src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "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_ElecEw_VdwCSTab_GeomW3W3_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_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     __m128i          ewitab;
 117     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 118     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 119     real             *ewtab;
 120     __m128           dummy_mask,cutoff_mask;
 121     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 122     __m128           one     = _mm_set1_ps(1.0);
 123     __m128           two     = _mm_set1_ps(2.0);
 124     x                = xx[0];
 125     f                = ff[0];
 126
 127     nri              = nlist->nri;
 128     iinr             = nlist->iinr;
 129     jindex           = nlist->jindex;
 130     jjnr             = nlist->jjnr;
 131     shiftidx         = nlist->shift;
 132     gid              = nlist->gid;
 133     shiftvec         = fr->shift_vec[0];
 134     fshift           = fr->fshift[0];
 135     facel            = _mm_set1_ps(fr->epsfac);
 136     charge           = mdatoms->chargeA;
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140
 141     vftab            = kernel_data->table_vdw->data;
 142     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 143
 144     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 145     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 146     beta2            = _mm_mul_ps(beta,beta);
 147     beta3            = _mm_mul_ps(beta,beta2);
 148     ewtab            = fr->ic->tabq_coul_FDV0;
 149     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 150     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 151
 152     /* Setup water-specific parameters */
 153     inr              = nlist->iinr[0];
 154     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 155     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 156     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 157     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159     jq0              = _mm_set1_ps(charge[inr+0]);
 160     jq1              = _mm_set1_ps(charge[inr+1]);
 161     jq2              = _mm_set1_ps(charge[inr+2]);
 162     vdwjidx0A        = 2*vdwtype[inr+0];
 163     qq00             = _mm_mul_ps(iq0,jq0);
 164     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 165     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 166     qq01             = _mm_mul_ps(iq0,jq1);
 167     qq02             = _mm_mul_ps(iq0,jq2);
 168     qq10             = _mm_mul_ps(iq1,jq0);
 169     qq11             = _mm_mul_ps(iq1,jq1);
 170     qq12             = _mm_mul_ps(iq1,jq2);
 171     qq20             = _mm_mul_ps(iq2,jq0);
 172     qq21             = _mm_mul_ps(iq2,jq1);
 173     qq22             = _mm_mul_ps(iq2,jq2);
 174
 175     /* Avoid stupid compiler warnings */
 176     jnrA = jnrB = jnrC = jnrD = 0;
 177     j_coord_offsetA = 0;
 178     j_coord_offsetB = 0;
 179     j_coord_offsetC = 0;
 180     j_coord_offsetD = 0;
 181
 182     outeriter        = 0;
 183     inneriter        = 0;
 184
 185     for(iidx=0;iidx<4*DIM;iidx++)
 186     {
 187         scratch[iidx] = 0.0;
 188     }
 189
 190     /* Start outer loop over neighborlists */
 191     for(iidx=0; iidx<nri; iidx++)
 192     {
 193         /* Load shift vector for this list */
 194         i_shift_offset   = DIM*shiftidx[iidx];
 195
 196         /* Load limits for loop over neighbors */
 197         j_index_start    = jindex[iidx];
 198         j_index_end      = jindex[iidx+1];
 199
 200         /* Get outer coordinate index */
 201         inr              = iinr[iidx];
 202         i_coord_offset   = DIM*inr;
 203
 204         /* Load i particle coords and add shift vector */
 205         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 206                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 207
 208         fix0             = _mm_setzero_ps();
 209         fiy0             = _mm_setzero_ps();
 210         fiz0             = _mm_setzero_ps();
 211         fix1             = _mm_setzero_ps();
 212         fiy1             = _mm_setzero_ps();
 213         fiz1             = _mm_setzero_ps();
 214         fix2             = _mm_setzero_ps();
 215         fiy2             = _mm_setzero_ps();
 216         fiz2             = _mm_setzero_ps();
 217
 218         /* Reset potential sums */
 219         velecsum         = _mm_setzero_ps();
 220         vvdwsum          = _mm_setzero_ps();
 221
 222         /* Start inner kernel loop */
 223         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 224         {
 225
 226             /* Get j neighbor index, and coordinate index */
 227             jnrA             = jjnr[jidx];
 228             jnrB             = jjnr[jidx+1];
 229             jnrC             = jjnr[jidx+2];
 230             jnrD             = jjnr[jidx+3];
 231             j_coord_offsetA  = DIM*jnrA;
 232             j_coord_offsetB  = DIM*jnrB;
 233             j_coord_offsetC  = DIM*jnrC;
 234             j_coord_offsetD  = DIM*jnrD;
 235
 236             /* load j atom coordinates */
 237             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 238                                               x+j_coord_offsetC,x+j_coord_offsetD,
 239                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 240
 241             /* Calculate displacement vector */
 242             dx00             = _mm_sub_ps(ix0,jx0);
 243             dy00             = _mm_sub_ps(iy0,jy0);
 244             dz00             = _mm_sub_ps(iz0,jz0);
 245             dx01             = _mm_sub_ps(ix0,jx1);
 246             dy01             = _mm_sub_ps(iy0,jy1);
 247             dz01             = _mm_sub_ps(iz0,jz1);
 248             dx02             = _mm_sub_ps(ix0,jx2);
 249             dy02             = _mm_sub_ps(iy0,jy2);
 250             dz02             = _mm_sub_ps(iz0,jz2);
 251             dx10             = _mm_sub_ps(ix1,jx0);
 252             dy10             = _mm_sub_ps(iy1,jy0);
 253             dz10             = _mm_sub_ps(iz1,jz0);
 254             dx11             = _mm_sub_ps(ix1,jx1);
 255             dy11             = _mm_sub_ps(iy1,jy1);
 256             dz11             = _mm_sub_ps(iz1,jz1);
 257             dx12             = _mm_sub_ps(ix1,jx2);
 258             dy12             = _mm_sub_ps(iy1,jy2);
 259             dz12             = _mm_sub_ps(iz1,jz2);
 260             dx20             = _mm_sub_ps(ix2,jx0);
 261             dy20             = _mm_sub_ps(iy2,jy0);
 262             dz20             = _mm_sub_ps(iz2,jz0);
 263             dx21             = _mm_sub_ps(ix2,jx1);
 264             dy21             = _mm_sub_ps(iy2,jy1);
 265             dz21             = _mm_sub_ps(iz2,jz1);
 266             dx22             = _mm_sub_ps(ix2,jx2);
 267             dy22             = _mm_sub_ps(iy2,jy2);
 268             dz22             = _mm_sub_ps(iz2,jz2);
 269
 270             /* Calculate squared distance and things based on it */
 271             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 272             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 273             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 274             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 275             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 276             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 277             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 278             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 279             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 280
 281             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 282             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 283             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 284             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 285             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 286             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 287             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 288             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 289             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 290
 291             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 292             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 293             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 294             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 295             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 296             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 297             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 298             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 299             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 300
 301             fjx0             = _mm_setzero_ps();
 302             fjy0             = _mm_setzero_ps();
 303             fjz0             = _mm_setzero_ps();
 304             fjx1             = _mm_setzero_ps();
 305             fjy1             = _mm_setzero_ps();
 306             fjz1             = _mm_setzero_ps();
 307             fjx2             = _mm_setzero_ps();
 308             fjy2             = _mm_setzero_ps();
 309             fjz2             = _mm_setzero_ps();
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             r00              = _mm_mul_ps(rsq00,rinv00);
 316
 317             /* Calculate table index by multiplying r with table scale and truncate to integer */
 318             rt               = _mm_mul_ps(r00,vftabscale);
 319             vfitab           = _mm_cvttps_epi32(rt);
 320 #ifdef __XOP__
 321             vfeps            = _mm_frcz_ps(rt);
 322 #else
 323             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 324 #endif
 325             twovfeps         = _mm_add_ps(vfeps,vfeps);
 326             vfitab           = _mm_slli_epi32(vfitab,3);
 327
 328             /* EWALD ELECTROSTATICS */
 329
 330             /* Analytical PME correction */
 331             zeta2            = _mm_mul_ps(beta2,rsq00);
 332             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 333             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 334             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 335             felec            = _mm_mul_ps(qq00,felec);
 336             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 337             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 338             velec            = _mm_mul_ps(qq00,velec);
 339
 340             /* CUBIC SPLINE TABLE DISPERSION */
 341             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 342             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 343             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 344             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 345             _MM_TRANSPOSE4_PS(Y,F,G,H);
 346             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 347             VV               = _mm_macc_ps(vfeps,Fp,Y);
 348             vvdw6            = _mm_mul_ps(c6_00,VV);
 349             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 350             fvdw6            = _mm_mul_ps(c6_00,FF);
 351
 352             /* CUBIC SPLINE TABLE REPULSION */
 353             vfitab           = _mm_add_epi32(vfitab,ifour);
 354             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 355             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 356             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 357             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 358             _MM_TRANSPOSE4_PS(Y,F,G,H);
 359             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 360             VV               = _mm_macc_ps(vfeps,Fp,Y);
 361             vvdw12           = _mm_mul_ps(c12_00,VV);
 362             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 363             fvdw12           = _mm_mul_ps(c12_00,FF);
 364             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 365             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velecsum         = _mm_add_ps(velecsum,velec);
 369             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 370
 371             fscal            = _mm_add_ps(felec,fvdw);
 372
 373              /* Update vectorial force */
 374             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 375             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 376             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 377
 378             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 379             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 380             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 381
 382             /**************************
 383              * CALCULATE INTERACTIONS *
 384              **************************/
 385
 386             r01              = _mm_mul_ps(rsq01,rinv01);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Analytical PME correction */
 391             zeta2            = _mm_mul_ps(beta2,rsq01);
 392             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
 393             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 394             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 395             felec            = _mm_mul_ps(qq01,felec);
 396             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 397             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv01);
 398             velec            = _mm_mul_ps(qq01,velec);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velecsum         = _mm_add_ps(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405              /* Update vectorial force */
 406             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 407             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 408             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 409
 410             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 411             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 412             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 413
 414             /**************************
 415              * CALCULATE INTERACTIONS *
 416              **************************/
 417
 418             r02              = _mm_mul_ps(rsq02,rinv02);
 419
 420             /* EWALD ELECTROSTATICS */
 421
 422             /* Analytical PME correction */
 423             zeta2            = _mm_mul_ps(beta2,rsq02);
 424             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
 425             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 426             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 427             felec            = _mm_mul_ps(qq02,felec);
 428             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 429             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv02);
 430             velec            = _mm_mul_ps(qq02,velec);
 431
 432             /* Update potential sum for this i atom from the interaction with this j atom. */
 433             velecsum         = _mm_add_ps(velecsum,velec);
 434
 435             fscal            = felec;
 436
 437              /* Update vectorial force */
 438             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 439             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 440             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 441
 442             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 443             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 444             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 445
 446             /**************************
 447              * CALCULATE INTERACTIONS *
 448              **************************/
 449
 450             r10              = _mm_mul_ps(rsq10,rinv10);
 451
 452             /* EWALD ELECTROSTATICS */
 453
 454             /* Analytical PME correction */
 455             zeta2            = _mm_mul_ps(beta2,rsq10);
 456             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 457             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 458             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 459             felec            = _mm_mul_ps(qq10,felec);
 460             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 461             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 462             velec            = _mm_mul_ps(qq10,velec);
 463
 464             /* Update potential sum for this i atom from the interaction with this j atom. */
 465             velecsum         = _mm_add_ps(velecsum,velec);
 466
 467             fscal            = felec;
 468
 469              /* Update vectorial force */
 470             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 471             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 472             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 473
 474             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 475             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 476             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             r11              = _mm_mul_ps(rsq11,rinv11);
 483
 484             /* EWALD ELECTROSTATICS */
 485
 486             /* Analytical PME correction */
 487             zeta2            = _mm_mul_ps(beta2,rsq11);
 488             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 489             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 490             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 491             felec            = _mm_mul_ps(qq11,felec);
 492             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 493             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 494             velec            = _mm_mul_ps(qq11,velec);
 495
 496             /* Update potential sum for this i atom from the interaction with this j atom. */
 497             velecsum         = _mm_add_ps(velecsum,velec);
 498
 499             fscal            = felec;
 500
 501              /* Update vectorial force */
 502             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 503             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 504             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 505
 506             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 507             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 508             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             r12              = _mm_mul_ps(rsq12,rinv12);
 515
 516             /* EWALD ELECTROSTATICS */
 517
 518             /* Analytical PME correction */
 519             zeta2            = _mm_mul_ps(beta2,rsq12);
 520             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 521             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 522             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 523             felec            = _mm_mul_ps(qq12,felec);
 524             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 525             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 526             velec            = _mm_mul_ps(qq12,velec);
 527
 528             /* Update potential sum for this i atom from the interaction with this j atom. */
 529             velecsum         = _mm_add_ps(velecsum,velec);
 530
 531             fscal            = felec;
 532
 533              /* Update vectorial force */
 534             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 535             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 536             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 537
 538             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 539             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 540             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             r20              = _mm_mul_ps(rsq20,rinv20);
 547
 548             /* EWALD ELECTROSTATICS */
 549
 550             /* Analytical PME correction */
 551             zeta2            = _mm_mul_ps(beta2,rsq20);
 552             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 553             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 554             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 555             felec            = _mm_mul_ps(qq20,felec);
 556             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 557             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 558             velec            = _mm_mul_ps(qq20,velec);
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velecsum         = _mm_add_ps(velecsum,velec);
 562
 563             fscal            = felec;
 564
 565              /* Update vectorial force */
 566             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 567             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 568             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 569
 570             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 571             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 572             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 573
 574             /**************************
 575              * CALCULATE INTERACTIONS *
 576              **************************/
 577
 578             r21              = _mm_mul_ps(rsq21,rinv21);
 579
 580             /* EWALD ELECTROSTATICS */
 581
 582             /* Analytical PME correction */
 583             zeta2            = _mm_mul_ps(beta2,rsq21);
 584             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
 585             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 586             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 587             felec            = _mm_mul_ps(qq21,felec);
 588             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 589             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
 590             velec            = _mm_mul_ps(qq21,velec);
 591
 592             /* Update potential sum for this i atom from the interaction with this j atom. */
 593             velecsum         = _mm_add_ps(velecsum,velec);
 594
 595             fscal            = felec;
 596
 597              /* Update vectorial force */
 598             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 599             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 600             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 601
 602             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 603             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 604             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             r22              = _mm_mul_ps(rsq22,rinv22);
 611
 612             /* EWALD ELECTROSTATICS */
 613
 614             /* Analytical PME correction */
 615             zeta2            = _mm_mul_ps(beta2,rsq22);
 616             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
 617             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 618             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 619             felec            = _mm_mul_ps(qq22,felec);
 620             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 621             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
 622             velec            = _mm_mul_ps(qq22,velec);
 623
 624             /* Update potential sum for this i atom from the interaction with this j atom. */
 625             velecsum         = _mm_add_ps(velecsum,velec);
 626
 627             fscal            = felec;
 628
 629              /* Update vectorial force */
 630             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 631             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 632             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 633
 634             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 635             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 636             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 637
 638             fjptrA             = f+j_coord_offsetA;
 639             fjptrB             = f+j_coord_offsetB;
 640             fjptrC             = f+j_coord_offsetC;
 641             fjptrD             = f+j_coord_offsetD;
 642
 643             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 644                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 645
 646             /* Inner loop uses 295 flops */
 647         }
 648
 649         if(jidx<j_index_end)
 650         {
 651
 652             /* Get j neighbor index, and coordinate index */
 653             jnrlistA         = jjnr[jidx];
 654             jnrlistB         = jjnr[jidx+1];
 655             jnrlistC         = jjnr[jidx+2];
 656             jnrlistD         = jjnr[jidx+3];
 657             /* Sign of each element will be negative for non-real atoms.
 658              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 659              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 660              */
 661             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 662             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 663             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 664             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 665             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 666             j_coord_offsetA  = DIM*jnrA;
 667             j_coord_offsetB  = DIM*jnrB;
 668             j_coord_offsetC  = DIM*jnrC;
 669             j_coord_offsetD  = DIM*jnrD;
 670
 671             /* load j atom coordinates */
 672             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 673                                               x+j_coord_offsetC,x+j_coord_offsetD,
 674                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 675
 676             /* Calculate displacement vector */
 677             dx00             = _mm_sub_ps(ix0,jx0);
 678             dy00             = _mm_sub_ps(iy0,jy0);
 679             dz00             = _mm_sub_ps(iz0,jz0);
 680             dx01             = _mm_sub_ps(ix0,jx1);
 681             dy01             = _mm_sub_ps(iy0,jy1);
 682             dz01             = _mm_sub_ps(iz0,jz1);
 683             dx02             = _mm_sub_ps(ix0,jx2);
 684             dy02             = _mm_sub_ps(iy0,jy2);
 685             dz02             = _mm_sub_ps(iz0,jz2);
 686             dx10             = _mm_sub_ps(ix1,jx0);
 687             dy10             = _mm_sub_ps(iy1,jy0);
 688             dz10             = _mm_sub_ps(iz1,jz0);
 689             dx11             = _mm_sub_ps(ix1,jx1);
 690             dy11             = _mm_sub_ps(iy1,jy1);
 691             dz11             = _mm_sub_ps(iz1,jz1);
 692             dx12             = _mm_sub_ps(ix1,jx2);
 693             dy12             = _mm_sub_ps(iy1,jy2);
 694             dz12             = _mm_sub_ps(iz1,jz2);
 695             dx20             = _mm_sub_ps(ix2,jx0);
 696             dy20             = _mm_sub_ps(iy2,jy0);
 697             dz20             = _mm_sub_ps(iz2,jz0);
 698             dx21             = _mm_sub_ps(ix2,jx1);
 699             dy21             = _mm_sub_ps(iy2,jy1);
 700             dz21             = _mm_sub_ps(iz2,jz1);
 701             dx22             = _mm_sub_ps(ix2,jx2);
 702             dy22             = _mm_sub_ps(iy2,jy2);
 703             dz22             = _mm_sub_ps(iz2,jz2);
 704
 705             /* Calculate squared distance and things based on it */
 706             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 707             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 708             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 709             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 710             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 711             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 712             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 713             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 714             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 715
 716             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 717             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 718             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 719             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 720             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 721             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 722             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 723             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 724             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 725
 726             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 727             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 728             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 729             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 730             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 731             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 732             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 733             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 734             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 735
 736             fjx0             = _mm_setzero_ps();
 737             fjy0             = _mm_setzero_ps();
 738             fjz0             = _mm_setzero_ps();
 739             fjx1             = _mm_setzero_ps();
 740             fjy1             = _mm_setzero_ps();
 741             fjz1             = _mm_setzero_ps();
 742             fjx2             = _mm_setzero_ps();
 743             fjy2             = _mm_setzero_ps();
 744             fjz2             = _mm_setzero_ps();
 745
 746             /**************************
 747              * CALCULATE INTERACTIONS *
 748              **************************/
 749
 750             r00              = _mm_mul_ps(rsq00,rinv00);
 751             r00              = _mm_andnot_ps(dummy_mask,r00);
 752
 753             /* Calculate table index by multiplying r with table scale and truncate to integer */
 754             rt               = _mm_mul_ps(r00,vftabscale);
 755             vfitab           = _mm_cvttps_epi32(rt);
 756 #ifdef __XOP__
 757             vfeps            = _mm_frcz_ps(rt);
 758 #else
 759             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 760 #endif
 761             twovfeps         = _mm_add_ps(vfeps,vfeps);
 762             vfitab           = _mm_slli_epi32(vfitab,3);
 763
 764             /* EWALD ELECTROSTATICS */
 765
 766             /* Analytical PME correction */
 767             zeta2            = _mm_mul_ps(beta2,rsq00);
 768             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 769             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 770             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 771             felec            = _mm_mul_ps(qq00,felec);
 772             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 773             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 774             velec            = _mm_mul_ps(qq00,velec);
 775
 776             /* CUBIC SPLINE TABLE DISPERSION */
 777             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 778             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 779             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 780             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 781             _MM_TRANSPOSE4_PS(Y,F,G,H);
 782             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 783             VV               = _mm_macc_ps(vfeps,Fp,Y);
 784             vvdw6            = _mm_mul_ps(c6_00,VV);
 785             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 786             fvdw6            = _mm_mul_ps(c6_00,FF);
 787
 788             /* CUBIC SPLINE TABLE REPULSION */
 789             vfitab           = _mm_add_epi32(vfitab,ifour);
 790             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 791             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 792             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 793             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 794             _MM_TRANSPOSE4_PS(Y,F,G,H);
 795             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 796             VV               = _mm_macc_ps(vfeps,Fp,Y);
 797             vvdw12           = _mm_mul_ps(c12_00,VV);
 798             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 799             fvdw12           = _mm_mul_ps(c12_00,FF);
 800             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 801             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 802
 803             /* Update potential sum for this i atom from the interaction with this j atom. */
 804             velec            = _mm_andnot_ps(dummy_mask,velec);
 805             velecsum         = _mm_add_ps(velecsum,velec);
 806             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 807             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 808
 809             fscal            = _mm_add_ps(felec,fvdw);
 810
 811             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 812
 813              /* Update vectorial force */
 814             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 815             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 816             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 817
 818             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 819             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 820             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             r01              = _mm_mul_ps(rsq01,rinv01);
 827             r01              = _mm_andnot_ps(dummy_mask,r01);
 828
 829             /* EWALD ELECTROSTATICS */
 830
 831             /* Analytical PME correction */
 832             zeta2            = _mm_mul_ps(beta2,rsq01);
 833             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
 834             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 835             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 836             felec            = _mm_mul_ps(qq01,felec);
 837             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 838             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv01);
 839             velec            = _mm_mul_ps(qq01,velec);
 840
 841             /* Update potential sum for this i atom from the interaction with this j atom. */
 842             velec            = _mm_andnot_ps(dummy_mask,velec);
 843             velecsum         = _mm_add_ps(velecsum,velec);
 844
 845             fscal            = felec;
 846
 847             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 848
 849              /* Update vectorial force */
 850             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 851             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 852             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 853
 854             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 855             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 856             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             r02              = _mm_mul_ps(rsq02,rinv02);
 863             r02              = _mm_andnot_ps(dummy_mask,r02);
 864
 865             /* EWALD ELECTROSTATICS */
 866
 867             /* Analytical PME correction */
 868             zeta2            = _mm_mul_ps(beta2,rsq02);
 869             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
 870             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 871             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 872             felec            = _mm_mul_ps(qq02,felec);
 873             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 874             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv02);
 875             velec            = _mm_mul_ps(qq02,velec);
 876
 877             /* Update potential sum for this i atom from the interaction with this j atom. */
 878             velec            = _mm_andnot_ps(dummy_mask,velec);
 879             velecsum         = _mm_add_ps(velecsum,velec);
 880
 881             fscal            = felec;
 882
 883             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 884
 885              /* Update vectorial force */
 886             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 887             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 888             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 889
 890             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 891             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 892             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             r10              = _mm_mul_ps(rsq10,rinv10);
 899             r10              = _mm_andnot_ps(dummy_mask,r10);
 900
 901             /* EWALD ELECTROSTATICS */
 902
 903             /* Analytical PME correction */
 904             zeta2            = _mm_mul_ps(beta2,rsq10);
 905             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 906             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 907             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 908             felec            = _mm_mul_ps(qq10,felec);
 909             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 910             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 911             velec            = _mm_mul_ps(qq10,velec);
 912
 913             /* Update potential sum for this i atom from the interaction with this j atom. */
 914             velec            = _mm_andnot_ps(dummy_mask,velec);
 915             velecsum         = _mm_add_ps(velecsum,velec);
 916
 917             fscal            = felec;
 918
 919             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 920
 921              /* Update vectorial force */
 922             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 923             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 924             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 925
 926             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 927             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 928             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 929
 930             /**************************
 931              * CALCULATE INTERACTIONS *
 932              **************************/
 933
 934             r11              = _mm_mul_ps(rsq11,rinv11);
 935             r11              = _mm_andnot_ps(dummy_mask,r11);
 936
 937             /* EWALD ELECTROSTATICS */
 938
 939             /* Analytical PME correction */
 940             zeta2            = _mm_mul_ps(beta2,rsq11);
 941             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 942             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 943             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 944             felec            = _mm_mul_ps(qq11,felec);
 945             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 946             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv11);
 947             velec            = _mm_mul_ps(qq11,velec);
 948
 949             /* Update potential sum for this i atom from the interaction with this j atom. */
 950             velec            = _mm_andnot_ps(dummy_mask,velec);
 951             velecsum         = _mm_add_ps(velecsum,velec);
 952
 953             fscal            = felec;
 954
 955             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 956
 957              /* Update vectorial force */
 958             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 959             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 960             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 961
 962             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 963             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 964             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             r12              = _mm_mul_ps(rsq12,rinv12);
 971             r12              = _mm_andnot_ps(dummy_mask,r12);
 972
 973             /* EWALD ELECTROSTATICS */
 974
 975             /* Analytical PME correction */
 976             zeta2            = _mm_mul_ps(beta2,rsq12);
 977             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 978             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 979             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 980             felec            = _mm_mul_ps(qq12,felec);
 981             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 982             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv12);
 983             velec            = _mm_mul_ps(qq12,velec);
 984
 985             /* Update potential sum for this i atom from the interaction with this j atom. */
 986             velec            = _mm_andnot_ps(dummy_mask,velec);
 987             velecsum         = _mm_add_ps(velecsum,velec);
 988
 989             fscal            = felec;
 990
 991             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 992
 993              /* Update vectorial force */
 994             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 995             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 996             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 997
 998             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 999             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1000             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1001
1002             /**************************
1003              * CALCULATE INTERACTIONS *
1004              **************************/
1005
1006             r20              = _mm_mul_ps(rsq20,rinv20);
1007             r20              = _mm_andnot_ps(dummy_mask,r20);
1008
1009             /* EWALD ELECTROSTATICS */
1010
1011             /* Analytical PME correction */
1012             zeta2            = _mm_mul_ps(beta2,rsq20);
1013             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1014             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1015             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1016             felec            = _mm_mul_ps(qq20,felec);
1017             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1018             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
1019             velec            = _mm_mul_ps(qq20,velec);
1020
1021             /* Update potential sum for this i atom from the interaction with this j atom. */
1022             velec            = _mm_andnot_ps(dummy_mask,velec);
1023             velecsum         = _mm_add_ps(velecsum,velec);
1024
1025             fscal            = felec;
1026
1027             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1028
1029              /* Update vectorial force */
1030             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1031             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1032             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1033
1034             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1035             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1036             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             r21              = _mm_mul_ps(rsq21,rinv21);
1043             r21              = _mm_andnot_ps(dummy_mask,r21);
1044
1045             /* EWALD ELECTROSTATICS */
1046
1047             /* Analytical PME correction */
1048             zeta2            = _mm_mul_ps(beta2,rsq21);
1049             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1050             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1051             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1052             felec            = _mm_mul_ps(qq21,felec);
1053             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1054             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv21);
1055             velec            = _mm_mul_ps(qq21,velec);
1056
1057             /* Update potential sum for this i atom from the interaction with this j atom. */
1058             velec            = _mm_andnot_ps(dummy_mask,velec);
1059             velecsum         = _mm_add_ps(velecsum,velec);
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1064
1065              /* Update vectorial force */
1066             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1067             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1068             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1069
1070             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1071             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1072             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             r22              = _mm_mul_ps(rsq22,rinv22);
1079             r22              = _mm_andnot_ps(dummy_mask,r22);
1080
1081             /* EWALD ELECTROSTATICS */
1082
1083             /* Analytical PME correction */
1084             zeta2            = _mm_mul_ps(beta2,rsq22);
1085             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1086             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1087             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1088             felec            = _mm_mul_ps(qq22,felec);
1089             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1090             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv22);
1091             velec            = _mm_mul_ps(qq22,velec);
1092
1093             /* Update potential sum for this i atom from the interaction with this j atom. */
1094             velec            = _mm_andnot_ps(dummy_mask,velec);
1095             velecsum         = _mm_add_ps(velecsum,velec);
1096
1097             fscal            = felec;
1098
1099             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1100
1101              /* Update vectorial force */
1102             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1103             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1104             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1105
1106             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1107             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1108             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1109
1110             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1111             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1112             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1113             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1114
1115             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1116                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1117
1118             /* Inner loop uses 304 flops */
1119         }
1120
1121         /* End of innermost loop */
1122
1123         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1124                                               f+i_coord_offset,fshift+i_shift_offset);
1125
1126         ggid                        = gid[iidx];
1127         /* Update potential energies */
1128         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1129         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1130
1131         /* Increment number of inner iterations */
1132         inneriter                  += j_index_end - j_index_start;
1133
1134         /* Outer loop uses 20 flops */
1135     }
1136
1137     /* Increment number of outer iterations */
1138     outeriter        += nri;
1139
1140     /* Update outer/inner flops */
1141
1142     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*304);
1143 }
1144 /*
1145  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_avx_128_fma_single
1146  * Electrostatics interaction: Ewald
1147  * VdW interaction:            CubicSplineTable
1148  * Geometry:                   Water3-Water3
1149  * Calculate force/pot:        Force
1150  */
1151 void
1152 nb_kernel_ElecEw_VdwCSTab_GeomW3W3_F_avx_128_fma_single
1153                     (t_nblist                    * gmx_restrict       nlist,
1154                      rvec                        * gmx_restrict          xx,
1155                      rvec                        * gmx_restrict          ff,
1156                      t_forcerec                  * gmx_restrict          fr,
1157                      t_mdatoms                   * gmx_restrict     mdatoms,
1158                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1159                      t_nrnb                      * gmx_restrict        nrnb)
1160 {
1161     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1162      * just 0 for non-waters.
1163      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1164      * jnr indices corresponding to data put in the four positions in the SIMD register.
1165      */
1166     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1167     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1168     int              jnrA,jnrB,jnrC,jnrD;
1169     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1170     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1171     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1172     real             rcutoff_scalar;
1173     real             *shiftvec,*fshift,*x,*f;
1174     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1175     real             scratch[4*DIM];
1176     __m128           fscal,rcutoff,rcutoff2,jidxall;
1177     int              vdwioffset0;
1178     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1179     int              vdwioffset1;
1180     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1181     int              vdwioffset2;
1182     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1183     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1184     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1185     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1186     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1187     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1188     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1189     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1190     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1191     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1192     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1193     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1194     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1195     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1196     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1197     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1198     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1199     real             *charge;
1200     int              nvdwtype;
1201     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1202     int              *vdwtype;
1203     real             *vdwparam;
1204     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1205     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1206     __m128i          vfitab;
1207     __m128i          ifour       = _mm_set1_epi32(4);
1208     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
1209     real             *vftab;
1210     __m128i          ewitab;
1211     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1212     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1213     real             *ewtab;
1214     __m128           dummy_mask,cutoff_mask;
1215     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1216     __m128           one     = _mm_set1_ps(1.0);
1217     __m128           two     = _mm_set1_ps(2.0);
1218     x                = xx[0];
1219     f                = ff[0];
1220
1221     nri              = nlist->nri;
1222     iinr             = nlist->iinr;
1223     jindex           = nlist->jindex;
1224     jjnr             = nlist->jjnr;
1225     shiftidx         = nlist->shift;
1226     gid              = nlist->gid;
1227     shiftvec         = fr->shift_vec[0];
1228     fshift           = fr->fshift[0];
1229     facel            = _mm_set1_ps(fr->epsfac);
1230     charge           = mdatoms->chargeA;
1231     nvdwtype         = fr->ntype;
1232     vdwparam         = fr->nbfp;
1233     vdwtype          = mdatoms->typeA;
1234
1235     vftab            = kernel_data->table_vdw->data;
1236     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
1237
1238     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1239     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
1240     beta2            = _mm_mul_ps(beta,beta);
1241     beta3            = _mm_mul_ps(beta,beta2);
1242     ewtab            = fr->ic->tabq_coul_F;
1243     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1244     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1245
1246     /* Setup water-specific parameters */
1247     inr              = nlist->iinr[0];
1248     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1249     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1250     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1251     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1252
1253     jq0              = _mm_set1_ps(charge[inr+0]);
1254     jq1              = _mm_set1_ps(charge[inr+1]);
1255     jq2              = _mm_set1_ps(charge[inr+2]);
1256     vdwjidx0A        = 2*vdwtype[inr+0];
1257     qq00             = _mm_mul_ps(iq0,jq0);
1258     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1259     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1260     qq01             = _mm_mul_ps(iq0,jq1);
1261     qq02             = _mm_mul_ps(iq0,jq2);
1262     qq10             = _mm_mul_ps(iq1,jq0);
1263     qq11             = _mm_mul_ps(iq1,jq1);
1264     qq12             = _mm_mul_ps(iq1,jq2);
1265     qq20             = _mm_mul_ps(iq2,jq0);
1266     qq21             = _mm_mul_ps(iq2,jq1);
1267     qq22             = _mm_mul_ps(iq2,jq2);
1268
1269     /* Avoid stupid compiler warnings */
1270     jnrA = jnrB = jnrC = jnrD = 0;
1271     j_coord_offsetA = 0;
1272     j_coord_offsetB = 0;
1273     j_coord_offsetC = 0;
1274     j_coord_offsetD = 0;
1275
1276     outeriter        = 0;
1277     inneriter        = 0;
1278
1279     for(iidx=0;iidx<4*DIM;iidx++)
1280     {
1281         scratch[iidx] = 0.0;
1282     }
1283
1284     /* Start outer loop over neighborlists */
1285     for(iidx=0; iidx<nri; iidx++)
1286     {
1287         /* Load shift vector for this list */
1288         i_shift_offset   = DIM*shiftidx[iidx];
1289
1290         /* Load limits for loop over neighbors */
1291         j_index_start    = jindex[iidx];
1292         j_index_end      = jindex[iidx+1];
1293
1294         /* Get outer coordinate index */
1295         inr              = iinr[iidx];
1296         i_coord_offset   = DIM*inr;
1297
1298         /* Load i particle coords and add shift vector */
1299         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1300                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1301
1302         fix0             = _mm_setzero_ps();
1303         fiy0             = _mm_setzero_ps();
1304         fiz0             = _mm_setzero_ps();
1305         fix1             = _mm_setzero_ps();
1306         fiy1             = _mm_setzero_ps();
1307         fiz1             = _mm_setzero_ps();
1308         fix2             = _mm_setzero_ps();
1309         fiy2             = _mm_setzero_ps();
1310         fiz2             = _mm_setzero_ps();
1311
1312         /* Start inner kernel loop */
1313         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1314         {
1315
1316             /* Get j neighbor index, and coordinate index */
1317             jnrA             = jjnr[jidx];
1318             jnrB             = jjnr[jidx+1];
1319             jnrC             = jjnr[jidx+2];
1320             jnrD             = jjnr[jidx+3];
1321             j_coord_offsetA  = DIM*jnrA;
1322             j_coord_offsetB  = DIM*jnrB;
1323             j_coord_offsetC  = DIM*jnrC;
1324             j_coord_offsetD  = DIM*jnrD;
1325
1326             /* load j atom coordinates */
1327             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1328                                               x+j_coord_offsetC,x+j_coord_offsetD,
1329                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1330
1331             /* Calculate displacement vector */
1332             dx00             = _mm_sub_ps(ix0,jx0);
1333             dy00             = _mm_sub_ps(iy0,jy0);
1334             dz00             = _mm_sub_ps(iz0,jz0);
1335             dx01             = _mm_sub_ps(ix0,jx1);
1336             dy01             = _mm_sub_ps(iy0,jy1);
1337             dz01             = _mm_sub_ps(iz0,jz1);
1338             dx02             = _mm_sub_ps(ix0,jx2);
1339             dy02             = _mm_sub_ps(iy0,jy2);
1340             dz02             = _mm_sub_ps(iz0,jz2);
1341             dx10             = _mm_sub_ps(ix1,jx0);
1342             dy10             = _mm_sub_ps(iy1,jy0);
1343             dz10             = _mm_sub_ps(iz1,jz0);
1344             dx11             = _mm_sub_ps(ix1,jx1);
1345             dy11             = _mm_sub_ps(iy1,jy1);
1346             dz11             = _mm_sub_ps(iz1,jz1);
1347             dx12             = _mm_sub_ps(ix1,jx2);
1348             dy12             = _mm_sub_ps(iy1,jy2);
1349             dz12             = _mm_sub_ps(iz1,jz2);
1350             dx20             = _mm_sub_ps(ix2,jx0);
1351             dy20             = _mm_sub_ps(iy2,jy0);
1352             dz20             = _mm_sub_ps(iz2,jz0);
1353             dx21             = _mm_sub_ps(ix2,jx1);
1354             dy21             = _mm_sub_ps(iy2,jy1);
1355             dz21             = _mm_sub_ps(iz2,jz1);
1356             dx22             = _mm_sub_ps(ix2,jx2);
1357             dy22             = _mm_sub_ps(iy2,jy2);
1358             dz22             = _mm_sub_ps(iz2,jz2);
1359
1360             /* Calculate squared distance and things based on it */
1361             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1362             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1363             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1364             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1365             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1366             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1367             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1368             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1369             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1370
1371             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1372             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1373             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1374             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1375             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1376             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1377             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1378             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1379             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1380
1381             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1382             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1383             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1384             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1385             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1386             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1387             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1388             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1389             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1390
1391             fjx0             = _mm_setzero_ps();
1392             fjy0             = _mm_setzero_ps();
1393             fjz0             = _mm_setzero_ps();
1394             fjx1             = _mm_setzero_ps();
1395             fjy1             = _mm_setzero_ps();
1396             fjz1             = _mm_setzero_ps();
1397             fjx2             = _mm_setzero_ps();
1398             fjy2             = _mm_setzero_ps();
1399             fjz2             = _mm_setzero_ps();
1400
1401             /**************************
1402              * CALCULATE INTERACTIONS *
1403              **************************/
1404
1405             r00              = _mm_mul_ps(rsq00,rinv00);
1406
1407             /* Calculate table index by multiplying r with table scale and truncate to integer */
1408             rt               = _mm_mul_ps(r00,vftabscale);
1409             vfitab           = _mm_cvttps_epi32(rt);
1410 #ifdef __XOP__
1411             vfeps            = _mm_frcz_ps(rt);
1412 #else
1413             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1414 #endif
1415             twovfeps         = _mm_add_ps(vfeps,vfeps);
1416             vfitab           = _mm_slli_epi32(vfitab,3);
1417
1418             /* EWALD ELECTROSTATICS */
1419
1420             /* Analytical PME correction */
1421             zeta2            = _mm_mul_ps(beta2,rsq00);
1422             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1423             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1424             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1425             felec            = _mm_mul_ps(qq00,felec);
1426
1427             /* CUBIC SPLINE TABLE DISPERSION */
1428             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1429             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1430             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1431             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1432             _MM_TRANSPOSE4_PS(Y,F,G,H);
1433             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1434             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1435             fvdw6            = _mm_mul_ps(c6_00,FF);
1436
1437             /* CUBIC SPLINE TABLE REPULSION */
1438             vfitab           = _mm_add_epi32(vfitab,ifour);
1439             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1440             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1441             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1442             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1443             _MM_TRANSPOSE4_PS(Y,F,G,H);
1444             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1445             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1446             fvdw12           = _mm_mul_ps(c12_00,FF);
1447             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1448
1449             fscal            = _mm_add_ps(felec,fvdw);
1450
1451              /* Update vectorial force */
1452             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1453             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1454             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1455
1456             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1457             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1458             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1459
1460             /**************************
1461              * CALCULATE INTERACTIONS *
1462              **************************/
1463
1464             r01              = _mm_mul_ps(rsq01,rinv01);
1465
1466             /* EWALD ELECTROSTATICS */
1467
1468             /* Analytical PME correction */
1469             zeta2            = _mm_mul_ps(beta2,rsq01);
1470             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
1471             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1472             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1473             felec            = _mm_mul_ps(qq01,felec);
1474
1475             fscal            = felec;
1476
1477              /* Update vectorial force */
1478             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1479             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1480             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1481
1482             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1483             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1484             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1485
1486             /**************************
1487              * CALCULATE INTERACTIONS *
1488              **************************/
1489
1490             r02              = _mm_mul_ps(rsq02,rinv02);
1491
1492             /* EWALD ELECTROSTATICS */
1493
1494             /* Analytical PME correction */
1495             zeta2            = _mm_mul_ps(beta2,rsq02);
1496             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
1497             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1498             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1499             felec            = _mm_mul_ps(qq02,felec);
1500
1501             fscal            = felec;
1502
1503              /* Update vectorial force */
1504             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1505             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1506             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1507
1508             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1509             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1510             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1511
1512             /**************************
1513              * CALCULATE INTERACTIONS *
1514              **************************/
1515
1516             r10              = _mm_mul_ps(rsq10,rinv10);
1517
1518             /* EWALD ELECTROSTATICS */
1519
1520             /* Analytical PME correction */
1521             zeta2            = _mm_mul_ps(beta2,rsq10);
1522             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1523             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1524             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1525             felec            = _mm_mul_ps(qq10,felec);
1526
1527             fscal            = felec;
1528
1529              /* Update vectorial force */
1530             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1531             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1532             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1533
1534             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1535             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1536             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1537
1538             /**************************
1539              * CALCULATE INTERACTIONS *
1540              **************************/
1541
1542             r11              = _mm_mul_ps(rsq11,rinv11);
1543
1544             /* EWALD ELECTROSTATICS */
1545
1546             /* Analytical PME correction */
1547             zeta2            = _mm_mul_ps(beta2,rsq11);
1548             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1549             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1550             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1551             felec            = _mm_mul_ps(qq11,felec);
1552
1553             fscal            = felec;
1554
1555              /* Update vectorial force */
1556             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1557             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1558             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1559
1560             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1561             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1562             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1563
1564             /**************************
1565              * CALCULATE INTERACTIONS *
1566              **************************/
1567
1568             r12              = _mm_mul_ps(rsq12,rinv12);
1569
1570             /* EWALD ELECTROSTATICS */
1571
1572             /* Analytical PME correction */
1573             zeta2            = _mm_mul_ps(beta2,rsq12);
1574             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1575             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1576             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1577             felec            = _mm_mul_ps(qq12,felec);
1578
1579             fscal            = felec;
1580
1581              /* Update vectorial force */
1582             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1583             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1584             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1585
1586             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1587             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1588             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1589
1590             /**************************
1591              * CALCULATE INTERACTIONS *
1592              **************************/
1593
1594             r20              = _mm_mul_ps(rsq20,rinv20);
1595
1596             /* EWALD ELECTROSTATICS */
1597
1598             /* Analytical PME correction */
1599             zeta2            = _mm_mul_ps(beta2,rsq20);
1600             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1601             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1602             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1603             felec            = _mm_mul_ps(qq20,felec);
1604
1605             fscal            = felec;
1606
1607              /* Update vectorial force */
1608             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1609             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1610             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1611
1612             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1613             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1614             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1615
1616             /**************************
1617              * CALCULATE INTERACTIONS *
1618              **************************/
1619
1620             r21              = _mm_mul_ps(rsq21,rinv21);
1621
1622             /* EWALD ELECTROSTATICS */
1623
1624             /* Analytical PME correction */
1625             zeta2            = _mm_mul_ps(beta2,rsq21);
1626             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1627             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1628             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1629             felec            = _mm_mul_ps(qq21,felec);
1630
1631             fscal            = felec;
1632
1633              /* Update vectorial force */
1634             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1635             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1636             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1637
1638             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1639             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1640             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1641
1642             /**************************
1643              * CALCULATE INTERACTIONS *
1644              **************************/
1645
1646             r22              = _mm_mul_ps(rsq22,rinv22);
1647
1648             /* EWALD ELECTROSTATICS */
1649
1650             /* Analytical PME correction */
1651             zeta2            = _mm_mul_ps(beta2,rsq22);
1652             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1653             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1654             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1655             felec            = _mm_mul_ps(qq22,felec);
1656
1657             fscal            = felec;
1658
1659              /* Update vectorial force */
1660             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1661             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1662             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1663
1664             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1665             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1666             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1667
1668             fjptrA             = f+j_coord_offsetA;
1669             fjptrB             = f+j_coord_offsetB;
1670             fjptrC             = f+j_coord_offsetC;
1671             fjptrD             = f+j_coord_offsetD;
1672
1673             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1674                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1675
1676             /* Inner loop uses 278 flops */
1677         }
1678
1679         if(jidx<j_index_end)
1680         {
1681
1682             /* Get j neighbor index, and coordinate index */
1683             jnrlistA         = jjnr[jidx];
1684             jnrlistB         = jjnr[jidx+1];
1685             jnrlistC         = jjnr[jidx+2];
1686             jnrlistD         = jjnr[jidx+3];
1687             /* Sign of each element will be negative for non-real atoms.
1688              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1689              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1690              */
1691             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1692             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1693             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1694             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1695             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1696             j_coord_offsetA  = DIM*jnrA;
1697             j_coord_offsetB  = DIM*jnrB;
1698             j_coord_offsetC  = DIM*jnrC;
1699             j_coord_offsetD  = DIM*jnrD;
1700
1701             /* load j atom coordinates */
1702             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1703                                               x+j_coord_offsetC,x+j_coord_offsetD,
1704                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1705
1706             /* Calculate displacement vector */
1707             dx00             = _mm_sub_ps(ix0,jx0);
1708             dy00             = _mm_sub_ps(iy0,jy0);
1709             dz00             = _mm_sub_ps(iz0,jz0);
1710             dx01             = _mm_sub_ps(ix0,jx1);
1711             dy01             = _mm_sub_ps(iy0,jy1);
1712             dz01             = _mm_sub_ps(iz0,jz1);
1713             dx02             = _mm_sub_ps(ix0,jx2);
1714             dy02             = _mm_sub_ps(iy0,jy2);
1715             dz02             = _mm_sub_ps(iz0,jz2);
1716             dx10             = _mm_sub_ps(ix1,jx0);
1717             dy10             = _mm_sub_ps(iy1,jy0);
1718             dz10             = _mm_sub_ps(iz1,jz0);
1719             dx11             = _mm_sub_ps(ix1,jx1);
1720             dy11             = _mm_sub_ps(iy1,jy1);
1721             dz11             = _mm_sub_ps(iz1,jz1);
1722             dx12             = _mm_sub_ps(ix1,jx2);
1723             dy12             = _mm_sub_ps(iy1,jy2);
1724             dz12             = _mm_sub_ps(iz1,jz2);
1725             dx20             = _mm_sub_ps(ix2,jx0);
1726             dy20             = _mm_sub_ps(iy2,jy0);
1727             dz20             = _mm_sub_ps(iz2,jz0);
1728             dx21             = _mm_sub_ps(ix2,jx1);
1729             dy21             = _mm_sub_ps(iy2,jy1);
1730             dz21             = _mm_sub_ps(iz2,jz1);
1731             dx22             = _mm_sub_ps(ix2,jx2);
1732             dy22             = _mm_sub_ps(iy2,jy2);
1733             dz22             = _mm_sub_ps(iz2,jz2);
1734
1735             /* Calculate squared distance and things based on it */
1736             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1737             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1738             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1739             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1740             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1741             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1742             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1743             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1744             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1745
1746             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1747             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1748             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1749             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1750             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1751             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1752             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1753             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1754             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1755
1756             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1757             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1758             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1759             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1760             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1761             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1762             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1763             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1764             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1765
1766             fjx0             = _mm_setzero_ps();
1767             fjy0             = _mm_setzero_ps();
1768             fjz0             = _mm_setzero_ps();
1769             fjx1             = _mm_setzero_ps();
1770             fjy1             = _mm_setzero_ps();
1771             fjz1             = _mm_setzero_ps();
1772             fjx2             = _mm_setzero_ps();
1773             fjy2             = _mm_setzero_ps();
1774             fjz2             = _mm_setzero_ps();
1775
1776             /**************************
1777              * CALCULATE INTERACTIONS *
1778              **************************/
1779
1780             r00              = _mm_mul_ps(rsq00,rinv00);
1781             r00              = _mm_andnot_ps(dummy_mask,r00);
1782
1783             /* Calculate table index by multiplying r with table scale and truncate to integer */
1784             rt               = _mm_mul_ps(r00,vftabscale);
1785             vfitab           = _mm_cvttps_epi32(rt);
1786 #ifdef __XOP__
1787             vfeps            = _mm_frcz_ps(rt);
1788 #else
1789             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1790 #endif
1791             twovfeps         = _mm_add_ps(vfeps,vfeps);
1792             vfitab           = _mm_slli_epi32(vfitab,3);
1793
1794             /* EWALD ELECTROSTATICS */
1795
1796             /* Analytical PME correction */
1797             zeta2            = _mm_mul_ps(beta2,rsq00);
1798             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1799             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1800             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1801             felec            = _mm_mul_ps(qq00,felec);
1802
1803             /* CUBIC SPLINE TABLE DISPERSION */
1804             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1805             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1806             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1807             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1808             _MM_TRANSPOSE4_PS(Y,F,G,H);
1809             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1810             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1811             fvdw6            = _mm_mul_ps(c6_00,FF);
1812
1813             /* CUBIC SPLINE TABLE REPULSION */
1814             vfitab           = _mm_add_epi32(vfitab,ifour);
1815             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1816             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1817             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1818             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1819             _MM_TRANSPOSE4_PS(Y,F,G,H);
1820             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1821             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1822             fvdw12           = _mm_mul_ps(c12_00,FF);
1823             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1824
1825             fscal            = _mm_add_ps(felec,fvdw);
1826
1827             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1828
1829              /* Update vectorial force */
1830             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1831             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1832             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1833
1834             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1835             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1836             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1837
1838             /**************************
1839              * CALCULATE INTERACTIONS *
1840              **************************/
1841
1842             r01              = _mm_mul_ps(rsq01,rinv01);
1843             r01              = _mm_andnot_ps(dummy_mask,r01);
1844
1845             /* EWALD ELECTROSTATICS */
1846
1847             /* Analytical PME correction */
1848             zeta2            = _mm_mul_ps(beta2,rsq01);
1849             rinv3            = _mm_mul_ps(rinvsq01,rinv01);
1850             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1851             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1852             felec            = _mm_mul_ps(qq01,felec);
1853
1854             fscal            = felec;
1855
1856             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1857
1858              /* Update vectorial force */
1859             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1860             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1861             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1862
1863             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1864             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1865             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1866
1867             /**************************
1868              * CALCULATE INTERACTIONS *
1869              **************************/
1870
1871             r02              = _mm_mul_ps(rsq02,rinv02);
1872             r02              = _mm_andnot_ps(dummy_mask,r02);
1873
1874             /* EWALD ELECTROSTATICS */
1875
1876             /* Analytical PME correction */
1877             zeta2            = _mm_mul_ps(beta2,rsq02);
1878             rinv3            = _mm_mul_ps(rinvsq02,rinv02);
1879             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1880             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1881             felec            = _mm_mul_ps(qq02,felec);
1882
1883             fscal            = felec;
1884
1885             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1886
1887              /* Update vectorial force */
1888             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1889             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1890             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1891
1892             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1893             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1894             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1895
1896             /**************************
1897              * CALCULATE INTERACTIONS *
1898              **************************/
1899
1900             r10              = _mm_mul_ps(rsq10,rinv10);
1901             r10              = _mm_andnot_ps(dummy_mask,r10);
1902
1903             /* EWALD ELECTROSTATICS */
1904
1905             /* Analytical PME correction */
1906             zeta2            = _mm_mul_ps(beta2,rsq10);
1907             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1908             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1909             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1910             felec            = _mm_mul_ps(qq10,felec);
1911
1912             fscal            = felec;
1913
1914             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1915
1916              /* Update vectorial force */
1917             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1918             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1919             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1920
1921             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1922             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1923             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1924
1925             /**************************
1926              * CALCULATE INTERACTIONS *
1927              **************************/
1928
1929             r11              = _mm_mul_ps(rsq11,rinv11);
1930             r11              = _mm_andnot_ps(dummy_mask,r11);
1931
1932             /* EWALD ELECTROSTATICS */
1933
1934             /* Analytical PME correction */
1935             zeta2            = _mm_mul_ps(beta2,rsq11);
1936             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1937             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1938             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1939             felec            = _mm_mul_ps(qq11,felec);
1940
1941             fscal            = felec;
1942
1943             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1944
1945              /* Update vectorial force */
1946             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1947             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1948             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1949
1950             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1951             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1952             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1953
1954             /**************************
1955              * CALCULATE INTERACTIONS *
1956              **************************/
1957
1958             r12              = _mm_mul_ps(rsq12,rinv12);
1959             r12              = _mm_andnot_ps(dummy_mask,r12);
1960
1961             /* EWALD ELECTROSTATICS */
1962
1963             /* Analytical PME correction */
1964             zeta2            = _mm_mul_ps(beta2,rsq12);
1965             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1966             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1967             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1968             felec            = _mm_mul_ps(qq12,felec);
1969
1970             fscal            = felec;
1971
1972             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1973
1974              /* Update vectorial force */
1975             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1976             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1977             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1978
1979             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1980             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1981             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1982
1983             /**************************
1984              * CALCULATE INTERACTIONS *
1985              **************************/
1986
1987             r20              = _mm_mul_ps(rsq20,rinv20);
1988             r20              = _mm_andnot_ps(dummy_mask,r20);
1989
1990             /* EWALD ELECTROSTATICS */
1991
1992             /* Analytical PME correction */
1993             zeta2            = _mm_mul_ps(beta2,rsq20);
1994             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1995             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1996             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1997             felec            = _mm_mul_ps(qq20,felec);
1998
1999             fscal            = felec;
2000
2001             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2002
2003              /* Update vectorial force */
2004             fix2             = _mm_macc_ps(dx20,fscal,fix2);
2005             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
2006             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
2007
2008             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
2009             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
2010             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
2011
2012             /**************************
2013              * CALCULATE INTERACTIONS *
2014              **************************/
2015
2016             r21              = _mm_mul_ps(rsq21,rinv21);
2017             r21              = _mm_andnot_ps(dummy_mask,r21);
2018
2019             /* EWALD ELECTROSTATICS */
2020
2021             /* Analytical PME correction */
2022             zeta2            = _mm_mul_ps(beta2,rsq21);
2023             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
2024             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2025             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2026             felec            = _mm_mul_ps(qq21,felec);
2027
2028             fscal            = felec;
2029
2030             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2031
2032              /* Update vectorial force */
2033             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2034             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2035             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2036
2037             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2038             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2039             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2040
2041             /**************************
2042              * CALCULATE INTERACTIONS *
2043              **************************/
2044
2045             r22              = _mm_mul_ps(rsq22,rinv22);
2046             r22              = _mm_andnot_ps(dummy_mask,r22);
2047
2048             /* EWALD ELECTROSTATICS */
2049
2050             /* Analytical PME correction */
2051             zeta2            = _mm_mul_ps(beta2,rsq22);
2052             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
2053             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2054             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2055             felec            = _mm_mul_ps(qq22,felec);
2056
2057             fscal            = felec;
2058
2059             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2060
2061              /* Update vectorial force */
2062             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2063             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2064             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2065
2066             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2067             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2068             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2069
2070             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2071             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2072             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2073             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2074
2075             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2076                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2077
2078             /* Inner loop uses 287 flops */
2079         }
2080
2081         /* End of innermost loop */
2082
2083         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2084                                               f+i_coord_offset,fshift+i_shift_offset);
2085
2086         /* Increment number of inner iterations */
2087         inneriter                  += j_index_end - j_index_start;
2088
2089         /* Outer loop uses 18 flops */
2090     }
2091
2092     /* Increment number of outer iterations */
2093     outeriter        += nri;
2094
2095     /* Update outer/inner flops */
2096
2097     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*287);
2098 }