2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2012,2013, by the GROMACS development team, led by
6 * David van der Spoel, Berk Hess, Erik Lindahl, and including many
7 * others, as listed in the AUTHORS file in the top-level source
8 * directory and at http://www.gromacs.org.
10 * GROMACS is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public License
12 * as published by the Free Software Foundation; either version 2.1
13 * of the License, or (at your option) any later version.
15 * GROMACS is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with GROMACS; if not, see
22 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
23 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
25 * If you want to redistribute modifications to GROMACS, please
26 * consider that scientific software is very special. Version
27 * control is crucial - bugs must be traceable. We will be happy to
28 * consider code for inclusion in the official distribution, but
29 * derived work must not be called official GROMACS. Details are found
30 * in the README & COPYING files - if they are missing, get the
31 * official version at http://www.gromacs.org.
33 * To help us fund GROMACS development, we humbly ask that you cite
34 * the research papers on the package. Check out http://www.gromacs.org.
41 #include "types/simple.h"
50 #include "genborn_allvsall.h"
53 #if 0 && defined (GMX_X86_SSE2)
55 #include <gmx_sse2_double.h>
73 int ** prologue_mask_gb;
86 gmx_allvsallgb2_data_t;
90 calc_maxoffset(int i,int natoms)
94 if ((natoms % 2) == 1)
96 /* Odd number of atoms, easy */
99 else if ((natoms % 4) == 0)
101 /* Multiple of four is hard */
110 maxoffset=natoms/2-1;
121 maxoffset=natoms/2-1;
134 maxoffset=natoms/2-1;
142 setup_gb_exclusions_and_indices(gmx_allvsallgb2_data_t * aadata,
153 int ni0,ni1,nj0,nj1,nj;
157 int firstinteraction;
161 /* This routine can appear to be a bit complex, but it is mostly book-keeping.
162 * To enable the fast all-vs-all kernel we need to be able to stream through all coordinates
163 * whether they should interact or not.
165 * To avoid looping over the exclusions, we create a simple mask that is 1 if the interaction
166 * should be present, otherwise 0. Since exclusions typically only occur when i & j are close,
167 * we create a jindex array with three elements per i atom: the starting point, the point to
168 * which we need to check exclusions, and the end point.
169 * This way we only have to allocate a short exclusion mask per i atom.
172 ni0 = (start/UNROLLI)*UNROLLI;
173 ni1 = ((end+UNROLLI-1)/UNROLLI)*UNROLLI;
175 /* Set the interaction mask to only enable the i atoms we want to include */
176 snew(pi,2*(natoms+UNROLLI+2*SIMD_WIDTH));
177 aadata->imask = (int *) (((size_t) pi + 16) & (~((size_t) 15)));
178 for(i=0;i<natoms+UNROLLI;i++)
180 aadata->imask[2*i] = (i>=start && i<end) ? 0xFFFFFFFF : 0;
181 aadata->imask[2*i+1] = (i>=start && i<end) ? 0xFFFFFFFF : 0;
184 /* Allocate memory for our modified jindex array */
185 snew(aadata->jindex_gb,4*(natoms+UNROLLI));
186 for(i=0;i<4*(natoms+UNROLLI);i++)
188 aadata->jindex_gb[i] = 0;
191 /* Create the exclusion masks for the prologue part */
192 snew(aadata->prologue_mask_gb,natoms+UNROLLI); /* list of pointers */
194 /* First zero everything to avoid uninitialized data */
195 for(i=0;i<natoms+UNROLLI;i++)
197 aadata->prologue_mask_gb[i] = NULL;
200 /* Calculate the largest exclusion range we need for each UNROLLI-tuplet of i atoms. */
201 for(ibase=ni0;ibase<ni1;ibase+=UNROLLI)
203 max_excl_offset = -1;
205 /* First find maxoffset for the next 4 atoms (or fewer if we are close to end) */
206 imax = ((ibase+UNROLLI) < end) ? (ibase+UNROLLI) : end;
208 /* Which atom is the first we (might) interact with? */
209 imin = natoms; /* Guaranteed to be overwritten by one of 'firstinteraction' */
210 for(i=ibase;i<imax;i++)
212 /* Before exclusions, which atom is the first we (might) interact with? */
213 firstinteraction = i+1;
214 max_offset = calc_maxoffset(i,natoms);
218 for(j=0;j<ilist[F_GB12].nr;j+=3)
220 a1 = ilist[F_GB12].iatoms[j+1];
221 a2 = ilist[F_GB12].iatoms[j+2];
236 if(k==firstinteraction)
244 for(j=0;j<ilist[F_GB13].nr;j+=3)
246 a1 = ilist[F_GB13].iatoms[j+1];
247 a2 = ilist[F_GB13].iatoms[j+2];
262 if(k==firstinteraction)
270 for(j=0;j<ilist[F_GB14].nr;j+=3)
272 a1 = ilist[F_GB14].iatoms[j+1];
273 a2 = ilist[F_GB14].iatoms[j+2];
287 if(k==firstinteraction)
293 imin = (firstinteraction < imin) ? firstinteraction : imin;
295 /* round down to j unrolling factor */
296 imin = (imin/UNROLLJ)*UNROLLJ;
298 for(i=ibase;i<imax;i++)
300 max_offset = calc_maxoffset(i,natoms);
304 for(j=0;j<ilist[F_GB12].nr;j+=3)
306 a1 = ilist[F_GB12].iatoms[j+1];
307 a2 = ilist[F_GB12].iatoms[j+2];
334 if( k+natoms <= max_offset )
338 max_excl_offset = (k > max_excl_offset) ? k : max_excl_offset;
343 for(j=0;j<ilist[F_GB13].nr;j+=3)
345 a1 = ilist[F_GB13].iatoms[j+1];
346 a2 = ilist[F_GB13].iatoms[j+2];
373 if( k+natoms <= max_offset )
377 max_excl_offset = (k > max_excl_offset) ? k : max_excl_offset;
382 for(j=0;j<ilist[F_GB14].nr;j+=3)
384 a1 = ilist[F_GB14].iatoms[j+1];
385 a2 = ilist[F_GB14].iatoms[j+2];
412 if( k+natoms <= max_offset )
416 max_excl_offset = (k > max_excl_offset) ? k : max_excl_offset;
421 /* The offset specifies the last atom to be excluded, so add one unit to get an upper loop limit */
423 /* round up to j unrolling factor */
424 max_excl_offset = (max_excl_offset/UNROLLJ+1)*UNROLLJ;
426 /* Set all the prologue masks length to this value (even for i>end) */
427 for(i=ibase;i<ibase+UNROLLI;i++)
429 aadata->jindex_gb[4*i] = imin;
430 aadata->jindex_gb[4*i+1] = imin+max_excl_offset;
434 /* Now the hard part, loop over it all again to calculate the actual contents of the prologue masks */
435 for(ibase=ni0;ibase<ni1;ibase+=UNROLLI)
437 for(i=ibase;i<ibase+UNROLLI;i++)
439 nj = aadata->jindex_gb[4*i+1] - aadata->jindex_gb[4*i];
440 imin = aadata->jindex_gb[4*i];
442 /* Allocate aligned memory */
443 snew(pi,2*(nj+2*SIMD_WIDTH));
444 aadata->prologue_mask_gb[i] = (int *) (((size_t) pi + 16) & (~((size_t) 15)));
446 max_offset = calc_maxoffset(i,natoms);
448 /* Include interactions i+1 <= j < i+maxoffset */
453 if( (j>i) && (j<=i+max_offset) )
455 aadata->prologue_mask_gb[i][2*k] = 0xFFFFFFFF;
456 aadata->prologue_mask_gb[i][2*k+1] = 0xFFFFFFFF;
460 aadata->prologue_mask_gb[i][2*k] = 0;
461 aadata->prologue_mask_gb[i][2*k+1] = 0;
465 /* Clear out the explicit exclusions */
470 for(j=0;j<ilist[F_GB12].nr;j+=3)
472 a1 = ilist[F_GB12].iatoms[j+1];
473 a2 = ilist[F_GB12].iatoms[j+2];
494 if( k+natoms <= max_offset )
502 aadata->prologue_mask_gb[i][2*k] = 0;
503 aadata->prologue_mask_gb[i][2*k+1] = 0;
509 for(j=0;j<ilist[F_GB13].nr;j+=3)
511 a1 = ilist[F_GB13].iatoms[j+1];
512 a2 = ilist[F_GB13].iatoms[j+2];
533 if( k+natoms <= max_offset )
541 aadata->prologue_mask_gb[i][2*k] = 0;
542 aadata->prologue_mask_gb[i][2*k+1] = 0;
548 for(j=0;j<ilist[F_GB14].nr;j+=3)
550 a1 = ilist[F_GB14].iatoms[j+1];
551 a2 = ilist[F_GB14].iatoms[j+2];
572 if( k+natoms <= max_offset )
580 aadata->prologue_mask_gb[i][2*k] = 0;
581 aadata->prologue_mask_gb[i][2*k+1] = 0;
589 /* Construct the epilogue mask - this just contains the check for maxoffset */
590 snew(aadata->epilogue_mask,natoms+UNROLLI);
592 /* First zero everything to avoid uninitialized data */
593 for(i=0;i<natoms+UNROLLI;i++)
595 aadata->jindex_gb[4*i+2] = aadata->jindex_gb[4*i+1];
596 aadata->jindex_gb[4*i+3] = aadata->jindex_gb[4*i+1];
597 aadata->epilogue_mask[i] = NULL;
600 for(ibase=ni0;ibase<ni1;ibase+=UNROLLI)
602 /* Find the lowest index for which we need to use the epilogue */
604 max_offset = calc_maxoffset(imin,natoms);
606 imin = imin + 1 + max_offset;
608 /* Find largest index for which we need to use the epilogue */
609 imax = ibase + UNROLLI-1;
610 imax = (imax < end) ? imax : end;
612 max_offset = calc_maxoffset(imax,natoms);
613 imax = imax + 1 + max_offset + UNROLLJ - 1;
615 for(i=ibase;i<ibase+UNROLLI;i++)
617 /* Start of epilogue - round down to j tile limit */
618 aadata->jindex_gb[4*i+2] = (imin/UNROLLJ)*UNROLLJ;
619 /* Make sure we dont overlap - for small systems everything is done in the prologue */
620 aadata->jindex_gb[4*i+2] = (aadata->jindex_gb[4*i+1] > aadata->jindex_gb[4*i+2]) ? aadata->jindex_gb[4*i+1] : aadata->jindex_gb[4*i+2];
621 /* Round upwards to j tile limit */
622 aadata->jindex_gb[4*i+3] = (imax/UNROLLJ)*UNROLLJ;
623 /* Make sure we dont have a negative range for the epilogue */
624 aadata->jindex_gb[4*i+3] = (aadata->jindex_gb[4*i+2] > aadata->jindex_gb[4*i+3]) ? aadata->jindex_gb[4*i+2] : aadata->jindex_gb[4*i+3];
628 /* And fill it with data... */
630 for(ibase=ni0;ibase<ni1;ibase+=UNROLLI)
632 for(i=ibase;i<ibase+UNROLLI;i++)
635 nj = aadata->jindex_gb[4*i+3] - aadata->jindex_gb[4*i+2];
637 /* Allocate aligned memory */
638 snew(pi,2*(nj+2*SIMD_WIDTH));
639 aadata->epilogue_mask[i] = (int *) (((size_t) pi + 16) & (~((size_t) 15)));
641 max_offset = calc_maxoffset(i,natoms);
645 j = aadata->jindex_gb[4*i+2] + k;
646 aadata->epilogue_mask[i][2*k] = (j <= i+max_offset) ? 0xFFFFFFFF : 0;
647 aadata->epilogue_mask[i][2*k+1] = (j <= i+max_offset) ? 0xFFFFFFFF : 0;
655 genborn_allvsall_setup(gmx_allvsallgb2_data_t ** p_aadata,
656 gmx_localtop_t * top,
657 gmx_genborn_t * born,
659 double radius_offset,
667 gmx_allvsallgb2_data_t *aadata;
670 natoms = mdatoms->nr;
675 snew(p,2*natoms+2*SIMD_WIDTH);
676 aadata->x_align = (double *) (((size_t) p + 16) & (~((size_t) 15)));
677 snew(p,2*natoms+2*SIMD_WIDTH);
678 aadata->y_align = (double *) (((size_t) p + 16) & (~((size_t) 15)));
679 snew(p,2*natoms+2*SIMD_WIDTH);
680 aadata->z_align = (double *) (((size_t) p + 16) & (~((size_t) 15)));
681 snew(p,2*natoms+2*SIMD_WIDTH);
682 aadata->fx_align = (double *) (((size_t) p + 16) & (~((size_t) 15)));
683 snew(p,2*natoms+2*SIMD_WIDTH);
684 aadata->fy_align = (double *) (((size_t) p + 16) & (~((size_t) 15)));
685 snew(p,2*natoms+2*SIMD_WIDTH);
686 aadata->fz_align = (double *) (((size_t) p + 16) & (~((size_t) 15)));
688 snew(p,2*natoms+UNROLLJ+SIMD_WIDTH);
689 aadata->gb_radius = (double *) (((size_t) p + 16) & (~((size_t) 15)));
691 snew(p,2*natoms+UNROLLJ+SIMD_WIDTH);
692 aadata->workparam = (double *) (((size_t) p + 16) & (~((size_t) 15)));
694 snew(p,2*natoms+UNROLLJ+SIMD_WIDTH);
695 aadata->work = (double *) (((size_t) p + 16) & (~((size_t) 15)));
697 for(i=0;i<mdatoms->nr;i++)
699 aadata->gb_radius[i] = top->atomtypes.gb_radius[mdatoms->typeA[i]] - radius_offset;
700 if(gb_algorithm==egbSTILL)
702 aadata->workparam[i] = born->vsolv[i];
704 else if(gb_algorithm==egbOBC)
706 aadata->workparam[i] = born->param[i];
708 aadata->work[i] = 0.0;
710 for(i=0;i<mdatoms->nr;i++)
712 aadata->gb_radius[natoms+i] = aadata->gb_radius[i];
713 aadata->workparam[natoms+i] = aadata->workparam[i];
714 aadata->work[natoms+i] = aadata->work[i];
717 for(i=0;i<2*natoms+SIMD_WIDTH;i++)
719 aadata->x_align[i] = 0.0;
720 aadata->y_align[i] = 0.0;
721 aadata->z_align[i] = 0.0;
722 aadata->fx_align[i] = 0.0;
723 aadata->fy_align[i] = 0.0;
724 aadata->fz_align[i] = 0.0;
727 setup_gb_exclusions_and_indices(aadata,top->idef.il,mdatoms->start,mdatoms->start+mdatoms->homenr,mdatoms->nr,
728 bInclude12,bInclude13,bInclude14);
733 * This routine apparently hits a compiler bug visual studio has had 'forever'.
734 * It is present both in VS2005 and VS2008, and the only way around it is to
735 * decrease optimization. We do that with at pragma, and only for MSVC, so it
736 * will not hurt any of the well-behaving and supported compilers out there.
737 * MS: Fix your compiler, it sucks like a black hole!
740 #pragma optimize("t",off)
744 genborn_allvsall_calc_still_radii_sse2_double(t_forcerec * fr,
746 gmx_genborn_t * born,
747 gmx_localtop_t * top,
752 gmx_allvsallgb2_data_t *aadata;
768 double irsq,idr4,idr6;
769 double raj,rvdw,ratio;
770 double vaj,ccf,dccf,theta,cosq;
771 double term,prod,icf4,icf6,gpi2,factor,sinq;
782 __m128d ix_SSE0,iy_SSE0,iz_SSE0;
783 __m128d ix_SSE1,iy_SSE1,iz_SSE1;
784 __m128d gpi_SSE0,rai_SSE0,prod_ai_SSE0;
785 __m128d gpi_SSE1,rai_SSE1,prod_ai_SSE1;
786 __m128d imask_SSE0,jmask_SSE0;
787 __m128d imask_SSE1,jmask_SSE1;
788 __m128d jx_SSE,jy_SSE,jz_SSE;
789 __m128d dx_SSE0,dy_SSE0,dz_SSE0;
790 __m128d dx_SSE1,dy_SSE1,dz_SSE1;
791 __m128d rsq_SSE0,rinv_SSE0,irsq_SSE0,idr4_SSE0,idr6_SSE0;
792 __m128d rsq_SSE1,rinv_SSE1,irsq_SSE1,idr4_SSE1,idr6_SSE1;
793 __m128d raj_SSE,vaj_SSE,prod_SSE;
794 __m128d rvdw_SSE0,ratio_SSE0;
795 __m128d rvdw_SSE1,ratio_SSE1;
796 __m128d theta_SSE0,sinq_SSE0,cosq_SSE0,term_SSE0;
797 __m128d theta_SSE1,sinq_SSE1,cosq_SSE1,term_SSE1;
798 __m128d ccf_SSE0,dccf_SSE0;
799 __m128d ccf_SSE1,dccf_SSE1;
800 __m128d icf4_SSE0,icf6_SSE0;
801 __m128d icf4_SSE1,icf6_SSE1;
802 __m128d half_SSE,one_SSE,two_SSE,four_SSE;
803 __m128d still_p4_SSE,still_p5inv_SSE,still_pip5_SSE;
805 natoms = mdatoms->nr;
806 ni0 = (mdatoms->start/SIMD_WIDTH)*SIMD_WIDTH;
807 ni1 = mdatoms->start+mdatoms->homenr;
811 aadata = *((gmx_allvsallgb2_data_t **)paadata);
816 genborn_allvsall_setup(&aadata,top,born,mdatoms,0.0,
817 egbSTILL,FALSE,FALSE,TRUE);
818 *((gmx_allvsallgb2_data_t **)paadata) = aadata;
821 x_align = aadata->x_align;
822 y_align = aadata->y_align;
823 z_align = aadata->z_align;
825 gb_radius = aadata->gb_radius;
826 vsolv = aadata->workparam;
828 jindex = aadata->jindex_gb;
831 still_p4_SSE = _mm_set1_pd(STILL_P4);
832 still_p5inv_SSE = _mm_set1_pd(STILL_P5INV);
833 still_pip5_SSE = _mm_set1_pd(STILL_PIP5);
834 half_SSE = _mm_set1_pd(0.5);
835 one_SSE = _mm_set1_pd(1.0);
836 two_SSE = _mm_set1_pd(2.0);
837 four_SSE = _mm_set1_pd(4.0);
839 /* This will be summed, so it has to extend to natoms + buffer */
840 for(i=0;i<natoms+1+natoms/2;i++)
845 for(i=ni0;i<ni1+1+natoms/2;i++)
849 y_align[i] = x[3*k+1];
850 z_align[i] = x[3*k+2];
854 for(i=ni0; i<ni1; i+=UNROLLI)
856 /* We assume shifts are NOT used for all-vs-all interactions */
857 /* Load i atom data */
858 ix_SSE0 = _mm_load1_pd(x_align+i);
859 iy_SSE0 = _mm_load1_pd(y_align+i);
860 iz_SSE0 = _mm_load1_pd(z_align+i);
861 ix_SSE1 = _mm_load1_pd(x_align+i+1);
862 iy_SSE1 = _mm_load1_pd(y_align+i+1);
863 iz_SSE1 = _mm_load1_pd(z_align+i+1);
865 gpi_SSE0 = _mm_setzero_pd();
866 gpi_SSE1 = _mm_setzero_pd();
868 rai_SSE0 = _mm_load1_pd(gb_radius+i);
869 rai_SSE1 = _mm_load1_pd(gb_radius+i+1);
871 prod_ai_SSE0 = _mm_set1_pd(STILL_P4*vsolv[i]);
872 prod_ai_SSE1 = _mm_set1_pd(STILL_P4*vsolv[i+1]);
874 /* Load limits for loop over neighbors */
880 pmask0 = aadata->prologue_mask_gb[i];
881 pmask1 = aadata->prologue_mask_gb[i+1];
882 emask0 = aadata->epilogue_mask[i];
883 emask1 = aadata->epilogue_mask[i+1];
885 imask_SSE0 = _mm_load1_pd((double *)(aadata->imask+2*i));
886 imask_SSE1 = _mm_load1_pd((double *)(aadata->imask+2*i+2));
888 /* Prologue part, including exclusion mask */
889 for(j=nj0; j<nj1; j+=UNROLLJ)
891 jmask_SSE0 = _mm_load_pd((double *)pmask0);
892 jmask_SSE1 = _mm_load_pd((double *)pmask1);
896 /* load j atom coordinates */
897 jx_SSE = _mm_load_pd(x_align+j);
898 jy_SSE = _mm_load_pd(y_align+j);
899 jz_SSE = _mm_load_pd(z_align+j);
901 /* Calculate distance */
902 dx_SSE0 = _mm_sub_pd(ix_SSE0,jx_SSE);
903 dy_SSE0 = _mm_sub_pd(iy_SSE0,jy_SSE);
904 dz_SSE0 = _mm_sub_pd(iz_SSE0,jz_SSE);
905 dx_SSE1 = _mm_sub_pd(ix_SSE1,jx_SSE);
906 dy_SSE1 = _mm_sub_pd(iy_SSE1,jy_SSE);
907 dz_SSE1 = _mm_sub_pd(iz_SSE1,jz_SSE);
909 /* rsq = dx*dx+dy*dy+dz*dz */
910 rsq_SSE0 = gmx_mm_calc_rsq_pd(dx_SSE0,dy_SSE0,dz_SSE0);
911 rsq_SSE1 = gmx_mm_calc_rsq_pd(dx_SSE1,dy_SSE1,dz_SSE1);
914 jmask_SSE0 = _mm_and_pd(jmask_SSE0,imask_SSE0);
915 jmask_SSE1 = _mm_and_pd(jmask_SSE1,imask_SSE1);
917 /* Calculate 1/r and 1/r2 */
918 rinv_SSE0 = gmx_mm_invsqrt_pd(rsq_SSE0);
919 rinv_SSE1 = gmx_mm_invsqrt_pd(rsq_SSE1);
922 rinv_SSE0 = _mm_and_pd(rinv_SSE0,jmask_SSE0);
923 rinv_SSE1 = _mm_and_pd(rinv_SSE1,jmask_SSE1);
925 irsq_SSE0 = _mm_mul_pd(rinv_SSE0,rinv_SSE0);
926 irsq_SSE1 = _mm_mul_pd(rinv_SSE1,rinv_SSE1);
927 idr4_SSE0 = _mm_mul_pd(irsq_SSE0,irsq_SSE0);
928 idr4_SSE1 = _mm_mul_pd(irsq_SSE1,irsq_SSE1);
929 idr6_SSE0 = _mm_mul_pd(idr4_SSE0,irsq_SSE0);
930 idr6_SSE1 = _mm_mul_pd(idr4_SSE1,irsq_SSE1);
932 raj_SSE = _mm_load_pd(gb_radius+j);
933 vaj_SSE = _mm_load_pd(vsolv+j);
935 rvdw_SSE0 = _mm_add_pd(rai_SSE0,raj_SSE);
936 rvdw_SSE1 = _mm_add_pd(rai_SSE1,raj_SSE);
938 ratio_SSE0 = _mm_mul_pd(rsq_SSE0, gmx_mm_inv_pd( _mm_mul_pd(rvdw_SSE0,rvdw_SSE0)));
939 ratio_SSE1 = _mm_mul_pd(rsq_SSE1, gmx_mm_inv_pd( _mm_mul_pd(rvdw_SSE1,rvdw_SSE1)));
941 ratio_SSE0 = _mm_min_pd(ratio_SSE0,still_p5inv_SSE);
942 ratio_SSE1 = _mm_min_pd(ratio_SSE1,still_p5inv_SSE);
943 theta_SSE0 = _mm_mul_pd(ratio_SSE0,still_pip5_SSE);
944 theta_SSE1 = _mm_mul_pd(ratio_SSE1,still_pip5_SSE);
945 gmx_mm_sincos_pd(theta_SSE0,&sinq_SSE0,&cosq_SSE0);
946 gmx_mm_sincos_pd(theta_SSE1,&sinq_SSE1,&cosq_SSE1);
947 term_SSE0 = _mm_mul_pd(half_SSE,_mm_sub_pd(one_SSE,cosq_SSE0));
948 term_SSE1 = _mm_mul_pd(half_SSE,_mm_sub_pd(one_SSE,cosq_SSE1));
949 ccf_SSE0 = _mm_mul_pd(term_SSE0,term_SSE0);
950 ccf_SSE1 = _mm_mul_pd(term_SSE1,term_SSE1);
951 dccf_SSE0 = _mm_mul_pd(_mm_mul_pd(two_SSE,term_SSE0),
952 _mm_mul_pd(sinq_SSE0,theta_SSE0));
953 dccf_SSE1 = _mm_mul_pd(_mm_mul_pd(two_SSE,term_SSE1),
954 _mm_mul_pd(sinq_SSE1,theta_SSE1));
956 prod_SSE = _mm_mul_pd(still_p4_SSE,vaj_SSE);
957 icf4_SSE0 = _mm_mul_pd(ccf_SSE0,idr4_SSE0);
958 icf4_SSE1 = _mm_mul_pd(ccf_SSE1,idr4_SSE1);
959 icf6_SSE0 = _mm_mul_pd( _mm_sub_pd( _mm_mul_pd(four_SSE,ccf_SSE0),dccf_SSE0), idr6_SSE0);
960 icf6_SSE1 = _mm_mul_pd( _mm_sub_pd( _mm_mul_pd(four_SSE,ccf_SSE1),dccf_SSE1), idr6_SSE1);
962 _mm_store_pd(work+j , _mm_add_pd(_mm_load_pd(work+j),
963 _mm_add_pd(_mm_mul_pd(prod_ai_SSE0,icf4_SSE0),
964 _mm_mul_pd(prod_ai_SSE1,icf4_SSE1))));
967 gpi_SSE0 = _mm_add_pd(gpi_SSE0, _mm_mul_pd(prod_SSE,icf4_SSE0));
968 gpi_SSE1 = _mm_add_pd(gpi_SSE1, _mm_mul_pd(prod_SSE,icf4_SSE1));
970 /* Save ai->aj and aj->ai chain rule terms */
971 _mm_store_pd(dadx,_mm_mul_pd(prod_SSE,icf6_SSE0));
973 _mm_store_pd(dadx,_mm_mul_pd(prod_SSE,icf6_SSE1));
976 _mm_store_pd(dadx,_mm_mul_pd(prod_ai_SSE0,icf6_SSE0));
978 _mm_store_pd(dadx,_mm_mul_pd(prod_ai_SSE1,icf6_SSE1));
982 /* Main part, no exclusions */
983 for(j=nj1; j<nj2; j+=UNROLLJ)
986 /* load j atom coordinates */
987 jx_SSE = _mm_load_pd(x_align+j);
988 jy_SSE = _mm_load_pd(y_align+j);
989 jz_SSE = _mm_load_pd(z_align+j);
991 /* Calculate distance */
992 dx_SSE0 = _mm_sub_pd(ix_SSE0,jx_SSE);
993 dy_SSE0 = _mm_sub_pd(iy_SSE0,jy_SSE);
994 dz_SSE0 = _mm_sub_pd(iz_SSE0,jz_SSE);
995 dx_SSE1 = _mm_sub_pd(ix_SSE1,jx_SSE);
996 dy_SSE1 = _mm_sub_pd(iy_SSE1,jy_SSE);
997 dz_SSE1 = _mm_sub_pd(iz_SSE1,jz_SSE);
999 /* rsq = dx*dx+dy*dy+dz*dz */
1000 rsq_SSE0 = gmx_mm_calc_rsq_pd(dx_SSE0,dy_SSE0,dz_SSE0);
1001 rsq_SSE1 = gmx_mm_calc_rsq_pd(dx_SSE1,dy_SSE1,dz_SSE1);
1003 /* Calculate 1/r and 1/r2 */
1004 rinv_SSE0 = gmx_mm_invsqrt_pd(rsq_SSE0);
1005 rinv_SSE1 = gmx_mm_invsqrt_pd(rsq_SSE1);
1008 rinv_SSE0 = _mm_and_pd(rinv_SSE0,imask_SSE0);
1009 rinv_SSE1 = _mm_and_pd(rinv_SSE1,imask_SSE1);
1011 irsq_SSE0 = _mm_mul_pd(rinv_SSE0,rinv_SSE0);
1012 irsq_SSE1 = _mm_mul_pd(rinv_SSE1,rinv_SSE1);
1013 idr4_SSE0 = _mm_mul_pd(irsq_SSE0,irsq_SSE0);
1014 idr4_SSE1 = _mm_mul_pd(irsq_SSE1,irsq_SSE1);
1015 idr6_SSE0 = _mm_mul_pd(idr4_SSE0,irsq_SSE0);
1016 idr6_SSE1 = _mm_mul_pd(idr4_SSE1,irsq_SSE1);
1018 raj_SSE = _mm_load_pd(gb_radius+j);
1020 rvdw_SSE0 = _mm_add_pd(rai_SSE0,raj_SSE);
1021 rvdw_SSE1 = _mm_add_pd(rai_SSE1,raj_SSE);
1022 vaj_SSE = _mm_load_pd(vsolv+j);
1024 ratio_SSE0 = _mm_mul_pd(rsq_SSE0, gmx_mm_inv_pd( _mm_mul_pd(rvdw_SSE0,rvdw_SSE0)));
1025 ratio_SSE1 = _mm_mul_pd(rsq_SSE1, gmx_mm_inv_pd( _mm_mul_pd(rvdw_SSE1,rvdw_SSE1)));
1027 ratio_SSE0 = _mm_min_pd(ratio_SSE0,still_p5inv_SSE);
1028 ratio_SSE1 = _mm_min_pd(ratio_SSE1,still_p5inv_SSE);
1029 theta_SSE0 = _mm_mul_pd(ratio_SSE0,still_pip5_SSE);
1030 theta_SSE1 = _mm_mul_pd(ratio_SSE1,still_pip5_SSE);
1031 gmx_mm_sincos_pd(theta_SSE0,&sinq_SSE0,&cosq_SSE0);
1032 gmx_mm_sincos_pd(theta_SSE1,&sinq_SSE1,&cosq_SSE1);
1033 term_SSE0 = _mm_mul_pd(half_SSE,_mm_sub_pd(one_SSE,cosq_SSE0));
1034 term_SSE1 = _mm_mul_pd(half_SSE,_mm_sub_pd(one_SSE,cosq_SSE1));
1035 ccf_SSE0 = _mm_mul_pd(term_SSE0,term_SSE0);
1036 ccf_SSE1 = _mm_mul_pd(term_SSE1,term_SSE1);
1037 dccf_SSE0 = _mm_mul_pd(_mm_mul_pd(two_SSE,term_SSE0),
1038 _mm_mul_pd(sinq_SSE0,theta_SSE0));
1039 dccf_SSE1 = _mm_mul_pd(_mm_mul_pd(two_SSE,term_SSE1),
1040 _mm_mul_pd(sinq_SSE1,theta_SSE1));
1042 prod_SSE = _mm_mul_pd(still_p4_SSE,vaj_SSE );
1043 icf4_SSE0 = _mm_mul_pd(ccf_SSE0,idr4_SSE0);
1044 icf4_SSE1 = _mm_mul_pd(ccf_SSE1,idr4_SSE1);
1045 icf6_SSE0 = _mm_mul_pd( _mm_sub_pd( _mm_mul_pd(four_SSE,ccf_SSE0),dccf_SSE0), idr6_SSE0);
1046 icf6_SSE1 = _mm_mul_pd( _mm_sub_pd( _mm_mul_pd(four_SSE,ccf_SSE1),dccf_SSE1), idr6_SSE1);
1048 _mm_store_pd(work+j , _mm_add_pd(_mm_load_pd(work+j),
1049 _mm_add_pd(_mm_mul_pd(prod_ai_SSE0,icf4_SSE0),
1050 _mm_mul_pd(prod_ai_SSE1,icf4_SSE1))));
1052 gpi_SSE0 = _mm_add_pd(gpi_SSE0, _mm_mul_pd(prod_SSE,icf4_SSE0));
1053 gpi_SSE1 = _mm_add_pd(gpi_SSE1, _mm_mul_pd(prod_SSE,icf4_SSE1));
1055 /* Save ai->aj and aj->ai chain rule terms */
1056 _mm_store_pd(dadx,_mm_mul_pd(prod_SSE,icf6_SSE0));
1058 _mm_store_pd(dadx,_mm_mul_pd(prod_SSE,icf6_SSE1));
1061 _mm_store_pd(dadx,_mm_mul_pd(prod_ai_SSE0,icf6_SSE0));
1063 _mm_store_pd(dadx,_mm_mul_pd(prod_ai_SSE1,icf6_SSE1));
1066 /* Epilogue part, including exclusion mask */
1067 for(j=nj2; j<nj3; j+=UNROLLJ)
1069 jmask_SSE0 = _mm_load_pd((double *)emask0);
1070 jmask_SSE1 = _mm_load_pd((double *)emask1);
1071 emask0 += 2*UNROLLJ;
1072 emask1 += 2*UNROLLJ;
1074 /* load j atom coordinates */
1075 jx_SSE = _mm_load_pd(x_align+j);
1076 jy_SSE = _mm_load_pd(y_align+j);
1077 jz_SSE = _mm_load_pd(z_align+j);
1079 /* Calculate distance */
1080 dx_SSE0 = _mm_sub_pd(ix_SSE0,jx_SSE);
1081 dy_SSE0 = _mm_sub_pd(iy_SSE0,jy_SSE);
1082 dz_SSE0 = _mm_sub_pd(iz_SSE0,jz_SSE);
1083 dx_SSE1 = _mm_sub_pd(ix_SSE1,jx_SSE);
1084 dy_SSE1 = _mm_sub_pd(iy_SSE1,jy_SSE);
1085 dz_SSE1 = _mm_sub_pd(iz_SSE1,jz_SSE);
1087 /* rsq = dx*dx+dy*dy+dz*dz */
1088 rsq_SSE0 = gmx_mm_calc_rsq_pd(dx_SSE0,dy_SSE0,dz_SSE0);
1089 rsq_SSE1 = gmx_mm_calc_rsq_pd(dx_SSE1,dy_SSE1,dz_SSE1);
1092 jmask_SSE0 = _mm_and_pd(jmask_SSE0,imask_SSE0);
1093 jmask_SSE1 = _mm_and_pd(jmask_SSE1,imask_SSE1);
1095 /* Calculate 1/r and 1/r2 */
1096 rinv_SSE0 = gmx_mm_invsqrt_pd(rsq_SSE0);
1097 rinv_SSE1 = gmx_mm_invsqrt_pd(rsq_SSE1);
1100 rinv_SSE0 = _mm_and_pd(rinv_SSE0,jmask_SSE0);
1101 rinv_SSE1 = _mm_and_pd(rinv_SSE1,jmask_SSE1);
1103 irsq_SSE0 = _mm_mul_pd(rinv_SSE0,rinv_SSE0);
1104 irsq_SSE1 = _mm_mul_pd(rinv_SSE1,rinv_SSE1);
1105 idr4_SSE0 = _mm_mul_pd(irsq_SSE0,irsq_SSE0);
1106 idr4_SSE1 = _mm_mul_pd(irsq_SSE1,irsq_SSE1);
1107 idr6_SSE0 = _mm_mul_pd(idr4_SSE0,irsq_SSE0);
1108 idr6_SSE1 = _mm_mul_pd(idr4_SSE1,irsq_SSE1);
1110 raj_SSE = _mm_load_pd(gb_radius+j);
1111 vaj_SSE = _mm_load_pd(vsolv+j);
1113 rvdw_SSE0 = _mm_add_pd(rai_SSE0,raj_SSE);
1114 rvdw_SSE1 = _mm_add_pd(rai_SSE1,raj_SSE);
1116 ratio_SSE0 = _mm_mul_pd(rsq_SSE0, gmx_mm_inv_pd( _mm_mul_pd(rvdw_SSE0,rvdw_SSE0)));
1117 ratio_SSE1 = _mm_mul_pd(rsq_SSE1, gmx_mm_inv_pd( _mm_mul_pd(rvdw_SSE1,rvdw_SSE1)));
1119 ratio_SSE0 = _mm_min_pd(ratio_SSE0,still_p5inv_SSE);
1120 ratio_SSE1 = _mm_min_pd(ratio_SSE1,still_p5inv_SSE);
1121 theta_SSE0 = _mm_mul_pd(ratio_SSE0,still_pip5_SSE);
1122 theta_SSE1 = _mm_mul_pd(ratio_SSE1,still_pip5_SSE);
1123 gmx_mm_sincos_pd(theta_SSE0,&sinq_SSE0,&cosq_SSE0);
1124 gmx_mm_sincos_pd(theta_SSE1,&sinq_SSE1,&cosq_SSE1);
1125 term_SSE0 = _mm_mul_pd(half_SSE,_mm_sub_pd(one_SSE,cosq_SSE0));
1126 term_SSE1 = _mm_mul_pd(half_SSE,_mm_sub_pd(one_SSE,cosq_SSE1));
1127 ccf_SSE0 = _mm_mul_pd(term_SSE0,term_SSE0);
1128 ccf_SSE1 = _mm_mul_pd(term_SSE1,term_SSE1);
1129 dccf_SSE0 = _mm_mul_pd(_mm_mul_pd(two_SSE,term_SSE0),
1130 _mm_mul_pd(sinq_SSE0,theta_SSE0));
1131 dccf_SSE1 = _mm_mul_pd(_mm_mul_pd(two_SSE,term_SSE1),
1132 _mm_mul_pd(sinq_SSE1,theta_SSE1));
1134 prod_SSE = _mm_mul_pd(still_p4_SSE,vaj_SSE);
1135 icf4_SSE0 = _mm_mul_pd(ccf_SSE0,idr4_SSE0);
1136 icf4_SSE1 = _mm_mul_pd(ccf_SSE1,idr4_SSE1);
1137 icf6_SSE0 = _mm_mul_pd( _mm_sub_pd( _mm_mul_pd(four_SSE,ccf_SSE0),dccf_SSE0), idr6_SSE0);
1138 icf6_SSE1 = _mm_mul_pd( _mm_sub_pd( _mm_mul_pd(four_SSE,ccf_SSE1),dccf_SSE1), idr6_SSE1);
1140 _mm_store_pd(work+j , _mm_add_pd(_mm_load_pd(work+j),
1141 _mm_add_pd(_mm_mul_pd(prod_ai_SSE0,icf4_SSE0),
1142 _mm_mul_pd(prod_ai_SSE1,icf4_SSE1))));
1144 gpi_SSE0 = _mm_add_pd(gpi_SSE0, _mm_mul_pd(prod_SSE,icf4_SSE0));
1145 gpi_SSE1 = _mm_add_pd(gpi_SSE1, _mm_mul_pd(prod_SSE,icf4_SSE1));
1147 /* Save ai->aj and aj->ai chain rule terms */
1148 _mm_store_pd(dadx,_mm_mul_pd(prod_SSE,icf6_SSE0));
1150 _mm_store_pd(dadx,_mm_mul_pd(prod_SSE,icf6_SSE1));
1153 _mm_store_pd(dadx,_mm_mul_pd(prod_ai_SSE0,icf6_SSE0));
1155 _mm_store_pd(dadx,_mm_mul_pd(prod_ai_SSE1,icf6_SSE1));
1158 GMX_MM_TRANSPOSE2_PD(gpi_SSE0,gpi_SSE1);
1159 gpi_SSE0 = _mm_add_pd(gpi_SSE0,gpi_SSE1);
1160 _mm_store_pd(work+i, _mm_add_pd(gpi_SSE0, _mm_load_pd(work+i)));
1163 /* In case we have written anything beyond natoms, move it back.
1164 * Never mind that we leave stuff above natoms; that will not
1165 * be accessed later in the routine.
1166 * In principle this should be a move rather than sum, but this
1167 * way we dont have to worry about even/odd offsets...
1169 for(i=natoms;i<ni1+1+natoms/2;i++)
1171 work[i-natoms] += work[i];
1174 /* Parallel summations */
1177 gmx_sum(natoms,work,cr);
1180 factor = 0.5 * ONE_4PI_EPS0;
1181 /* Calculate the radii - should we do all atoms, or just our local ones? */
1182 for(i=0;i<natoms;i++)
1184 if(born->use[i] != 0)
1186 gpi = born->gpol[i]+work[i];
1188 born->bRad[i] = factor*gmx_invsqrt(gpi2);
1189 fr->invsqrta[i] = gmx_invsqrt(born->bRad[i]);
1195 /* Reinstate MSVC optimization */
1197 #pragma optimize("",on)
1202 genborn_allvsall_calc_hct_obc_radii_sse2_double(t_forcerec * fr,
1203 t_mdatoms * mdatoms,
1204 gmx_genborn_t * born,
1206 gmx_localtop_t * top,
1211 gmx_allvsallgb2_data_t *aadata;
1214 int nj0,nj1,nj2,nj3;
1232 double rai,rai_inv,rai_inv2,sum_ai,sum_ai2,sum_ai3,tsum,tchain;
1234 __m128d ix_SSE0,iy_SSE0,iz_SSE0;
1235 __m128d ix_SSE1,iy_SSE1,iz_SSE1;
1236 __m128d gpi_SSE0,rai_SSE0,prod_ai_SSE0;
1237 __m128d gpi_SSE1,rai_SSE1,prod_ai_SSE1;
1238 __m128d imask_SSE0,jmask_SSE0;
1239 __m128d imask_SSE1,jmask_SSE1;
1240 __m128d jx_SSE,jy_SSE,jz_SSE;
1241 __m128d dx_SSE0,dy_SSE0,dz_SSE0;
1242 __m128d dx_SSE1,dy_SSE1,dz_SSE1;
1243 __m128d rsq_SSE0,rinv_SSE0,irsq_SSE0,idr4_SSE0,idr6_SSE0;
1244 __m128d rsq_SSE1,rinv_SSE1,irsq_SSE1,idr4_SSE1,idr6_SSE1;
1245 __m128d raj_SSE,raj_inv_SSE,sk_aj_SSE,sk2_aj_SSE;
1246 __m128d ccf_SSE0,dccf_SSE0,prod_SSE0;
1247 __m128d ccf_SSE1,dccf_SSE1,prod_SSE1;
1248 __m128d icf4_SSE0,icf6_SSE0;
1249 __m128d icf4_SSE1,icf6_SSE1;
1250 __m128d oneeighth_SSE,onefourth_SSE,half_SSE,one_SSE,two_SSE,four_SSE;
1251 __m128d still_p4_SSE,still_p5inv_SSE,still_pip5_SSE;
1252 __m128d rai_inv_SSE0;
1253 __m128d rai_inv_SSE1;
1254 __m128d sk_ai_SSE0,sk2_ai_SSE0,sum_ai_SSE0;
1255 __m128d sk_ai_SSE1,sk2_ai_SSE1,sum_ai_SSE1;
1256 __m128d lij_inv_SSE0,sk2_rinv_SSE0;
1257 __m128d lij_inv_SSE1,sk2_rinv_SSE1;
1260 __m128d t1_SSE0,t2_SSE0,t3_SSE0,t4_SSE0;
1261 __m128d t1_SSE1,t2_SSE1,t3_SSE1,t4_SSE1;
1262 __m128d obc_mask1_SSE0,obc_mask2_SSE0,obc_mask3_SSE0;
1263 __m128d obc_mask1_SSE1,obc_mask2_SSE1,obc_mask3_SSE1;
1264 __m128d uij_SSE0,uij2_SSE0,uij3_SSE0;
1265 __m128d uij_SSE1,uij2_SSE1,uij3_SSE1;
1266 __m128d lij_SSE0,lij2_SSE0,lij3_SSE0;
1267 __m128d lij_SSE1,lij2_SSE1,lij3_SSE1;
1268 __m128d dlij_SSE0,diff2_SSE0,logterm_SSE0;
1269 __m128d dlij_SSE1,diff2_SSE1,logterm_SSE1;
1270 __m128d doffset_SSE,tmpSSE;
1272 natoms = mdatoms->nr;
1273 ni0 = (mdatoms->start/SIMD_WIDTH)*SIMD_WIDTH;
1274 ni1 = mdatoms->start+mdatoms->homenr;
1278 aadata = *((gmx_allvsallgb2_data_t **)paadata);
1283 genborn_allvsall_setup(&aadata,top,born,mdatoms,born->gb_doffset,
1284 egbOBC,TRUE,TRUE,TRUE);
1285 *((gmx_allvsallgb2_data_t **)paadata) = aadata;
1288 x_align = aadata->x_align;
1289 y_align = aadata->y_align;
1290 z_align = aadata->z_align;
1292 gb_radius = aadata->gb_radius;
1293 work = aadata->work;
1294 jindex = aadata->jindex_gb;
1296 obc_param = aadata->workparam;
1298 oneeighth_SSE = _mm_set1_pd(0.125);
1299 onefourth_SSE = _mm_set1_pd(0.25);
1300 half_SSE = _mm_set1_pd(0.5);
1301 one_SSE = _mm_set1_pd(1.0);
1302 two_SSE = _mm_set1_pd(2.0);
1303 four_SSE = _mm_set1_pd(4.0);
1304 doffset_SSE = _mm_set1_pd(born->gb_doffset);
1306 for(i=0;i<natoms;i++)
1308 x_align[i] = x[3*i];
1309 y_align[i] = x[3*i+1];
1310 z_align[i] = x[3*i+2];
1314 for(i=0;i<natoms/2+1;i++)
1316 x_align[natoms+i] = x_align[i];
1317 y_align[natoms+i] = y_align[i];
1318 z_align[natoms+i] = z_align[i];
1321 for(i=0;i<natoms+natoms/2+1;i++)
1326 for(i=ni0; i<ni1; i+= UNROLLI)
1328 /* We assume shifts are NOT used for all-vs-all interactions */
1330 /* Load i atom data */
1331 ix_SSE0 = _mm_load1_pd(x_align+i);
1332 iy_SSE0 = _mm_load1_pd(y_align+i);
1333 iz_SSE0 = _mm_load1_pd(z_align+i);
1334 ix_SSE1 = _mm_load1_pd(x_align+i+1);
1335 iy_SSE1 = _mm_load1_pd(y_align+i+1);
1336 iz_SSE1 = _mm_load1_pd(z_align+i+1);
1338 rai_SSE0 = _mm_load1_pd(gb_radius+i);
1339 rai_SSE1 = _mm_load1_pd(gb_radius+i+1);
1340 rai_inv_SSE0 = gmx_mm_inv_pd(rai_SSE0);
1341 rai_inv_SSE1 = gmx_mm_inv_pd(rai_SSE1);
1343 sk_ai_SSE0 = _mm_load1_pd(obc_param+i);
1344 sk_ai_SSE1 = _mm_load1_pd(obc_param+i+1);
1345 sk2_ai_SSE0 = _mm_mul_pd(sk_ai_SSE0,sk_ai_SSE0);
1346 sk2_ai_SSE1 = _mm_mul_pd(sk_ai_SSE1,sk_ai_SSE1);
1348 sum_ai_SSE0 = _mm_setzero_pd();
1349 sum_ai_SSE1 = _mm_setzero_pd();
1351 /* Load limits for loop over neighbors */
1353 nj1 = jindex[4*i+1];
1354 nj2 = jindex[4*i+2];
1355 nj3 = jindex[4*i+3];
1357 pmask0 = aadata->prologue_mask_gb[i];
1358 pmask1 = aadata->prologue_mask_gb[i+1];
1359 emask0 = aadata->epilogue_mask[i];
1360 emask1 = aadata->epilogue_mask[i+1];
1362 imask_SSE0 = _mm_load1_pd((double *)(aadata->imask+2*i));
1363 imask_SSE1 = _mm_load1_pd((double *)(aadata->imask+2*i+2));
1365 /* Prologue part, including exclusion mask */
1366 for(j=nj0; j<nj1; j+=UNROLLJ)
1368 jmask_SSE0 = _mm_load_pd((double *)pmask0);
1369 jmask_SSE1 = _mm_load_pd((double *)pmask1);
1370 pmask0 += 2*UNROLLJ;
1371 pmask1 += 2*UNROLLJ;
1373 /* load j atom coordinates */
1374 jx_SSE = _mm_load_pd(x_align+j);
1375 jy_SSE = _mm_load_pd(y_align+j);
1376 jz_SSE = _mm_load_pd(z_align+j);
1378 /* Calculate distance */
1379 dx_SSE0 = _mm_sub_pd(ix_SSE0,jx_SSE);
1380 dy_SSE0 = _mm_sub_pd(iy_SSE0,jy_SSE);
1381 dz_SSE0 = _mm_sub_pd(iz_SSE0,jz_SSE);
1382 dx_SSE1 = _mm_sub_pd(ix_SSE1,jx_SSE);
1383 dy_SSE1 = _mm_sub_pd(iy_SSE1,jy_SSE);
1384 dz_SSE1 = _mm_sub_pd(iz_SSE1,jz_SSE);
1386 /* rsq = dx*dx+dy*dy+dz*dz */
1387 rsq_SSE0 = gmx_mm_calc_rsq_pd(dx_SSE0,dy_SSE0,dz_SSE0);
1388 rsq_SSE1 = gmx_mm_calc_rsq_pd(dx_SSE1,dy_SSE1,dz_SSE1);
1391 jmask_SSE0 = _mm_and_pd(jmask_SSE0,imask_SSE0);
1392 jmask_SSE1 = _mm_and_pd(jmask_SSE1,imask_SSE1);
1394 /* Calculate 1/r and 1/r2 */
1395 rinv_SSE0 = gmx_mm_invsqrt_pd(rsq_SSE0);
1396 rinv_SSE1 = gmx_mm_invsqrt_pd(rsq_SSE1);
1399 rinv_SSE0 = _mm_and_pd(rinv_SSE0,jmask_SSE0);
1400 rinv_SSE1 = _mm_and_pd(rinv_SSE1,jmask_SSE1);
1402 dr_SSE0 = _mm_mul_pd(rsq_SSE0,rinv_SSE0);
1403 dr_SSE1 = _mm_mul_pd(rsq_SSE1,rinv_SSE1);
1405 sk_aj_SSE = _mm_load_pd(obc_param+j);
1406 raj_SSE = _mm_load_pd(gb_radius+j);
1407 raj_inv_SSE = gmx_mm_inv_pd(raj_SSE);
1409 /* Evaluate influence of atom aj -> ai */
1410 t1_SSE0 = _mm_add_pd(dr_SSE0,sk_aj_SSE);
1411 t1_SSE1 = _mm_add_pd(dr_SSE1,sk_aj_SSE);
1412 t2_SSE0 = _mm_sub_pd(dr_SSE0,sk_aj_SSE);
1413 t2_SSE1 = _mm_sub_pd(dr_SSE1,sk_aj_SSE);
1414 t3_SSE0 = _mm_sub_pd(sk_aj_SSE,dr_SSE0);
1415 t3_SSE1 = _mm_sub_pd(sk_aj_SSE,dr_SSE1);
1417 obc_mask1_SSE0 = _mm_cmplt_pd(rai_SSE0, t1_SSE0);
1418 obc_mask1_SSE1 = _mm_cmplt_pd(rai_SSE1, t1_SSE1);
1419 obc_mask2_SSE0 = _mm_cmplt_pd(rai_SSE0, t2_SSE0);
1420 obc_mask2_SSE1 = _mm_cmplt_pd(rai_SSE1, t2_SSE1);
1421 obc_mask3_SSE0 = _mm_cmplt_pd(rai_SSE0, t3_SSE0);
1422 obc_mask3_SSE1 = _mm_cmplt_pd(rai_SSE1, t3_SSE1);
1423 obc_mask1_SSE0 = _mm_and_pd(obc_mask1_SSE0,jmask_SSE0);
1424 obc_mask1_SSE1 = _mm_and_pd(obc_mask1_SSE1,jmask_SSE1);
1426 uij_SSE0 = gmx_mm_inv_pd(t1_SSE0);
1427 uij_SSE1 = gmx_mm_inv_pd(t1_SSE1);
1428 lij_SSE0 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE0,gmx_mm_inv_pd(t2_SSE0)),
1429 _mm_andnot_pd(obc_mask2_SSE0,rai_inv_SSE0));
1430 lij_SSE1 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE1,gmx_mm_inv_pd(t2_SSE1)),
1431 _mm_andnot_pd(obc_mask2_SSE1,rai_inv_SSE1));
1432 dlij_SSE0 = _mm_and_pd(one_SSE,obc_mask2_SSE0);
1433 dlij_SSE1 = _mm_and_pd(one_SSE,obc_mask2_SSE1);
1435 uij2_SSE0 = _mm_mul_pd(uij_SSE0, uij_SSE0);
1436 uij2_SSE1 = _mm_mul_pd(uij_SSE1, uij_SSE1);
1437 uij3_SSE0 = _mm_mul_pd(uij2_SSE0,uij_SSE0);
1438 uij3_SSE1 = _mm_mul_pd(uij2_SSE1,uij_SSE1);
1439 lij2_SSE0 = _mm_mul_pd(lij_SSE0, lij_SSE0);
1440 lij2_SSE1 = _mm_mul_pd(lij_SSE1, lij_SSE1);
1441 lij3_SSE0 = _mm_mul_pd(lij2_SSE0,lij_SSE0);
1442 lij3_SSE1 = _mm_mul_pd(lij2_SSE1,lij_SSE1);
1444 diff2_SSE0 = _mm_sub_pd(uij2_SSE0,lij2_SSE0);
1445 diff2_SSE1 = _mm_sub_pd(uij2_SSE1,lij2_SSE1);
1446 lij_inv_SSE0 = gmx_mm_invsqrt_pd(lij2_SSE0);
1447 lij_inv_SSE1 = gmx_mm_invsqrt_pd(lij2_SSE1);
1448 sk2_aj_SSE = _mm_mul_pd(sk_aj_SSE,sk_aj_SSE);
1449 sk2_rinv_SSE0 = _mm_mul_pd(sk2_aj_SSE,rinv_SSE0);
1450 sk2_rinv_SSE1 = _mm_mul_pd(sk2_aj_SSE,rinv_SSE1);
1451 prod_SSE0 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE0);
1452 prod_SSE1 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE1);
1454 logterm_SSE0 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE0,lij_inv_SSE0));
1455 logterm_SSE1 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE1,lij_inv_SSE1));
1457 t1_SSE0 = _mm_sub_pd(lij_SSE0,uij_SSE0);
1458 t1_SSE1 = _mm_sub_pd(lij_SSE1,uij_SSE1);
1459 t2_SSE0 = _mm_mul_pd(diff2_SSE0,
1460 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE0),
1462 t2_SSE1 = _mm_mul_pd(diff2_SSE1,
1463 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE1),
1466 t3_SSE0 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE0,logterm_SSE0));
1467 t3_SSE1 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE1,logterm_SSE1));
1468 t1_SSE0 = _mm_add_pd(t1_SSE0,_mm_add_pd(t2_SSE0,t3_SSE0));
1469 t1_SSE1 = _mm_add_pd(t1_SSE1,_mm_add_pd(t2_SSE1,t3_SSE1));
1470 t4_SSE0 = _mm_mul_pd(two_SSE,_mm_sub_pd(rai_inv_SSE0,lij_SSE0));
1471 t4_SSE1 = _mm_mul_pd(two_SSE,_mm_sub_pd(rai_inv_SSE1,lij_SSE1));
1472 t4_SSE0 = _mm_and_pd(t4_SSE0,obc_mask3_SSE0);
1473 t4_SSE1 = _mm_and_pd(t4_SSE1,obc_mask3_SSE1);
1474 t1_SSE0 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE0,t4_SSE0));
1475 t1_SSE1 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE1,t4_SSE1));
1477 sum_ai_SSE0 = _mm_add_pd(sum_ai_SSE0,_mm_and_pd(t1_SSE0,obc_mask1_SSE0));
1478 sum_ai_SSE1 = _mm_add_pd(sum_ai_SSE1,_mm_and_pd(t1_SSE1,obc_mask1_SSE1));
1480 t1_SSE0 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE0),
1481 _mm_mul_pd(prod_SSE0,lij3_SSE0));
1482 t1_SSE1 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE1),
1483 _mm_mul_pd(prod_SSE1,lij3_SSE1));
1484 t1_SSE0 = _mm_sub_pd(t1_SSE0,
1485 _mm_mul_pd(onefourth_SSE,
1486 _mm_add_pd(_mm_mul_pd(lij_SSE0,rinv_SSE0),
1487 _mm_mul_pd(lij3_SSE0,dr_SSE0))));
1488 t1_SSE1 = _mm_sub_pd(t1_SSE1,
1489 _mm_mul_pd(onefourth_SSE,
1490 _mm_add_pd(_mm_mul_pd(lij_SSE1,rinv_SSE1),
1491 _mm_mul_pd(lij3_SSE1,dr_SSE1))));
1493 t2_SSE0 = _mm_mul_pd(onefourth_SSE,
1494 _mm_add_pd(_mm_mul_pd(uij_SSE0,rinv_SSE0),
1495 _mm_mul_pd(uij3_SSE0,dr_SSE0)));
1496 t2_SSE1 = _mm_mul_pd(onefourth_SSE,
1497 _mm_add_pd(_mm_mul_pd(uij_SSE1,rinv_SSE1),
1498 _mm_mul_pd(uij3_SSE1,dr_SSE1)));
1499 t2_SSE0 = _mm_sub_pd(t2_SSE0,
1500 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE0),
1501 _mm_mul_pd(prod_SSE0,uij3_SSE0)));
1502 t2_SSE1 = _mm_sub_pd(t2_SSE1,
1503 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE1),
1504 _mm_mul_pd(prod_SSE1,uij3_SSE1)));
1505 t3_SSE0 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE0),
1506 _mm_mul_pd(rinv_SSE0,rinv_SSE0));
1507 t3_SSE1 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE1),
1508 _mm_mul_pd(rinv_SSE1,rinv_SSE1));
1509 t3_SSE0 = _mm_sub_pd(t3_SSE0,
1510 _mm_mul_pd(_mm_mul_pd(diff2_SSE0,oneeighth_SSE),
1512 _mm_mul_pd(sk2_rinv_SSE0,rinv_SSE0))));
1513 t3_SSE1 = _mm_sub_pd(t3_SSE1,
1514 _mm_mul_pd(_mm_mul_pd(diff2_SSE1,oneeighth_SSE),
1516 _mm_mul_pd(sk2_rinv_SSE1,rinv_SSE1))));
1518 t1_SSE0 = _mm_mul_pd(rinv_SSE0,
1519 _mm_add_pd(_mm_mul_pd(dlij_SSE0,t1_SSE0),
1520 _mm_add_pd(t2_SSE0,t3_SSE0)));
1521 t1_SSE1 = _mm_mul_pd(rinv_SSE1,
1522 _mm_add_pd(_mm_mul_pd(dlij_SSE1,t1_SSE1),
1523 _mm_add_pd(t2_SSE1,t3_SSE1)));
1525 _mm_store_pd(dadx,_mm_and_pd(t1_SSE0,obc_mask1_SSE0));
1527 _mm_store_pd(dadx,_mm_and_pd(t1_SSE1,obc_mask1_SSE1));
1530 /* Evaluate influence of atom ai -> aj */
1531 t1_SSE0 = _mm_add_pd(dr_SSE0,sk_ai_SSE0);
1532 t1_SSE1 = _mm_add_pd(dr_SSE1,sk_ai_SSE1);
1533 t2_SSE0 = _mm_sub_pd(dr_SSE0,sk_ai_SSE0);
1534 t2_SSE1 = _mm_sub_pd(dr_SSE1,sk_ai_SSE1);
1535 t3_SSE0 = _mm_sub_pd(sk_ai_SSE0,dr_SSE0);
1536 t3_SSE1 = _mm_sub_pd(sk_ai_SSE1,dr_SSE1);
1538 obc_mask1_SSE0 = _mm_cmplt_pd(raj_SSE, t1_SSE0);
1539 obc_mask1_SSE1 = _mm_cmplt_pd(raj_SSE, t1_SSE1);
1540 obc_mask2_SSE0 = _mm_cmplt_pd(raj_SSE, t2_SSE0);
1541 obc_mask2_SSE1 = _mm_cmplt_pd(raj_SSE, t2_SSE1);
1542 obc_mask3_SSE0 = _mm_cmplt_pd(raj_SSE, t3_SSE0);
1543 obc_mask3_SSE1 = _mm_cmplt_pd(raj_SSE, t3_SSE1);
1544 obc_mask1_SSE0 = _mm_and_pd(obc_mask1_SSE0,jmask_SSE0);
1545 obc_mask1_SSE1 = _mm_and_pd(obc_mask1_SSE1,jmask_SSE1);
1547 uij_SSE0 = gmx_mm_inv_pd(t1_SSE0);
1548 uij_SSE1 = gmx_mm_inv_pd(t1_SSE1);
1549 lij_SSE0 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE0,gmx_mm_inv_pd(t2_SSE0)),
1550 _mm_andnot_pd(obc_mask2_SSE0,raj_inv_SSE));
1551 lij_SSE1 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE1,gmx_mm_inv_pd(t2_SSE1)),
1552 _mm_andnot_pd(obc_mask2_SSE1,raj_inv_SSE));
1553 dlij_SSE0 = _mm_and_pd(one_SSE,obc_mask2_SSE0);
1554 dlij_SSE1 = _mm_and_pd(one_SSE,obc_mask2_SSE1);
1556 uij2_SSE0 = _mm_mul_pd(uij_SSE0, uij_SSE0);
1557 uij2_SSE1 = _mm_mul_pd(uij_SSE1, uij_SSE1);
1558 uij3_SSE0 = _mm_mul_pd(uij2_SSE0,uij_SSE0);
1559 uij3_SSE1 = _mm_mul_pd(uij2_SSE1,uij_SSE1);
1560 lij2_SSE0 = _mm_mul_pd(lij_SSE0, lij_SSE0);
1561 lij2_SSE1 = _mm_mul_pd(lij_SSE1, lij_SSE1);
1562 lij3_SSE0 = _mm_mul_pd(lij2_SSE0,lij_SSE0);
1563 lij3_SSE1 = _mm_mul_pd(lij2_SSE1,lij_SSE1);
1565 diff2_SSE0 = _mm_sub_pd(uij2_SSE0,lij2_SSE0);
1566 diff2_SSE1 = _mm_sub_pd(uij2_SSE1,lij2_SSE1);
1567 lij_inv_SSE0 = gmx_mm_invsqrt_pd(lij2_SSE0);
1568 lij_inv_SSE1 = gmx_mm_invsqrt_pd(lij2_SSE1);
1569 sk2_rinv_SSE0 = _mm_mul_pd(sk2_ai_SSE0,rinv_SSE0);
1570 sk2_rinv_SSE1 = _mm_mul_pd(sk2_ai_SSE1,rinv_SSE1);
1571 prod_SSE0 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE0);
1572 prod_SSE1 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE1);
1574 logterm_SSE0 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE0,lij_inv_SSE0));
1575 logterm_SSE1 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE1,lij_inv_SSE1));
1576 t1_SSE0 = _mm_sub_pd(lij_SSE0,uij_SSE0);
1577 t1_SSE1 = _mm_sub_pd(lij_SSE1,uij_SSE1);
1578 t2_SSE0 = _mm_mul_pd(diff2_SSE0,
1579 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE0),
1581 t2_SSE1 = _mm_mul_pd(diff2_SSE1,
1582 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE1),
1584 t3_SSE0 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE0,logterm_SSE0));
1585 t3_SSE1 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE1,logterm_SSE1));
1586 t1_SSE0 = _mm_add_pd(t1_SSE0,_mm_add_pd(t2_SSE0,t3_SSE0));
1587 t1_SSE1 = _mm_add_pd(t1_SSE1,_mm_add_pd(t2_SSE1,t3_SSE1));
1588 t4_SSE0 = _mm_mul_pd(two_SSE,_mm_sub_pd(raj_inv_SSE,lij_SSE0));
1589 t4_SSE1 = _mm_mul_pd(two_SSE,_mm_sub_pd(raj_inv_SSE,lij_SSE1));
1590 t4_SSE0 = _mm_and_pd(t4_SSE0,obc_mask3_SSE0);
1591 t4_SSE1 = _mm_and_pd(t4_SSE1,obc_mask3_SSE1);
1592 t1_SSE0 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE0,t4_SSE0));
1593 t1_SSE1 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE1,t4_SSE1));
1595 _mm_store_pd(work+j, _mm_add_pd(_mm_load_pd(work+j),
1596 _mm_add_pd(_mm_and_pd(t1_SSE0,obc_mask1_SSE0),
1597 _mm_and_pd(t1_SSE1,obc_mask1_SSE1))));
1599 t1_SSE0 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE0),
1600 _mm_mul_pd(prod_SSE0,lij3_SSE0));
1601 t1_SSE1 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE1),
1602 _mm_mul_pd(prod_SSE1,lij3_SSE1));
1603 t1_SSE0 = _mm_sub_pd(t1_SSE0,
1604 _mm_mul_pd(onefourth_SSE,
1605 _mm_add_pd(_mm_mul_pd(lij_SSE0,rinv_SSE0),
1606 _mm_mul_pd(lij3_SSE0,dr_SSE0))));
1607 t1_SSE1 = _mm_sub_pd(t1_SSE1,
1608 _mm_mul_pd(onefourth_SSE,
1609 _mm_add_pd(_mm_mul_pd(lij_SSE1,rinv_SSE1),
1610 _mm_mul_pd(lij3_SSE1,dr_SSE1))));
1611 t2_SSE0 = _mm_mul_pd(onefourth_SSE,
1612 _mm_add_pd(_mm_mul_pd(uij_SSE0,rinv_SSE0),
1613 _mm_mul_pd(uij3_SSE0,dr_SSE0)));
1614 t2_SSE1 = _mm_mul_pd(onefourth_SSE,
1615 _mm_add_pd(_mm_mul_pd(uij_SSE1,rinv_SSE1),
1616 _mm_mul_pd(uij3_SSE1,dr_SSE1)));
1617 t2_SSE0 = _mm_sub_pd(t2_SSE0,
1618 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE0),
1619 _mm_mul_pd(prod_SSE0,uij3_SSE0)));
1620 t2_SSE1 = _mm_sub_pd(t2_SSE1,
1621 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE1),
1622 _mm_mul_pd(prod_SSE1,uij3_SSE1)));
1624 t3_SSE0 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE0),
1625 _mm_mul_pd(rinv_SSE0,rinv_SSE0));
1626 t3_SSE1 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE1),
1627 _mm_mul_pd(rinv_SSE1,rinv_SSE1));
1629 t3_SSE0 = _mm_sub_pd(t3_SSE0,
1630 _mm_mul_pd(_mm_mul_pd(diff2_SSE0,oneeighth_SSE),
1632 _mm_mul_pd(sk2_rinv_SSE0,rinv_SSE0))));
1633 t3_SSE1 = _mm_sub_pd(t3_SSE1,
1634 _mm_mul_pd(_mm_mul_pd(diff2_SSE1,oneeighth_SSE),
1636 _mm_mul_pd(sk2_rinv_SSE1,rinv_SSE1))));
1639 t1_SSE0 = _mm_mul_pd(rinv_SSE0,
1640 _mm_add_pd(_mm_mul_pd(dlij_SSE0,t1_SSE0),
1641 _mm_add_pd(t2_SSE0,t3_SSE0)));
1642 t1_SSE1 = _mm_mul_pd(rinv_SSE1,
1643 _mm_add_pd(_mm_mul_pd(dlij_SSE1,t1_SSE1),
1644 _mm_add_pd(t2_SSE1,t3_SSE1)));
1646 _mm_store_pd(dadx,_mm_and_pd(t1_SSE0,obc_mask1_SSE0));
1648 _mm_store_pd(dadx,_mm_and_pd(t1_SSE1,obc_mask1_SSE1));
1652 /* Main part, no exclusions */
1653 for(j=nj1; j<nj2; j+=UNROLLJ)
1655 /* load j atom coordinates */
1656 jx_SSE = _mm_load_pd(x_align+j);
1657 jy_SSE = _mm_load_pd(y_align+j);
1658 jz_SSE = _mm_load_pd(z_align+j);
1660 /* Calculate distance */
1661 dx_SSE0 = _mm_sub_pd(ix_SSE0,jx_SSE);
1662 dy_SSE0 = _mm_sub_pd(iy_SSE0,jy_SSE);
1663 dz_SSE0 = _mm_sub_pd(iz_SSE0,jz_SSE);
1664 dx_SSE1 = _mm_sub_pd(ix_SSE1,jx_SSE);
1665 dy_SSE1 = _mm_sub_pd(iy_SSE1,jy_SSE);
1666 dz_SSE1 = _mm_sub_pd(iz_SSE1,jz_SSE);
1668 /* rsq = dx*dx+dy*dy+dz*dz */
1669 rsq_SSE0 = gmx_mm_calc_rsq_pd(dx_SSE0,dy_SSE0,dz_SSE0);
1670 rsq_SSE1 = gmx_mm_calc_rsq_pd(dx_SSE1,dy_SSE1,dz_SSE1);
1672 /* Calculate 1/r and 1/r2 */
1673 rinv_SSE0 = gmx_mm_invsqrt_pd(rsq_SSE0);
1674 rinv_SSE1 = gmx_mm_invsqrt_pd(rsq_SSE1);
1677 rinv_SSE0 = _mm_and_pd(rinv_SSE0,imask_SSE0);
1678 rinv_SSE1 = _mm_and_pd(rinv_SSE1,imask_SSE1);
1680 dr_SSE0 = _mm_mul_pd(rsq_SSE0,rinv_SSE0);
1681 dr_SSE1 = _mm_mul_pd(rsq_SSE1,rinv_SSE1);
1683 sk_aj_SSE = _mm_load_pd(obc_param+j);
1684 raj_SSE = _mm_load_pd(gb_radius+j);
1686 raj_inv_SSE = gmx_mm_inv_pd(raj_SSE);
1688 /* Evaluate influence of atom aj -> ai */
1689 t1_SSE0 = _mm_add_pd(dr_SSE0,sk_aj_SSE);
1690 t1_SSE1 = _mm_add_pd(dr_SSE1,sk_aj_SSE);
1691 t2_SSE0 = _mm_sub_pd(dr_SSE0,sk_aj_SSE);
1692 t2_SSE1 = _mm_sub_pd(dr_SSE1,sk_aj_SSE);
1693 t3_SSE0 = _mm_sub_pd(sk_aj_SSE,dr_SSE0);
1694 t3_SSE1 = _mm_sub_pd(sk_aj_SSE,dr_SSE1);
1696 obc_mask1_SSE0 = _mm_cmplt_pd(rai_SSE0, t1_SSE0);
1697 obc_mask1_SSE1 = _mm_cmplt_pd(rai_SSE1, t1_SSE1);
1698 obc_mask2_SSE0 = _mm_cmplt_pd(rai_SSE0, t2_SSE0);
1699 obc_mask2_SSE1 = _mm_cmplt_pd(rai_SSE1, t2_SSE1);
1700 obc_mask3_SSE0 = _mm_cmplt_pd(rai_SSE0, t3_SSE0);
1701 obc_mask3_SSE1 = _mm_cmplt_pd(rai_SSE1, t3_SSE1);
1702 obc_mask1_SSE0 = _mm_and_pd(obc_mask1_SSE0,imask_SSE0);
1703 obc_mask1_SSE1 = _mm_and_pd(obc_mask1_SSE1,imask_SSE1);
1705 uij_SSE0 = gmx_mm_inv_pd(t1_SSE0);
1706 uij_SSE1 = gmx_mm_inv_pd(t1_SSE1);
1707 lij_SSE0 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE0,gmx_mm_inv_pd(t2_SSE0)),
1708 _mm_andnot_pd(obc_mask2_SSE0,rai_inv_SSE0));
1709 lij_SSE1 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE1,gmx_mm_inv_pd(t2_SSE1)),
1710 _mm_andnot_pd(obc_mask2_SSE1,rai_inv_SSE1));
1711 dlij_SSE0 = _mm_and_pd(one_SSE,obc_mask2_SSE0);
1712 dlij_SSE1 = _mm_and_pd(one_SSE,obc_mask2_SSE1);
1714 uij2_SSE0 = _mm_mul_pd(uij_SSE0, uij_SSE0);
1715 uij2_SSE1 = _mm_mul_pd(uij_SSE1, uij_SSE1);
1716 uij3_SSE0 = _mm_mul_pd(uij2_SSE0,uij_SSE0);
1717 uij3_SSE1 = _mm_mul_pd(uij2_SSE1,uij_SSE1);
1718 lij2_SSE0 = _mm_mul_pd(lij_SSE0, lij_SSE0);
1719 lij2_SSE1 = _mm_mul_pd(lij_SSE1, lij_SSE1);
1720 lij3_SSE0 = _mm_mul_pd(lij2_SSE0,lij_SSE0);
1721 lij3_SSE1 = _mm_mul_pd(lij2_SSE1,lij_SSE1);
1723 diff2_SSE0 = _mm_sub_pd(uij2_SSE0,lij2_SSE0);
1724 diff2_SSE1 = _mm_sub_pd(uij2_SSE1,lij2_SSE1);
1725 lij_inv_SSE0 = gmx_mm_invsqrt_pd(lij2_SSE0);
1726 lij_inv_SSE1 = gmx_mm_invsqrt_pd(lij2_SSE1);
1727 sk2_aj_SSE = _mm_mul_pd(sk_aj_SSE,sk_aj_SSE);
1728 sk2_rinv_SSE0 = _mm_mul_pd(sk2_aj_SSE,rinv_SSE0);
1729 sk2_rinv_SSE1 = _mm_mul_pd(sk2_aj_SSE,rinv_SSE1);
1730 prod_SSE0 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE0);
1731 prod_SSE1 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE1);
1733 logterm_SSE0 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE0,lij_inv_SSE0));
1734 logterm_SSE1 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE1,lij_inv_SSE1));
1736 t1_SSE0 = _mm_sub_pd(lij_SSE0,uij_SSE0);
1737 t1_SSE1 = _mm_sub_pd(lij_SSE1,uij_SSE1);
1738 t2_SSE0 = _mm_mul_pd(diff2_SSE0,
1739 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE0),
1741 t2_SSE1 = _mm_mul_pd(diff2_SSE1,
1742 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE1),
1745 t3_SSE0 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE0,logterm_SSE0));
1746 t3_SSE1 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE1,logterm_SSE1));
1747 t1_SSE0 = _mm_add_pd(t1_SSE0,_mm_add_pd(t2_SSE0,t3_SSE0));
1748 t1_SSE1 = _mm_add_pd(t1_SSE1,_mm_add_pd(t2_SSE1,t3_SSE1));
1749 t4_SSE0 = _mm_mul_pd(two_SSE,_mm_sub_pd(rai_inv_SSE0,lij_SSE0));
1750 t4_SSE1 = _mm_mul_pd(two_SSE,_mm_sub_pd(rai_inv_SSE1,lij_SSE1));
1751 t4_SSE0 = _mm_and_pd(t4_SSE0,obc_mask3_SSE0);
1752 t4_SSE1 = _mm_and_pd(t4_SSE1,obc_mask3_SSE1);
1753 t1_SSE0 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE0,t4_SSE0));
1754 t1_SSE1 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE1,t4_SSE1));
1756 sum_ai_SSE0 = _mm_add_pd(sum_ai_SSE0,_mm_and_pd(t1_SSE0,obc_mask1_SSE0));
1757 sum_ai_SSE1 = _mm_add_pd(sum_ai_SSE1,_mm_and_pd(t1_SSE1,obc_mask1_SSE1));
1759 t1_SSE0 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE0),
1760 _mm_mul_pd(prod_SSE0,lij3_SSE0));
1761 t1_SSE1 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE1),
1762 _mm_mul_pd(prod_SSE1,lij3_SSE1));
1764 t1_SSE0 = _mm_sub_pd(t1_SSE0,
1765 _mm_mul_pd(onefourth_SSE,
1766 _mm_add_pd(_mm_mul_pd(lij_SSE0,rinv_SSE0),
1767 _mm_mul_pd(lij3_SSE0,dr_SSE0))));
1768 t1_SSE1 = _mm_sub_pd(t1_SSE1,
1769 _mm_mul_pd(onefourth_SSE,
1770 _mm_add_pd(_mm_mul_pd(lij_SSE1,rinv_SSE1),
1771 _mm_mul_pd(lij3_SSE1,dr_SSE1))));
1773 t2_SSE0 = _mm_mul_pd(onefourth_SSE,
1774 _mm_add_pd(_mm_mul_pd(uij_SSE0,rinv_SSE0),
1775 _mm_mul_pd(uij3_SSE0,dr_SSE0)));
1776 t2_SSE1 = _mm_mul_pd(onefourth_SSE,
1777 _mm_add_pd(_mm_mul_pd(uij_SSE1,rinv_SSE1),
1778 _mm_mul_pd(uij3_SSE1,dr_SSE1)));
1779 t2_SSE0 = _mm_sub_pd(t2_SSE0,
1780 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE0),
1781 _mm_mul_pd(prod_SSE0,uij3_SSE0)));
1782 t2_SSE1 = _mm_sub_pd(t2_SSE1,
1783 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE1),
1784 _mm_mul_pd(prod_SSE1,uij3_SSE1)));
1785 t3_SSE0 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE0),
1786 _mm_mul_pd(rinv_SSE0,rinv_SSE0));
1787 t3_SSE1 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE1),
1788 _mm_mul_pd(rinv_SSE1,rinv_SSE1));
1789 t3_SSE0 = _mm_sub_pd(t3_SSE0,
1790 _mm_mul_pd(_mm_mul_pd(diff2_SSE0,oneeighth_SSE),
1792 _mm_mul_pd(sk2_rinv_SSE0,rinv_SSE0))));
1793 t3_SSE1 = _mm_sub_pd(t3_SSE1,
1794 _mm_mul_pd(_mm_mul_pd(diff2_SSE1,oneeighth_SSE),
1796 _mm_mul_pd(sk2_rinv_SSE1,rinv_SSE1))));
1798 t1_SSE0 = _mm_mul_pd(rinv_SSE0,
1799 _mm_add_pd(_mm_mul_pd(dlij_SSE0,t1_SSE0),
1800 _mm_add_pd(t2_SSE0,t3_SSE0)));
1801 t1_SSE1 = _mm_mul_pd(rinv_SSE1,
1802 _mm_add_pd(_mm_mul_pd(dlij_SSE1,t1_SSE1),
1803 _mm_add_pd(t2_SSE1,t3_SSE1)));
1805 _mm_store_pd(dadx,_mm_and_pd(t1_SSE0,obc_mask1_SSE0));
1807 _mm_store_pd(dadx,_mm_and_pd(t1_SSE1,obc_mask1_SSE1));
1810 /* Evaluate influence of atom ai -> aj */
1811 t1_SSE0 = _mm_add_pd(dr_SSE0,sk_ai_SSE0);
1812 t1_SSE1 = _mm_add_pd(dr_SSE1,sk_ai_SSE1);
1813 t2_SSE0 = _mm_sub_pd(dr_SSE0,sk_ai_SSE0);
1814 t2_SSE1 = _mm_sub_pd(dr_SSE1,sk_ai_SSE1);
1815 t3_SSE0 = _mm_sub_pd(sk_ai_SSE0,dr_SSE0);
1816 t3_SSE1 = _mm_sub_pd(sk_ai_SSE1,dr_SSE1);
1818 obc_mask1_SSE0 = _mm_cmplt_pd(raj_SSE, t1_SSE0);
1819 obc_mask1_SSE1 = _mm_cmplt_pd(raj_SSE, t1_SSE1);
1820 obc_mask2_SSE0 = _mm_cmplt_pd(raj_SSE, t2_SSE0);
1821 obc_mask2_SSE1 = _mm_cmplt_pd(raj_SSE, t2_SSE1);
1822 obc_mask3_SSE0 = _mm_cmplt_pd(raj_SSE, t3_SSE0);
1823 obc_mask3_SSE1 = _mm_cmplt_pd(raj_SSE, t3_SSE1);
1824 obc_mask1_SSE0 = _mm_and_pd(obc_mask1_SSE0,imask_SSE0);
1825 obc_mask1_SSE1 = _mm_and_pd(obc_mask1_SSE1,imask_SSE1);
1827 uij_SSE0 = gmx_mm_inv_pd(t1_SSE0);
1828 uij_SSE1 = gmx_mm_inv_pd(t1_SSE1);
1829 lij_SSE0 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE0,gmx_mm_inv_pd(t2_SSE0)),
1830 _mm_andnot_pd(obc_mask2_SSE0,raj_inv_SSE));
1831 lij_SSE1 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE1,gmx_mm_inv_pd(t2_SSE1)),
1832 _mm_andnot_pd(obc_mask2_SSE1,raj_inv_SSE));
1833 dlij_SSE0 = _mm_and_pd(one_SSE,obc_mask2_SSE0);
1834 dlij_SSE1 = _mm_and_pd(one_SSE,obc_mask2_SSE1);
1836 uij2_SSE0 = _mm_mul_pd(uij_SSE0, uij_SSE0);
1837 uij2_SSE1 = _mm_mul_pd(uij_SSE1, uij_SSE1);
1838 uij3_SSE0 = _mm_mul_pd(uij2_SSE0,uij_SSE0);
1839 uij3_SSE1 = _mm_mul_pd(uij2_SSE1,uij_SSE1);
1840 lij2_SSE0 = _mm_mul_pd(lij_SSE0, lij_SSE0);
1841 lij2_SSE1 = _mm_mul_pd(lij_SSE1, lij_SSE1);
1842 lij3_SSE0 = _mm_mul_pd(lij2_SSE0,lij_SSE0);
1843 lij3_SSE1 = _mm_mul_pd(lij2_SSE1,lij_SSE1);
1845 diff2_SSE0 = _mm_sub_pd(uij2_SSE0,lij2_SSE0);
1846 diff2_SSE1 = _mm_sub_pd(uij2_SSE1,lij2_SSE1);
1847 lij_inv_SSE0 = gmx_mm_invsqrt_pd(lij2_SSE0);
1848 lij_inv_SSE1 = gmx_mm_invsqrt_pd(lij2_SSE1);
1849 sk2_rinv_SSE0 = _mm_mul_pd(sk2_ai_SSE0,rinv_SSE0);
1850 sk2_rinv_SSE1 = _mm_mul_pd(sk2_ai_SSE1,rinv_SSE1);
1851 prod_SSE0 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE0);
1852 prod_SSE1 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE1);
1854 logterm_SSE0 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE0,lij_inv_SSE0));
1855 logterm_SSE1 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE1,lij_inv_SSE1));
1856 t1_SSE0 = _mm_sub_pd(lij_SSE0,uij_SSE0);
1857 t1_SSE1 = _mm_sub_pd(lij_SSE1,uij_SSE1);
1858 t2_SSE0 = _mm_mul_pd(diff2_SSE0,
1859 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE0),
1861 t2_SSE1 = _mm_mul_pd(diff2_SSE1,
1862 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE1),
1864 t3_SSE0 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE0,logterm_SSE0));
1865 t3_SSE1 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE1,logterm_SSE1));
1866 t1_SSE0 = _mm_add_pd(t1_SSE0,_mm_add_pd(t2_SSE0,t3_SSE0));
1867 t1_SSE1 = _mm_add_pd(t1_SSE1,_mm_add_pd(t2_SSE1,t3_SSE1));
1868 t4_SSE0 = _mm_mul_pd(two_SSE,_mm_sub_pd(raj_inv_SSE,lij_SSE0));
1869 t4_SSE1 = _mm_mul_pd(two_SSE,_mm_sub_pd(raj_inv_SSE,lij_SSE1));
1870 t4_SSE0 = _mm_and_pd(t4_SSE0,obc_mask3_SSE0);
1871 t4_SSE1 = _mm_and_pd(t4_SSE1,obc_mask3_SSE1);
1872 t1_SSE0 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE0,t4_SSE0));
1873 t1_SSE1 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE1,t4_SSE1));
1875 _mm_store_pd(work+j, _mm_add_pd(_mm_load_pd(work+j),
1876 _mm_add_pd(_mm_and_pd(t1_SSE0,obc_mask1_SSE0),
1877 _mm_and_pd(t1_SSE1,obc_mask1_SSE1))));
1879 t1_SSE0 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE0),
1880 _mm_mul_pd(prod_SSE0,lij3_SSE0));
1881 t1_SSE1 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE1),
1882 _mm_mul_pd(prod_SSE1,lij3_SSE1));
1883 t1_SSE0 = _mm_sub_pd(t1_SSE0,
1884 _mm_mul_pd(onefourth_SSE,
1885 _mm_add_pd(_mm_mul_pd(lij_SSE0,rinv_SSE0),
1886 _mm_mul_pd(lij3_SSE0,dr_SSE0))));
1887 t1_SSE1 = _mm_sub_pd(t1_SSE1,
1888 _mm_mul_pd(onefourth_SSE,
1889 _mm_add_pd(_mm_mul_pd(lij_SSE1,rinv_SSE1),
1890 _mm_mul_pd(lij3_SSE1,dr_SSE1))));
1891 t2_SSE0 = _mm_mul_pd(onefourth_SSE,
1892 _mm_add_pd(_mm_mul_pd(uij_SSE0,rinv_SSE0),
1893 _mm_mul_pd(uij3_SSE0,dr_SSE0)));
1894 t2_SSE1 = _mm_mul_pd(onefourth_SSE,
1895 _mm_add_pd(_mm_mul_pd(uij_SSE1,rinv_SSE1),
1896 _mm_mul_pd(uij3_SSE1,dr_SSE1)));
1897 t2_SSE0 = _mm_sub_pd(t2_SSE0,
1898 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE0),
1899 _mm_mul_pd(prod_SSE0,uij3_SSE0)));
1900 t2_SSE1 = _mm_sub_pd(t2_SSE1,
1901 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE1),
1902 _mm_mul_pd(prod_SSE1,uij3_SSE1)));
1904 t3_SSE0 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE0),
1905 _mm_mul_pd(rinv_SSE0,rinv_SSE0));
1906 t3_SSE1 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE1),
1907 _mm_mul_pd(rinv_SSE1,rinv_SSE1));
1909 t3_SSE0 = _mm_sub_pd(t3_SSE0,
1910 _mm_mul_pd(_mm_mul_pd(diff2_SSE0,oneeighth_SSE),
1912 _mm_mul_pd(sk2_rinv_SSE0,rinv_SSE0))));
1913 t3_SSE1 = _mm_sub_pd(t3_SSE1,
1914 _mm_mul_pd(_mm_mul_pd(diff2_SSE1,oneeighth_SSE),
1916 _mm_mul_pd(sk2_rinv_SSE1,rinv_SSE1))));
1918 t1_SSE0 = _mm_mul_pd(rinv_SSE0,
1919 _mm_add_pd(_mm_mul_pd(dlij_SSE0,t1_SSE0),
1920 _mm_add_pd(t2_SSE0,t3_SSE0)));
1921 t1_SSE1 = _mm_mul_pd(rinv_SSE1,
1922 _mm_add_pd(_mm_mul_pd(dlij_SSE1,t1_SSE1),
1923 _mm_add_pd(t2_SSE1,t3_SSE1)));
1925 _mm_store_pd(dadx,_mm_and_pd(t1_SSE0,obc_mask1_SSE0));
1927 _mm_store_pd(dadx,_mm_and_pd(t1_SSE1,obc_mask1_SSE1));
1931 /* Epilogue part, including exclusion mask */
1932 for(j=nj2; j<nj3; j+=UNROLLJ)
1934 jmask_SSE0 = _mm_load_pd((double *)emask0);
1935 jmask_SSE1 = _mm_load_pd((double *)emask1);
1936 emask0 += 2*UNROLLJ;
1937 emask1 += 2*UNROLLJ;
1939 /* load j atom coordinates */
1940 jx_SSE = _mm_load_pd(x_align+j);
1941 jy_SSE = _mm_load_pd(y_align+j);
1942 jz_SSE = _mm_load_pd(z_align+j);
1944 /* Calculate distance */
1945 dx_SSE0 = _mm_sub_pd(ix_SSE0,jx_SSE);
1946 dy_SSE0 = _mm_sub_pd(iy_SSE0,jy_SSE);
1947 dz_SSE0 = _mm_sub_pd(iz_SSE0,jz_SSE);
1948 dx_SSE1 = _mm_sub_pd(ix_SSE1,jx_SSE);
1949 dy_SSE1 = _mm_sub_pd(iy_SSE1,jy_SSE);
1950 dz_SSE1 = _mm_sub_pd(iz_SSE1,jz_SSE);
1952 /* rsq = dx*dx+dy*dy+dz*dz */
1953 rsq_SSE0 = gmx_mm_calc_rsq_pd(dx_SSE0,dy_SSE0,dz_SSE0);
1954 rsq_SSE1 = gmx_mm_calc_rsq_pd(dx_SSE1,dy_SSE1,dz_SSE1);
1957 jmask_SSE0 = _mm_and_pd(jmask_SSE0,imask_SSE0);
1958 jmask_SSE1 = _mm_and_pd(jmask_SSE1,imask_SSE1);
1960 /* Calculate 1/r and 1/r2 */
1961 rinv_SSE0 = gmx_mm_invsqrt_pd(rsq_SSE0);
1962 rinv_SSE1 = gmx_mm_invsqrt_pd(rsq_SSE1);
1965 rinv_SSE0 = _mm_and_pd(rinv_SSE0,jmask_SSE0);
1966 rinv_SSE1 = _mm_and_pd(rinv_SSE1,jmask_SSE1);
1968 dr_SSE0 = _mm_mul_pd(rsq_SSE0,rinv_SSE0);
1969 dr_SSE1 = _mm_mul_pd(rsq_SSE1,rinv_SSE1);
1971 sk_aj_SSE = _mm_load_pd(obc_param+j);
1972 raj_SSE = _mm_load_pd(gb_radius+j);
1974 raj_inv_SSE = gmx_mm_inv_pd(raj_SSE);
1976 /* Evaluate influence of atom aj -> ai */
1977 t1_SSE0 = _mm_add_pd(dr_SSE0,sk_aj_SSE);
1978 t1_SSE1 = _mm_add_pd(dr_SSE1,sk_aj_SSE);
1979 t2_SSE0 = _mm_sub_pd(dr_SSE0,sk_aj_SSE);
1980 t2_SSE1 = _mm_sub_pd(dr_SSE1,sk_aj_SSE);
1981 t3_SSE0 = _mm_sub_pd(sk_aj_SSE,dr_SSE0);
1982 t3_SSE1 = _mm_sub_pd(sk_aj_SSE,dr_SSE1);
1984 obc_mask1_SSE0 = _mm_cmplt_pd(rai_SSE0, t1_SSE0);
1985 obc_mask1_SSE1 = _mm_cmplt_pd(rai_SSE1, t1_SSE1);
1986 obc_mask2_SSE0 = _mm_cmplt_pd(rai_SSE0, t2_SSE0);
1987 obc_mask2_SSE1 = _mm_cmplt_pd(rai_SSE1, t2_SSE1);
1988 obc_mask3_SSE0 = _mm_cmplt_pd(rai_SSE0, t3_SSE0);
1989 obc_mask3_SSE1 = _mm_cmplt_pd(rai_SSE1, t3_SSE1);
1990 obc_mask1_SSE0 = _mm_and_pd(obc_mask1_SSE0,jmask_SSE0);
1991 obc_mask1_SSE1 = _mm_and_pd(obc_mask1_SSE1,jmask_SSE1);
1993 uij_SSE0 = gmx_mm_inv_pd(t1_SSE0);
1994 uij_SSE1 = gmx_mm_inv_pd(t1_SSE1);
1995 lij_SSE0 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE0,gmx_mm_inv_pd(t2_SSE0)),
1996 _mm_andnot_pd(obc_mask2_SSE0,rai_inv_SSE0));
1997 lij_SSE1 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE1,gmx_mm_inv_pd(t2_SSE1)),
1998 _mm_andnot_pd(obc_mask2_SSE1,rai_inv_SSE1));
2000 dlij_SSE0 = _mm_and_pd(one_SSE,obc_mask2_SSE0);
2001 dlij_SSE1 = _mm_and_pd(one_SSE,obc_mask2_SSE1);
2003 uij2_SSE0 = _mm_mul_pd(uij_SSE0, uij_SSE0);
2004 uij2_SSE1 = _mm_mul_pd(uij_SSE1, uij_SSE1);
2005 uij3_SSE0 = _mm_mul_pd(uij2_SSE0,uij_SSE0);
2006 uij3_SSE1 = _mm_mul_pd(uij2_SSE1,uij_SSE1);
2007 lij2_SSE0 = _mm_mul_pd(lij_SSE0, lij_SSE0);
2008 lij2_SSE1 = _mm_mul_pd(lij_SSE1, lij_SSE1);
2009 lij3_SSE0 = _mm_mul_pd(lij2_SSE0,lij_SSE0);
2010 lij3_SSE1 = _mm_mul_pd(lij2_SSE1,lij_SSE1);
2012 diff2_SSE0 = _mm_sub_pd(uij2_SSE0,lij2_SSE0);
2013 diff2_SSE1 = _mm_sub_pd(uij2_SSE1,lij2_SSE1);
2014 lij_inv_SSE0 = gmx_mm_invsqrt_pd(lij2_SSE0);
2015 lij_inv_SSE1 = gmx_mm_invsqrt_pd(lij2_SSE1);
2016 sk2_aj_SSE = _mm_mul_pd(sk_aj_SSE,sk_aj_SSE);
2017 sk2_rinv_SSE0 = _mm_mul_pd(sk2_aj_SSE,rinv_SSE0);
2018 sk2_rinv_SSE1 = _mm_mul_pd(sk2_aj_SSE,rinv_SSE1);
2019 prod_SSE0 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE0);
2020 prod_SSE1 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE1);
2022 logterm_SSE0 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE0,lij_inv_SSE0));
2023 logterm_SSE1 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE1,lij_inv_SSE1));
2025 t1_SSE0 = _mm_sub_pd(lij_SSE0,uij_SSE0);
2026 t1_SSE1 = _mm_sub_pd(lij_SSE1,uij_SSE1);
2027 t2_SSE0 = _mm_mul_pd(diff2_SSE0,
2028 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE0),
2030 t2_SSE1 = _mm_mul_pd(diff2_SSE1,
2031 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE1),
2034 t3_SSE0 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE0,logterm_SSE0));
2035 t3_SSE1 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE1,logterm_SSE1));
2036 t1_SSE0 = _mm_add_pd(t1_SSE0,_mm_add_pd(t2_SSE0,t3_SSE0));
2037 t1_SSE1 = _mm_add_pd(t1_SSE1,_mm_add_pd(t2_SSE1,t3_SSE1));
2038 t4_SSE0 = _mm_mul_pd(two_SSE,_mm_sub_pd(rai_inv_SSE0,lij_SSE0));
2039 t4_SSE1 = _mm_mul_pd(two_SSE,_mm_sub_pd(rai_inv_SSE1,lij_SSE1));
2040 t4_SSE0 = _mm_and_pd(t4_SSE0,obc_mask3_SSE0);
2041 t4_SSE1 = _mm_and_pd(t4_SSE1,obc_mask3_SSE1);
2042 t1_SSE0 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE0,t4_SSE0));
2043 t1_SSE1 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE1,t4_SSE1));
2045 sum_ai_SSE0 = _mm_add_pd(sum_ai_SSE0,_mm_and_pd(t1_SSE0,obc_mask1_SSE0));
2046 sum_ai_SSE1 = _mm_add_pd(sum_ai_SSE1,_mm_and_pd(t1_SSE1,obc_mask1_SSE1));
2048 t1_SSE0 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE0),
2049 _mm_mul_pd(prod_SSE0,lij3_SSE0));
2050 t1_SSE1 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE1),
2051 _mm_mul_pd(prod_SSE1,lij3_SSE1));
2052 t1_SSE0 = _mm_sub_pd(t1_SSE0,
2053 _mm_mul_pd(onefourth_SSE,
2054 _mm_add_pd(_mm_mul_pd(lij_SSE0,rinv_SSE0),
2055 _mm_mul_pd(lij3_SSE0,dr_SSE0))));
2056 t1_SSE1 = _mm_sub_pd(t1_SSE1,
2057 _mm_mul_pd(onefourth_SSE,
2058 _mm_add_pd(_mm_mul_pd(lij_SSE1,rinv_SSE1),
2059 _mm_mul_pd(lij3_SSE1,dr_SSE1))));
2061 t2_SSE0 = _mm_mul_pd(onefourth_SSE,
2062 _mm_add_pd(_mm_mul_pd(uij_SSE0,rinv_SSE0),
2063 _mm_mul_pd(uij3_SSE0,dr_SSE0)));
2064 t2_SSE1 = _mm_mul_pd(onefourth_SSE,
2065 _mm_add_pd(_mm_mul_pd(uij_SSE1,rinv_SSE1),
2066 _mm_mul_pd(uij3_SSE1,dr_SSE1)));
2067 t2_SSE0 = _mm_sub_pd(t2_SSE0,
2068 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE0),
2069 _mm_mul_pd(prod_SSE0,uij3_SSE0)));
2070 t2_SSE1 = _mm_sub_pd(t2_SSE1,
2071 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE1),
2072 _mm_mul_pd(prod_SSE1,uij3_SSE1)));
2073 t3_SSE0 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE0),
2074 _mm_mul_pd(rinv_SSE0,rinv_SSE0));
2075 t3_SSE1 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE1),
2076 _mm_mul_pd(rinv_SSE1,rinv_SSE1));
2077 t3_SSE0 = _mm_sub_pd(t3_SSE0,
2078 _mm_mul_pd(_mm_mul_pd(diff2_SSE0,oneeighth_SSE),
2080 _mm_mul_pd(sk2_rinv_SSE0,rinv_SSE0))));
2081 t3_SSE1 = _mm_sub_pd(t3_SSE1,
2082 _mm_mul_pd(_mm_mul_pd(diff2_SSE1,oneeighth_SSE),
2084 _mm_mul_pd(sk2_rinv_SSE1,rinv_SSE1))));
2086 t1_SSE0 = _mm_mul_pd(rinv_SSE0,
2087 _mm_add_pd(_mm_mul_pd(dlij_SSE0,t1_SSE0),
2088 _mm_add_pd(t2_SSE0,t3_SSE0)));
2089 t1_SSE1 = _mm_mul_pd(rinv_SSE1,
2090 _mm_add_pd(_mm_mul_pd(dlij_SSE1,t1_SSE1),
2091 _mm_add_pd(t2_SSE1,t3_SSE1)));
2093 _mm_store_pd(dadx,_mm_and_pd(t1_SSE0,obc_mask1_SSE0));
2095 _mm_store_pd(dadx,_mm_and_pd(t1_SSE1,obc_mask1_SSE1));
2098 /* Evaluate influence of atom ai -> aj */
2099 t1_SSE0 = _mm_add_pd(dr_SSE0,sk_ai_SSE0);
2100 t1_SSE1 = _mm_add_pd(dr_SSE1,sk_ai_SSE1);
2101 t2_SSE0 = _mm_sub_pd(dr_SSE0,sk_ai_SSE0);
2102 t2_SSE1 = _mm_sub_pd(dr_SSE1,sk_ai_SSE1);
2103 t3_SSE0 = _mm_sub_pd(sk_ai_SSE0,dr_SSE0);
2104 t3_SSE1 = _mm_sub_pd(sk_ai_SSE1,dr_SSE1);
2106 obc_mask1_SSE0 = _mm_cmplt_pd(raj_SSE, t1_SSE0);
2107 obc_mask1_SSE1 = _mm_cmplt_pd(raj_SSE, t1_SSE1);
2108 obc_mask2_SSE0 = _mm_cmplt_pd(raj_SSE, t2_SSE0);
2109 obc_mask2_SSE1 = _mm_cmplt_pd(raj_SSE, t2_SSE1);
2110 obc_mask3_SSE0 = _mm_cmplt_pd(raj_SSE, t3_SSE0);
2111 obc_mask3_SSE1 = _mm_cmplt_pd(raj_SSE, t3_SSE1);
2112 obc_mask1_SSE0 = _mm_and_pd(obc_mask1_SSE0,jmask_SSE0);
2113 obc_mask1_SSE1 = _mm_and_pd(obc_mask1_SSE1,jmask_SSE1);
2115 uij_SSE0 = gmx_mm_inv_pd(t1_SSE0);
2116 uij_SSE1 = gmx_mm_inv_pd(t1_SSE1);
2117 lij_SSE0 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE0,gmx_mm_inv_pd(t2_SSE0)),
2118 _mm_andnot_pd(obc_mask2_SSE0,raj_inv_SSE));
2119 lij_SSE1 = _mm_or_pd( _mm_and_pd(obc_mask2_SSE1,gmx_mm_inv_pd(t2_SSE1)),
2120 _mm_andnot_pd(obc_mask2_SSE1,raj_inv_SSE));
2122 dlij_SSE0 = _mm_and_pd(one_SSE,obc_mask2_SSE0);
2123 dlij_SSE1 = _mm_and_pd(one_SSE,obc_mask2_SSE1);
2125 uij2_SSE0 = _mm_mul_pd(uij_SSE0, uij_SSE0);
2126 uij2_SSE1 = _mm_mul_pd(uij_SSE1, uij_SSE1);
2127 uij3_SSE0 = _mm_mul_pd(uij2_SSE0,uij_SSE0);
2128 uij3_SSE1 = _mm_mul_pd(uij2_SSE1,uij_SSE1);
2129 lij2_SSE0 = _mm_mul_pd(lij_SSE0, lij_SSE0);
2130 lij2_SSE1 = _mm_mul_pd(lij_SSE1, lij_SSE1);
2131 lij3_SSE0 = _mm_mul_pd(lij2_SSE0,lij_SSE0);
2132 lij3_SSE1 = _mm_mul_pd(lij2_SSE1,lij_SSE1);
2134 diff2_SSE0 = _mm_sub_pd(uij2_SSE0,lij2_SSE0);
2135 diff2_SSE1 = _mm_sub_pd(uij2_SSE1,lij2_SSE1);
2136 lij_inv_SSE0 = gmx_mm_invsqrt_pd(lij2_SSE0);
2137 lij_inv_SSE1 = gmx_mm_invsqrt_pd(lij2_SSE1);
2138 sk2_rinv_SSE0 = _mm_mul_pd(sk2_ai_SSE0,rinv_SSE0);
2139 sk2_rinv_SSE1 = _mm_mul_pd(sk2_ai_SSE1,rinv_SSE1);
2140 prod_SSE0 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE0);
2141 prod_SSE1 = _mm_mul_pd(onefourth_SSE,sk2_rinv_SSE1);
2143 logterm_SSE0 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE0,lij_inv_SSE0));
2144 logterm_SSE1 = gmx_mm_log_pd(_mm_mul_pd(uij_SSE1,lij_inv_SSE1));
2145 t1_SSE0 = _mm_sub_pd(lij_SSE0,uij_SSE0);
2146 t1_SSE1 = _mm_sub_pd(lij_SSE1,uij_SSE1);
2147 t2_SSE0 = _mm_mul_pd(diff2_SSE0,
2148 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE0),
2150 t2_SSE1 = _mm_mul_pd(diff2_SSE1,
2151 _mm_sub_pd(_mm_mul_pd(onefourth_SSE,dr_SSE1),
2153 t3_SSE0 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE0,logterm_SSE0));
2154 t3_SSE1 = _mm_mul_pd(half_SSE,_mm_mul_pd(rinv_SSE1,logterm_SSE1));
2155 t1_SSE0 = _mm_add_pd(t1_SSE0,_mm_add_pd(t2_SSE0,t3_SSE0));
2156 t1_SSE1 = _mm_add_pd(t1_SSE1,_mm_add_pd(t2_SSE1,t3_SSE1));
2157 t4_SSE0 = _mm_mul_pd(two_SSE,_mm_sub_pd(raj_inv_SSE,lij_SSE0));
2158 t4_SSE1 = _mm_mul_pd(two_SSE,_mm_sub_pd(raj_inv_SSE,lij_SSE1));
2159 t4_SSE0 = _mm_and_pd(t4_SSE0,obc_mask3_SSE0);
2160 t4_SSE1 = _mm_and_pd(t4_SSE1,obc_mask3_SSE1);
2161 t1_SSE0 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE0,t4_SSE0));
2162 t1_SSE1 = _mm_mul_pd(half_SSE,_mm_add_pd(t1_SSE1,t4_SSE1));
2164 _mm_store_pd(work+j, _mm_add_pd(_mm_load_pd(work+j),
2165 _mm_add_pd(_mm_and_pd(t1_SSE0,obc_mask1_SSE0),
2166 _mm_and_pd(t1_SSE1,obc_mask1_SSE1))));
2168 t1_SSE0 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE0),
2169 _mm_mul_pd(prod_SSE0,lij3_SSE0));
2170 t1_SSE1 = _mm_add_pd(_mm_mul_pd(half_SSE,lij2_SSE1),
2171 _mm_mul_pd(prod_SSE1,lij3_SSE1));
2173 t1_SSE0 = _mm_sub_pd(t1_SSE0,
2174 _mm_mul_pd(onefourth_SSE,
2175 _mm_add_pd(_mm_mul_pd(lij_SSE0,rinv_SSE0),
2176 _mm_mul_pd(lij3_SSE0,dr_SSE0))));
2177 t1_SSE1 = _mm_sub_pd(t1_SSE1,
2178 _mm_mul_pd(onefourth_SSE,
2179 _mm_add_pd(_mm_mul_pd(lij_SSE1,rinv_SSE1),
2180 _mm_mul_pd(lij3_SSE1,dr_SSE1))));
2181 t2_SSE0 = _mm_mul_pd(onefourth_SSE,
2182 _mm_add_pd(_mm_mul_pd(uij_SSE0,rinv_SSE0),
2183 _mm_mul_pd(uij3_SSE0,dr_SSE0)));
2184 t2_SSE1 = _mm_mul_pd(onefourth_SSE,
2185 _mm_add_pd(_mm_mul_pd(uij_SSE1,rinv_SSE1),
2186 _mm_mul_pd(uij3_SSE1,dr_SSE1)));
2187 t2_SSE0 = _mm_sub_pd(t2_SSE0,
2188 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE0),
2189 _mm_mul_pd(prod_SSE0,uij3_SSE0)));
2190 t2_SSE1 = _mm_sub_pd(t2_SSE1,
2191 _mm_add_pd(_mm_mul_pd(half_SSE,uij2_SSE1),
2192 _mm_mul_pd(prod_SSE1,uij3_SSE1)));
2194 t3_SSE0 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE0),
2195 _mm_mul_pd(rinv_SSE0,rinv_SSE0));
2196 t3_SSE1 = _mm_mul_pd(_mm_mul_pd(onefourth_SSE,logterm_SSE1),
2197 _mm_mul_pd(rinv_SSE1,rinv_SSE1));
2199 t3_SSE0 = _mm_sub_pd(t3_SSE0,
2200 _mm_mul_pd(_mm_mul_pd(diff2_SSE0,oneeighth_SSE),
2202 _mm_mul_pd(sk2_rinv_SSE0,rinv_SSE0))));
2203 t3_SSE1 = _mm_sub_pd(t3_SSE1,
2204 _mm_mul_pd(_mm_mul_pd(diff2_SSE1,oneeighth_SSE),
2206 _mm_mul_pd(sk2_rinv_SSE1,rinv_SSE1))));
2208 t1_SSE0 = _mm_mul_pd(rinv_SSE0,
2209 _mm_add_pd(_mm_mul_pd(dlij_SSE0,t1_SSE0),
2210 _mm_add_pd(t2_SSE0,t3_SSE0)));
2211 t1_SSE1 = _mm_mul_pd(rinv_SSE1,
2212 _mm_add_pd(_mm_mul_pd(dlij_SSE1,t1_SSE1),
2213 _mm_add_pd(t2_SSE1,t3_SSE1)));
2215 _mm_store_pd(dadx,_mm_and_pd(t1_SSE0,obc_mask1_SSE0));
2217 _mm_store_pd(dadx,_mm_and_pd(t1_SSE1,obc_mask1_SSE1));
2220 GMX_MM_TRANSPOSE2_PD(sum_ai_SSE0,sum_ai_SSE1);
2221 sum_ai_SSE0 = _mm_add_pd(sum_ai_SSE0,sum_ai_SSE1);
2222 _mm_store_pd(work+i, _mm_add_pd(sum_ai_SSE0, _mm_load_pd(work+i)));
2226 for(i=0;i<natoms/2+1;i++)
2228 work[i] += work[natoms+i];
2231 /* Parallel summations */
2235 gmx_sum(natoms,work, cr);
2238 if(gb_algorithm==egbHCT)
2241 for(i=0;i<natoms;i++)
2243 if(born->use[i] != 0)
2245 rai = top->atomtypes.gb_radius[mdatoms->typeA[i]]-born->gb_doffset;
2246 sum_ai = 1.0/rai - work[i];
2247 min_rad = rai + born->gb_doffset;
2250 born->bRad[i] = rad > min_rad ? rad : min_rad;
2251 fr->invsqrta[i] = gmx_invsqrt(born->bRad[i]);
2260 /* Calculate the radii */
2261 for(i=0;i<natoms;i++)
2264 if(born->use[i] != 0)
2266 rai = top->atomtypes.gb_radius[mdatoms->typeA[i]];
2268 rai = rai-born->gb_doffset;
2270 sum_ai = rai * work[i];
2271 sum_ai2 = sum_ai * sum_ai;
2272 sum_ai3 = sum_ai2 * sum_ai;
2274 tsum = tanh(born->obc_alpha*sum_ai-born->obc_beta*sum_ai2+born->obc_gamma*sum_ai3);
2275 born->bRad[i] = rai_inv - tsum*rai_inv2;
2276 born->bRad[i] = 1.0 / born->bRad[i];
2278 fr->invsqrta[i]=gmx_invsqrt(born->bRad[i]);
2280 tchain = rai * (born->obc_alpha-2*born->obc_beta*sum_ai+3*born->obc_gamma*sum_ai2);
2281 born->drobc[i] = (1.0-tsum*tsum)*tchain*rai_inv2;
2297 genborn_allvsall_calc_chainrule_sse2_double(t_forcerec * fr,
2298 t_mdatoms * mdatoms,
2299 gmx_genborn_t * born,
2305 gmx_allvsallgb2_data_t *aadata;
2308 int nj0,nj1,nj2,nj3;
2321 double rbai,rbaj,fgb,fgb_ai,rbi;
2332 __m128d jmask_SSE0,jmask_SSE1;
2333 __m128d ix_SSE0,iy_SSE0,iz_SSE0;
2334 __m128d ix_SSE1,iy_SSE1,iz_SSE1;
2335 __m128d fix_SSE0,fiy_SSE0,fiz_SSE0;
2336 __m128d fix_SSE1,fiy_SSE1,fiz_SSE1;
2337 __m128d rbai_SSE0,rbai_SSE1;
2338 __m128d imask_SSE0,imask_SSE1;
2339 __m128d jx_SSE,jy_SSE,jz_SSE,rbaj_SSE;
2340 __m128d dx_SSE0,dy_SSE0,dz_SSE0;
2341 __m128d dx_SSE1,dy_SSE1,dz_SSE1;
2342 __m128d fgb_SSE0,fgb_ai_SSE0;
2343 __m128d fgb_SSE1,fgb_ai_SSE1;
2344 __m128d tx_SSE0,ty_SSE0,tz_SSE0;
2345 __m128d tx_SSE1,ty_SSE1,tz_SSE1;
2346 __m128d t1,t2,tmpSSE;
2348 natoms = mdatoms->nr;
2349 ni0 = (mdatoms->start/SIMD_WIDTH)*SIMD_WIDTH;
2350 ni1 = mdatoms->start+mdatoms->homenr;
2352 aadata = (gmx_allvsallgb2_data_t *)paadata;
2354 x_align = aadata->x_align;
2355 y_align = aadata->y_align;
2356 z_align = aadata->z_align;
2357 fx_align = aadata->fx_align;
2358 fy_align = aadata->fy_align;
2359 fz_align = aadata->fz_align;
2361 jindex = aadata->jindex_gb;
2367 /* Loop to get the proper form for the Born radius term */
2368 if(gb_algorithm==egbSTILL)
2370 for(i=0;i<natoms;i++)
2372 rbi = born->bRad[i];
2373 rb[i] = (2 * rbi * rbi * fr->dvda[i])/ONE_4PI_EPS0;
2376 else if(gb_algorithm==egbHCT)
2378 for(i=0;i<natoms;i++)
2380 rbi = born->bRad[i];
2381 rb[i] = rbi * rbi * fr->dvda[i];
2384 else if(gb_algorithm==egbOBC)
2386 for(idx=0;idx<natoms;idx++)
2388 rbi = born->bRad[idx];
2389 rb[idx] = rbi * rbi * born->drobc[idx] * fr->dvda[idx];
2393 for(i=0;i<2*natoms;i++)
2401 for(i=0;i<natoms;i++)
2406 for(i=ni0; i<ni1; i+=UNROLLI)
2408 /* We assume shifts are NOT used for all-vs-all interactions */
2410 /* Load i atom data */
2411 ix_SSE0 = _mm_load1_pd(x_align+i);
2412 iy_SSE0 = _mm_load1_pd(y_align+i);
2413 iz_SSE0 = _mm_load1_pd(z_align+i);
2414 ix_SSE1 = _mm_load1_pd(x_align+i+1);
2415 iy_SSE1 = _mm_load1_pd(y_align+i+1);
2416 iz_SSE1 = _mm_load1_pd(z_align+i+1);
2418 fix_SSE0 = _mm_setzero_pd();
2419 fiy_SSE0 = _mm_setzero_pd();
2420 fiz_SSE0 = _mm_setzero_pd();
2421 fix_SSE1 = _mm_setzero_pd();
2422 fiy_SSE1 = _mm_setzero_pd();
2423 fiz_SSE1 = _mm_setzero_pd();
2425 rbai_SSE0 = _mm_load1_pd(rb+i);
2426 rbai_SSE1 = _mm_load1_pd(rb+i+1);
2428 /* Load limits for loop over neighbors */
2430 nj3 = jindex[4*i+3];
2432 /* No masks necessary, since the stored chain rule derivatives will be zero in those cases! */
2433 for(j=nj0; j<nj3; j+=UNROLLJ)
2435 /* load j atom coordinates */
2436 jx_SSE = _mm_load_pd(x_align+j);
2437 jy_SSE = _mm_load_pd(y_align+j);
2438 jz_SSE = _mm_load_pd(z_align+j);
2440 /* Calculate distance */
2441 dx_SSE0 = _mm_sub_pd(ix_SSE0,jx_SSE);
2442 dy_SSE0 = _mm_sub_pd(iy_SSE0,jy_SSE);
2443 dz_SSE0 = _mm_sub_pd(iz_SSE0,jz_SSE);
2444 dx_SSE1 = _mm_sub_pd(ix_SSE1,jx_SSE);
2445 dy_SSE1 = _mm_sub_pd(iy_SSE1,jy_SSE);
2446 dz_SSE1 = _mm_sub_pd(iz_SSE1,jz_SSE);
2448 rbaj_SSE = _mm_load_pd(rb+j);
2450 fgb_SSE0 = _mm_mul_pd(rbai_SSE0,_mm_load_pd(dadx));
2452 fgb_SSE1 = _mm_mul_pd(rbai_SSE1,_mm_load_pd(dadx));
2455 fgb_ai_SSE0 = _mm_mul_pd(rbaj_SSE,_mm_load_pd(dadx));
2457 fgb_ai_SSE1 = _mm_mul_pd(rbaj_SSE,_mm_load_pd(dadx));
2460 /* Total force between ai and aj is the sum of ai->aj and aj->ai */
2461 fgb_SSE0 = _mm_add_pd(fgb_SSE0,fgb_ai_SSE0);
2462 fgb_SSE1 = _mm_add_pd(fgb_SSE1,fgb_ai_SSE1);
2464 /* Calculate temporary vectorial force */
2465 tx_SSE0 = _mm_mul_pd(fgb_SSE0,dx_SSE0);
2466 ty_SSE0 = _mm_mul_pd(fgb_SSE0,dy_SSE0);
2467 tz_SSE0 = _mm_mul_pd(fgb_SSE0,dz_SSE0);
2468 tx_SSE1 = _mm_mul_pd(fgb_SSE1,dx_SSE1);
2469 ty_SSE1 = _mm_mul_pd(fgb_SSE1,dy_SSE1);
2470 tz_SSE1 = _mm_mul_pd(fgb_SSE1,dz_SSE1);
2472 /* Increment i atom force */
2473 fix_SSE0 = _mm_add_pd(fix_SSE0,tx_SSE0);
2474 fiy_SSE0 = _mm_add_pd(fiy_SSE0,ty_SSE0);
2475 fiz_SSE0 = _mm_add_pd(fiz_SSE0,tz_SSE0);
2476 fix_SSE1 = _mm_add_pd(fix_SSE1,tx_SSE1);
2477 fiy_SSE1 = _mm_add_pd(fiy_SSE1,ty_SSE1);
2478 fiz_SSE1 = _mm_add_pd(fiz_SSE1,tz_SSE1);
2480 /* Decrement j atom force */
2481 _mm_store_pd(fx_align+j,
2482 _mm_sub_pd( _mm_load_pd(fx_align+j) , _mm_add_pd(tx_SSE0,tx_SSE1) ));
2483 _mm_store_pd(fy_align+j,
2484 _mm_sub_pd( _mm_load_pd(fy_align+j) , _mm_add_pd(ty_SSE0,ty_SSE1) ));
2485 _mm_store_pd(fz_align+j,
2486 _mm_sub_pd( _mm_load_pd(fz_align+j) , _mm_add_pd(tz_SSE0,tz_SSE1) ));
2489 /* Add i forces to mem */
2490 GMX_MM_TRANSPOSE2_PD(fix_SSE0,fix_SSE1);
2491 fix_SSE0 = _mm_add_pd(fix_SSE0,fix_SSE1);
2492 _mm_store_pd(fx_align+i, _mm_add_pd(fix_SSE0, _mm_load_pd(fx_align+i)));
2494 GMX_MM_TRANSPOSE2_PD(fiy_SSE0,fiy_SSE1);
2495 fiy_SSE0 = _mm_add_pd(fiy_SSE0,fiy_SSE1);
2496 _mm_store_pd(fy_align+i, _mm_add_pd(fiy_SSE0, _mm_load_pd(fy_align+i)));
2498 GMX_MM_TRANSPOSE2_PD(fiz_SSE0,fiz_SSE1);
2499 fiz_SSE0 = _mm_add_pd(fiz_SSE0,fiz_SSE1);
2500 _mm_store_pd(fz_align+i, _mm_add_pd(fiz_SSE0, _mm_load_pd(fz_align+i)));
2503 for(i=0;i<natoms;i++)
2505 f[3*i] += fx_align[i] + fx_align[natoms+i];
2506 f[3*i+1] += fy_align[i] + fy_align[natoms+i];
2507 f[3*i+2] += fz_align[i] + fz_align[natoms+i];
2514 /* dummy variable when not using SSE */
2515 int genborn_allvsall_sse2_double_dummy;