COULOMB ELECTROSTATICS */
1716             velec            = _mm_mul_ps(qq22,rinv22);
1717             felec            = _mm_mul_ps(velec,rinvsq22);
1718
1719             fscal            = felec;
1720
1721             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1722
1723              /* Update vectorial force */
1724             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1725             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1726             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1727
1728             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1729             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1730             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1731
1732             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1733             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1734             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1735             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1736
1737             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1738                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1739
1740             /* Inner loop uses 298 flops */
1741         }
1742
1743         /* End of innermost loop */
1744
1745         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1746                                               f+i_coord_offset,fshift+i_shift_offset);
1747
1748         /* Increment number of inner iterations */
1749         inneriter                  += j_index_end - j_index_start;
1750
1751         /* Outer loop uses 18 flops */
1752     }
1753
1754     /* Increment number of outer iterations */
1755     outeriter        += nri;
1756
1757     /* Update outer/inner flops */
1758
1759     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*298);
1760 }