2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2009, The GROMACS Development Team.
6 * Copyright (c) 2012,2014, by the GROMACS development team, led by
7 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
8 * and including many others, as listed in the AUTHORS file in the
9 * top-level source directory and at http://www.gromacs.org.
11 * GROMACS is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public License
13 * as published by the Free Software Foundation; either version 2.1
14 * of the License, or (at your option) any later version.
16 * GROMACS is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with GROMACS; if not, see
23 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
24 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
26 * If you want to redistribute modifications to GROMACS, please
27 * consider that scientific software is very special. Version
28 * control is crucial - bugs must be traceable. We will be happy to
29 * consider code for inclusion in the official distribution, but
30 * derived work must not be called official GROMACS. Details are found
31 * in the README & COPYING files - if they are missing, get the
32 * official version at http://www.gromacs.org.
34 * To help us fund GROMACS development, we humbly ask that you cite
35 * the research papers on the package. Check out http://www.gromacs.org.
40 #include "gromacs/legacyheaders/types/simple.h"
42 #include "gromacs/math/vec.h"
43 #include "gromacs/utility/smalloc.h"
45 #include "gromacs/legacyheaders/network.h"
46 #include "gromacs/math/units.h"
47 #include "gromacs/legacyheaders/genborn.h"
48 #include "genborn_allvsall.h"
50 #if 0 && defined (GMX_SIMD_X86_SSE2_OR_HIGHER)
52 #include <gmx_sse2_single.h>
70 int ** prologue_mask_gb;
83 gmx_allvsallgb2_data_t;
87 calc_maxoffset(int i, int natoms)
91 if ((natoms % 2) == 1)
93 /* Odd number of atoms, easy */
96 else if ((natoms % 4) == 0)
98 /* Multiple of four is hard */
103 maxoffset = natoms/2;
107 maxoffset = natoms/2-1;
114 maxoffset = natoms/2;
118 maxoffset = natoms/2-1;
127 maxoffset = natoms/2;
131 maxoffset = natoms/2-1;
139 setup_gb_exclusions_and_indices(gmx_allvsallgb2_data_t * aadata,
150 int ni0, ni1, nj0, nj1, nj;
151 int imin, imax, iexcl;
154 int firstinteraction;
158 /* This routine can appear to be a bit complex, but it is mostly book-keeping.
159 * To enable the fast all-vs-all kernel we need to be able to stream through all coordinates
160 * whether they should interact or not.
162 * To avoid looping over the exclusions, we create a simple mask that is 1 if the interaction
163 * should be present, otherwise 0. Since exclusions typically only occur when i & j are close,
164 * we create a jindex array with three elements per i atom: the starting point, the point to
165 * which we need to check exclusions, and the end point.
166 * This way we only have to allocate a short exclusion mask per i atom.
169 ni0 = (start/UNROLLI)*UNROLLI;
170 ni1 = ((end+UNROLLI-1)/UNROLLI)*UNROLLI;
172 /* Set the interaction mask to only enable the i atoms we want to include */
173 snew(pi, natoms+UNROLLI+2*SIMD_WIDTH);
174 aadata->imask = (int *) (((size_t) pi + 16) & (~((size_t) 15)));
175 for (i = 0; i < natoms+UNROLLI; i++)
177 aadata->imask[i] = (i >= start && i < end) ? 0xFFFFFFFF : 0;
180 /* Allocate memory for our modified jindex array */
181 snew(aadata->jindex_gb, 4*(natoms+UNROLLI));
182 for (i = 0; i < 4*(natoms+UNROLLI); i++)
184 aadata->jindex_gb[i] = 0;
187 /* Create the exclusion masks for the prologue part */
188 snew(aadata->prologue_mask_gb, natoms+UNROLLI); /* list of pointers */
190 /* First zero everything to avoid uninitialized data */
191 for (i = 0; i < natoms+UNROLLI; i++)
193 aadata->prologue_mask_gb[i] = NULL;
196 /* Calculate the largest exclusion range we need for each UNROLLI-tuplet of i atoms. */
197 for (ibase = ni0; ibase < ni1; ibase += UNROLLI)
199 max_excl_offset = -1;
201 /* First find maxoffset for the next 4 atoms (or fewer if we are close to end) */
202 imax = ((ibase+UNROLLI) < end) ? (ibase+UNROLLI) : end;
204 /* Which atom is the first we (might) interact with? */
205 imin = natoms; /* Guaranteed to be overwritten by one of 'firstinteraction' */
206 for (i = ibase; i < imax; i++)
208 /* Before exclusions, which atom is the first we (might) interact with? */
209 firstinteraction = i+1;
210 max_offset = calc_maxoffset(i, natoms);
214 for (j = 0; j < ilist[F_GB12].nr; j += 3)
216 a1 = ilist[F_GB12].iatoms[j+1];
217 a2 = ilist[F_GB12].iatoms[j+2];
232 if (k == firstinteraction)
240 for (j = 0; j < ilist[F_GB13].nr; j += 3)
242 a1 = ilist[F_GB13].iatoms[j+1];
243 a2 = ilist[F_GB13].iatoms[j+2];
258 if (k == firstinteraction)
266 for (j = 0; j < ilist[F_GB14].nr; j += 3)
268 a1 = ilist[F_GB14].iatoms[j+1];
269 a2 = ilist[F_GB14].iatoms[j+2];
283 if (k == firstinteraction)
289 imin = (firstinteraction < imin) ? firstinteraction : imin;
291 /* round down to j unrolling factor */
292 imin = (imin/UNROLLJ)*UNROLLJ;
294 for (i = ibase; i < imax; i++)
296 max_offset = calc_maxoffset(i, natoms);
300 for (j = 0; j < ilist[F_GB12].nr; j += 3)
302 a1 = ilist[F_GB12].iatoms[j+1];
303 a2 = ilist[F_GB12].iatoms[j+2];
323 if (k > i+max_offset)
330 if (k+natoms <= max_offset)
334 max_excl_offset = (k > max_excl_offset) ? k : max_excl_offset;
339 for (j = 0; j < ilist[F_GB13].nr; j += 3)
341 a1 = ilist[F_GB13].iatoms[j+1];
342 a2 = ilist[F_GB13].iatoms[j+2];
362 if (k > i+max_offset)
369 if (k+natoms <= max_offset)
373 max_excl_offset = (k > max_excl_offset) ? k : max_excl_offset;
378 for (j = 0; j < ilist[F_GB14].nr; j += 3)
380 a1 = ilist[F_GB14].iatoms[j+1];
381 a2 = ilist[F_GB14].iatoms[j+2];
401 if (k > i+max_offset)
408 if (k+natoms <= max_offset)
412 max_excl_offset = (k > max_excl_offset) ? k : max_excl_offset;
417 /* The offset specifies the last atom to be excluded, so add one unit to get an upper loop limit */
419 /* round up to j unrolling factor */
420 max_excl_offset = (max_excl_offset/UNROLLJ+1)*UNROLLJ;
422 /* Set all the prologue masks length to this value (even for i>end) */
423 for (i = ibase; i < ibase+UNROLLI; i++)
425 aadata->jindex_gb[4*i] = imin;
426 aadata->jindex_gb[4*i+1] = imin+max_excl_offset;
430 /* Now the hard part, loop over it all again to calculate the actual contents of the prologue masks */
431 for (ibase = ni0; ibase < ni1; ibase += UNROLLI)
433 for (i = ibase; i < ibase+UNROLLI; i++)
435 nj = aadata->jindex_gb[4*i+1] - aadata->jindex_gb[4*i];
436 imin = aadata->jindex_gb[4*i];
438 /* Allocate aligned memory */
439 snew(pi, nj+2*SIMD_WIDTH);
440 aadata->prologue_mask_gb[i] = (int *) (((size_t) pi + 16) & (~((size_t) 15)));
442 max_offset = calc_maxoffset(i, natoms);
444 /* Include interactions i+1 <= j < i+maxoffset */
445 for (k = 0; k < nj; k++)
449 if ( (j > i) && (j <= i+max_offset) )
451 aadata->prologue_mask_gb[i][k] = 0xFFFFFFFF;
455 aadata->prologue_mask_gb[i][k] = 0;
459 /* Clear out the explicit exclusions */
464 for (j = 0; j < ilist[F_GB12].nr; j += 3)
466 a1 = ilist[F_GB12].iatoms[j+1];
467 a2 = ilist[F_GB12].iatoms[j+2];
482 if (k > i+max_offset)
488 if (k+natoms <= max_offset)
496 aadata->prologue_mask_gb[i][k] = 0;
502 for (j = 0; j < ilist[F_GB13].nr; j += 3)
504 a1 = ilist[F_GB13].iatoms[j+1];
505 a2 = ilist[F_GB13].iatoms[j+2];
520 if (k > i+max_offset)
526 if (k+natoms <= max_offset)
534 aadata->prologue_mask_gb[i][k] = 0;
540 for (j = 0; j < ilist[F_GB14].nr; j += 3)
542 a1 = ilist[F_GB14].iatoms[j+1];
543 a2 = ilist[F_GB14].iatoms[j+2];
558 if (k > i+max_offset)
564 if (k+natoms <= max_offset)
572 aadata->prologue_mask_gb[i][k] = 0;
580 /* Construct the epilogue mask - this just contains the check for maxoffset */
581 snew(aadata->epilogue_mask, natoms+UNROLLI);
583 /* First zero everything to avoid uninitialized data */
584 for (i = 0; i < natoms+UNROLLI; i++)
586 aadata->jindex_gb[4*i+2] = aadata->jindex_gb[4*i+1];
587 aadata->jindex_gb[4*i+3] = aadata->jindex_gb[4*i+1];
588 aadata->epilogue_mask[i] = NULL;
591 for (ibase = ni0; ibase < ni1; ibase += UNROLLI)
593 /* Find the lowest index for which we need to use the epilogue */
595 max_offset = calc_maxoffset(imin, natoms);
597 imin = imin + 1 + max_offset;
599 /* Find largest index for which we need to use the epilogue */
600 imax = ibase + UNROLLI-1;
601 imax = (imax < end) ? imax : end;
603 max_offset = calc_maxoffset(imax, natoms);
604 imax = imax + 1 + max_offset + UNROLLJ - 1;
606 for (i = ibase; i < ibase+UNROLLI; i++)
608 /* Start of epilogue - round down to j tile limit */
609 aadata->jindex_gb[4*i+2] = (imin/UNROLLJ)*UNROLLJ;
610 /* Make sure we dont overlap - for small systems everything is done in the prologue */
611 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];
612 /* Round upwards to j tile limit */
613 aadata->jindex_gb[4*i+3] = (imax/UNROLLJ)*UNROLLJ;
614 /* Make sure we dont have a negative range for the epilogue */
615 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];
619 /* And fill it with data... */
621 for (ibase = ni0; ibase < ni1; ibase += UNROLLI)
623 for (i = ibase; i < ibase+UNROLLI; i++)
626 nj = aadata->jindex_gb[4*i+3] - aadata->jindex_gb[4*i+2];
628 /* Allocate aligned memory */
629 snew(pi, nj+2*SIMD_WIDTH);
630 aadata->epilogue_mask[i] = (int *) (((size_t) pi + 16) & (~((size_t) 15)));
632 max_offset = calc_maxoffset(i, natoms);
634 for (k = 0; k < nj; k++)
636 j = aadata->jindex_gb[4*i+2] + k;
637 aadata->epilogue_mask[i][k] = (j <= i+max_offset) ? 0xFFFFFFFF : 0;
645 genborn_allvsall_setup(gmx_allvsallgb2_data_t ** p_aadata,
646 gmx_localtop_t * top,
647 gmx_genborn_t * born,
657 gmx_allvsallgb2_data_t *aadata;
660 natoms = mdatoms->nr;
665 snew(p, 2*natoms+2*SIMD_WIDTH);
666 aadata->x_align = (real *) (((size_t) p + 16) & (~((size_t) 15)));
667 snew(p, 2*natoms+2*SIMD_WIDTH);
668 aadata->y_align = (real *) (((size_t) p + 16) & (~((size_t) 15)));
669 snew(p, 2*natoms+2*SIMD_WIDTH);
670 aadata->z_align = (real *) (((size_t) p + 16) & (~((size_t) 15)));
671 snew(p, 2*natoms+2*SIMD_WIDTH);
672 aadata->fx_align = (real *) (((size_t) p + 16) & (~((size_t) 15)));
673 snew(p, 2*natoms+2*SIMD_WIDTH);
674 aadata->fy_align = (real *) (((size_t) p + 16) & (~((size_t) 15)));
675 snew(p, 2*natoms+2*SIMD_WIDTH);
676 aadata->fz_align = (real *) (((size_t) p + 16) & (~((size_t) 15)));
678 snew(p, 2*natoms+UNROLLJ+SIMD_WIDTH);
679 aadata->gb_radius = (real *) (((size_t) p + 16) & (~((size_t) 15)));
681 snew(p, 2*natoms+UNROLLJ+SIMD_WIDTH);
682 aadata->workparam = (real *) (((size_t) p + 16) & (~((size_t) 15)));
684 snew(p, 2*natoms+UNROLLJ+SIMD_WIDTH);
685 aadata->work = (real *) (((size_t) p + 16) & (~((size_t) 15)));
687 for (i = 0; i < mdatoms->nr; i++)
689 aadata->gb_radius[i] = top->atomtypes.gb_radius[mdatoms->typeA[i]] - radius_offset;
690 if (gb_algorithm == egbSTILL)
692 aadata->workparam[i] = born->vsolv[i];
694 else if (gb_algorithm == egbOBC)
696 aadata->workparam[i] = born->param[i];
698 aadata->work[i] = 0.0;
700 for (i = 0; i < mdatoms->nr; i++)
702 aadata->gb_radius[natoms+i] = aadata->gb_radius[i];
703 aadata->workparam[natoms+i] = aadata->workparam[i];
704 aadata->work[natoms+i] = aadata->work[i];
707 for (i = 0; i < 2*natoms+SIMD_WIDTH; i++)
709 aadata->x_align[i] = 0.0;
710 aadata->y_align[i] = 0.0;
711 aadata->z_align[i] = 0.0;
712 aadata->fx_align[i] = 0.0;
713 aadata->fy_align[i] = 0.0;
714 aadata->fz_align[i] = 0.0;
717 setup_gb_exclusions_and_indices(aadata, top->idef.il, 0, mdatoms->homenr, mdatoms->nr,
718 bInclude12, bInclude13, bInclude14);
723 genborn_allvsall_calc_still_radii_sse2_single(t_forcerec * fr,
725 gmx_genborn_t * born,
726 gmx_localtop_t * top,
731 gmx_allvsallgb2_data_t *aadata;
734 int nj0, nj1, nj2, nj3;
751 real irsq, idr4, idr6;
752 real raj, rvdw, ratio;
753 real vaj, ccf, dccf, theta, cosq;
754 real term, prod, icf4, icf6, gpi2, factor, sinq;
765 __m128 ix_SSE0, iy_SSE0, iz_SSE0;
766 __m128 ix_SSE1, iy_SSE1, iz_SSE1;
767 __m128 ix_SSE2, iy_SSE2, iz_SSE2;
768 __m128 ix_SSE3, iy_SSE3, iz_SSE3;
769 __m128 gpi_SSE0, rai_SSE0, prod_ai_SSE0;
770 __m128 gpi_SSE1, rai_SSE1, prod_ai_SSE1;
771 __m128 gpi_SSE2, rai_SSE2, prod_ai_SSE2;
772 __m128 gpi_SSE3, rai_SSE3, prod_ai_SSE3;
773 __m128 imask_SSE0, jmask_SSE0;
774 __m128 imask_SSE1, jmask_SSE1;
775 __m128 imask_SSE2, jmask_SSE2;
776 __m128 imask_SSE3, jmask_SSE3;
777 __m128 jx_SSE, jy_SSE, jz_SSE;
778 __m128 dx_SSE0, dy_SSE0, dz_SSE0;
779 __m128 dx_SSE1, dy_SSE1, dz_SSE1;
780 __m128 dx_SSE2, dy_SSE2, dz_SSE2;
781 __m128 dx_SSE3, dy_SSE3, dz_SSE3;
782 __m128 rsq_SSE0, rinv_SSE0, irsq_SSE0, idr4_SSE0, idr6_SSE0;
783 __m128 rsq_SSE1, rinv_SSE1, irsq_SSE1, idr4_SSE1, idr6_SSE1;
784 __m128 rsq_SSE2, rinv_SSE2, irsq_SSE2, idr4_SSE2, idr6_SSE2;
785 __m128 rsq_SSE3, rinv_SSE3, irsq_SSE3, idr4_SSE3, idr6_SSE3;
786 __m128 raj_SSE, vaj_SSE, prod_SSE;
787 __m128 rvdw_SSE0, ratio_SSE0;
788 __m128 rvdw_SSE1, ratio_SSE1;
789 __m128 rvdw_SSE2, ratio_SSE2;
790 __m128 rvdw_SSE3, ratio_SSE3;
791 __m128 theta_SSE0, sinq_SSE0, cosq_SSE0, term_SSE0;
792 __m128 theta_SSE1, sinq_SSE1, cosq_SSE1, term_SSE1;
793 __m128 theta_SSE2, sinq_SSE2, cosq_SSE2, term_SSE2;
794 __m128 theta_SSE3, sinq_SSE3, cosq_SSE3, term_SSE3;
795 __m128 ccf_SSE0, dccf_SSE0;
796 __m128 ccf_SSE1, dccf_SSE1;
797 __m128 ccf_SSE2, dccf_SSE2;
798 __m128 ccf_SSE3, dccf_SSE3;
799 __m128 icf4_SSE0, icf6_SSE0;
800 __m128 icf4_SSE1, icf6_SSE1;
801 __m128 icf4_SSE2, icf6_SSE2;
802 __m128 icf4_SSE3, icf6_SSE3;
803 __m128 half_SSE, one_SSE, two_SSE, four_SSE;
804 __m128 still_p4_SSE, still_p5inv_SSE, still_pip5_SSE;
806 natoms = mdatoms->nr;
808 ni1 = mdatoms->homenr;
812 aadata = *((gmx_allvsallgb2_data_t **)paadata);
817 genborn_allvsall_setup(&aadata, top, born, mdatoms, 0.0,
818 egbSTILL, FALSE, FALSE, TRUE);
819 *((gmx_allvsallgb2_data_t **)paadata) = aadata;
822 x_align = aadata->x_align;
823 y_align = aadata->y_align;
824 z_align = aadata->z_align;
826 gb_radius = aadata->gb_radius;
827 vsolv = aadata->workparam;
829 jindex = aadata->jindex_gb;
832 still_p4_SSE = _mm_set1_ps(STILL_P4);
833 still_p5inv_SSE = _mm_set1_ps(STILL_P5INV);
834 still_pip5_SSE = _mm_set1_ps(STILL_PIP5);
835 half_SSE = _mm_set1_ps(0.5);
836 one_SSE = _mm_set1_ps(1.0);
837 two_SSE = _mm_set1_ps(2.0);
838 four_SSE = _mm_set1_ps(4.0);
840 /* This will be summed, so it has to extend to natoms + buffer */
841 for (i = 0; i < natoms+1+natoms/2; i++)
846 for (i = ni0; i < ni1+1+natoms/2; i++)
850 y_align[i] = x[3*k+1];
851 z_align[i] = x[3*k+2];
856 for (i = ni0; i < ni1; i += UNROLLI)
858 /* We assume shifts are NOT used for all-vs-all interactions */
860 /* Load i atom data */
861 ix_SSE0 = _mm_load1_ps(x_align+i);
862 iy_SSE0 = _mm_load1_ps(y_align+i);
863 iz_SSE0 = _mm_load1_ps(z_align+i);
864 ix_SSE1 = _mm_load1_ps(x_align+i+1);
865 iy_SSE1 = _mm_load1_ps(y_align+i+1);
866 iz_SSE1 = _mm_load1_ps(z_align+i+1);
867 ix_SSE2 = _mm_load1_ps(x_align+i+2);
868 iy_SSE2 = _mm_load1_ps(y_align+i+2);
869 iz_SSE2 = _mm_load1_ps(z_align+i+2);
870 ix_SSE3 = _mm_load1_ps(x_align+i+3);
871 iy_SSE3 = _mm_load1_ps(y_align+i+3);
872 iz_SSE3 = _mm_load1_ps(z_align+i+3);
874 gpi_SSE0 = _mm_setzero_ps();
875 gpi_SSE1 = _mm_setzero_ps();
876 gpi_SSE2 = _mm_setzero_ps();
877 gpi_SSE3 = _mm_setzero_ps();
879 rai_SSE0 = _mm_load1_ps(gb_radius+i);
880 rai_SSE1 = _mm_load1_ps(gb_radius+i+1);
881 rai_SSE2 = _mm_load1_ps(gb_radius+i+2);
882 rai_SSE3 = _mm_load1_ps(gb_radius+i+3);
884 prod_ai_SSE0 = _mm_set1_ps(STILL_P4*vsolv[i]);
885 prod_ai_SSE1 = _mm_set1_ps(STILL_P4*vsolv[i+1]);
886 prod_ai_SSE2 = _mm_set1_ps(STILL_P4*vsolv[i+2]);
887 prod_ai_SSE3 = _mm_set1_ps(STILL_P4*vsolv[i+3]);
889 /* Load limits for loop over neighbors */
895 pmask0 = aadata->prologue_mask_gb[i];
896 pmask1 = aadata->prologue_mask_gb[i+1];
897 pmask2 = aadata->prologue_mask_gb[i+2];
898 pmask3 = aadata->prologue_mask_gb[i+3];
899 emask0 = aadata->epilogue_mask[i];
900 emask1 = aadata->epilogue_mask[i+1];
901 emask2 = aadata->epilogue_mask[i+2];
902 emask3 = aadata->epilogue_mask[i+3];
904 imask_SSE0 = _mm_load1_ps((real *)(aadata->imask+i));
905 imask_SSE1 = _mm_load1_ps((real *)(aadata->imask+i+1));
906 imask_SSE2 = _mm_load1_ps((real *)(aadata->imask+i+2));
907 imask_SSE3 = _mm_load1_ps((real *)(aadata->imask+i+3));
909 /* Prologue part, including exclusion mask */
910 for (j = nj0; j < nj1; j += UNROLLJ)
912 jmask_SSE0 = _mm_load_ps((real *)pmask0);
913 jmask_SSE1 = _mm_load_ps((real *)pmask1);
914 jmask_SSE2 = _mm_load_ps((real *)pmask2);
915 jmask_SSE3 = _mm_load_ps((real *)pmask3);
921 /* load j atom coordinates */
922 jx_SSE = _mm_load_ps(x_align+j);
923 jy_SSE = _mm_load_ps(y_align+j);
924 jz_SSE = _mm_load_ps(z_align+j);
926 /* Calculate distance */
927 dx_SSE0 = _mm_sub_ps(ix_SSE0, jx_SSE);
928 dy_SSE0 = _mm_sub_ps(iy_SSE0, jy_SSE);
929 dz_SSE0 = _mm_sub_ps(iz_SSE0, jz_SSE);
930 dx_SSE1 = _mm_sub_ps(ix_SSE1, jx_SSE);
931 dy_SSE1 = _mm_sub_ps(iy_SSE1, jy_SSE);
932 dz_SSE1 = _mm_sub_ps(iz_SSE1, jz_SSE);
933 dx_SSE2 = _mm_sub_ps(ix_SSE2, jx_SSE);
934 dy_SSE2 = _mm_sub_ps(iy_SSE2, jy_SSE);
935 dz_SSE2 = _mm_sub_ps(iz_SSE2, jz_SSE);
936 dx_SSE3 = _mm_sub_ps(ix_SSE3, jx_SSE);
937 dy_SSE3 = _mm_sub_ps(iy_SSE3, jy_SSE);
938 dz_SSE3 = _mm_sub_ps(iz_SSE3, jz_SSE);
940 /* rsq = dx*dx+dy*dy+dz*dz */
941 rsq_SSE0 = gmx_mm_calc_rsq_ps(dx_SSE0, dy_SSE0, dz_SSE0);
942 rsq_SSE1 = gmx_mm_calc_rsq_ps(dx_SSE1, dy_SSE1, dz_SSE1);
943 rsq_SSE2 = gmx_mm_calc_rsq_ps(dx_SSE2, dy_SSE2, dz_SSE2);
944 rsq_SSE3 = gmx_mm_calc_rsq_ps(dx_SSE3, dy_SSE3, dz_SSE3);
947 jmask_SSE0 = _mm_and_ps(jmask_SSE0, imask_SSE0);
948 jmask_SSE1 = _mm_and_ps(jmask_SSE1, imask_SSE1);
949 jmask_SSE2 = _mm_and_ps(jmask_SSE2, imask_SSE2);
950 jmask_SSE3 = _mm_and_ps(jmask_SSE3, imask_SSE3);
952 /* Calculate 1/r and 1/r2 */
953 rinv_SSE0 = gmx_mm_invsqrt_ps(rsq_SSE0);
954 rinv_SSE1 = gmx_mm_invsqrt_ps(rsq_SSE1);
955 rinv_SSE2 = gmx_mm_invsqrt_ps(rsq_SSE2);
956 rinv_SSE3 = gmx_mm_invsqrt_ps(rsq_SSE3);
959 rinv_SSE0 = _mm_and_ps(rinv_SSE0, jmask_SSE0);
960 rinv_SSE1 = _mm_and_ps(rinv_SSE1, jmask_SSE1);
961 rinv_SSE2 = _mm_and_ps(rinv_SSE2, jmask_SSE2);
962 rinv_SSE3 = _mm_and_ps(rinv_SSE3, jmask_SSE3);
964 irsq_SSE0 = _mm_mul_ps(rinv_SSE0, rinv_SSE0);
965 irsq_SSE1 = _mm_mul_ps(rinv_SSE1, rinv_SSE1);
966 irsq_SSE2 = _mm_mul_ps(rinv_SSE2, rinv_SSE2);
967 irsq_SSE3 = _mm_mul_ps(rinv_SSE3, rinv_SSE3);
968 idr4_SSE0 = _mm_mul_ps(irsq_SSE0, irsq_SSE0);
969 idr4_SSE1 = _mm_mul_ps(irsq_SSE1, irsq_SSE1);
970 idr4_SSE2 = _mm_mul_ps(irsq_SSE2, irsq_SSE2);
971 idr4_SSE3 = _mm_mul_ps(irsq_SSE3, irsq_SSE3);
972 idr6_SSE0 = _mm_mul_ps(idr4_SSE0, irsq_SSE0);
973 idr6_SSE1 = _mm_mul_ps(idr4_SSE1, irsq_SSE1);
974 idr6_SSE2 = _mm_mul_ps(idr4_SSE2, irsq_SSE2);
975 idr6_SSE3 = _mm_mul_ps(idr4_SSE3, irsq_SSE3);
977 raj_SSE = _mm_load_ps(gb_radius+j);
978 vaj_SSE = _mm_load_ps(vsolv+j);
980 rvdw_SSE0 = _mm_add_ps(rai_SSE0, raj_SSE);
981 rvdw_SSE1 = _mm_add_ps(rai_SSE1, raj_SSE);
982 rvdw_SSE2 = _mm_add_ps(rai_SSE2, raj_SSE);
983 rvdw_SSE3 = _mm_add_ps(rai_SSE3, raj_SSE);
985 ratio_SSE0 = _mm_mul_ps(rsq_SSE0, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE0, rvdw_SSE0)));
986 ratio_SSE1 = _mm_mul_ps(rsq_SSE1, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE1, rvdw_SSE1)));
987 ratio_SSE2 = _mm_mul_ps(rsq_SSE2, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE2, rvdw_SSE2)));
988 ratio_SSE3 = _mm_mul_ps(rsq_SSE3, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE3, rvdw_SSE3)));
990 ratio_SSE0 = _mm_min_ps(ratio_SSE0, still_p5inv_SSE);
991 ratio_SSE1 = _mm_min_ps(ratio_SSE1, still_p5inv_SSE);
992 ratio_SSE2 = _mm_min_ps(ratio_SSE2, still_p5inv_SSE);
993 ratio_SSE3 = _mm_min_ps(ratio_SSE3, still_p5inv_SSE);
994 theta_SSE0 = _mm_mul_ps(ratio_SSE0, still_pip5_SSE);
995 theta_SSE1 = _mm_mul_ps(ratio_SSE1, still_pip5_SSE);
996 theta_SSE2 = _mm_mul_ps(ratio_SSE2, still_pip5_SSE);
997 theta_SSE3 = _mm_mul_ps(ratio_SSE3, still_pip5_SSE);
998 gmx_mm_sincos_ps(theta_SSE0, &sinq_SSE0, &cosq_SSE0);
999 gmx_mm_sincos_ps(theta_SSE1, &sinq_SSE1, &cosq_SSE1);
1000 gmx_mm_sincos_ps(theta_SSE2, &sinq_SSE2, &cosq_SSE2);
1001 gmx_mm_sincos_ps(theta_SSE3, &sinq_SSE3, &cosq_SSE3);
1002 term_SSE0 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE0));
1003 term_SSE1 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE1));
1004 term_SSE2 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE2));
1005 term_SSE3 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE3));
1006 ccf_SSE0 = _mm_mul_ps(term_SSE0, term_SSE0);
1007 ccf_SSE1 = _mm_mul_ps(term_SSE1, term_SSE1);
1008 ccf_SSE2 = _mm_mul_ps(term_SSE2, term_SSE2);
1009 ccf_SSE3 = _mm_mul_ps(term_SSE3, term_SSE3);
1010 dccf_SSE0 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE0),
1011 _mm_mul_ps(sinq_SSE0, theta_SSE0));
1012 dccf_SSE1 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE1),
1013 _mm_mul_ps(sinq_SSE1, theta_SSE1));
1014 dccf_SSE2 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE2),
1015 _mm_mul_ps(sinq_SSE2, theta_SSE2));
1016 dccf_SSE3 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE3),
1017 _mm_mul_ps(sinq_SSE3, theta_SSE3));
1019 prod_SSE = _mm_mul_ps(still_p4_SSE, vaj_SSE);
1020 icf4_SSE0 = _mm_mul_ps(ccf_SSE0, idr4_SSE0);
1021 icf4_SSE1 = _mm_mul_ps(ccf_SSE1, idr4_SSE1);
1022 icf4_SSE2 = _mm_mul_ps(ccf_SSE2, idr4_SSE2);
1023 icf4_SSE3 = _mm_mul_ps(ccf_SSE3, idr4_SSE3);
1024 icf6_SSE0 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE0), dccf_SSE0), idr6_SSE0);
1025 icf6_SSE1 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE1), dccf_SSE1), idr6_SSE1);
1026 icf6_SSE2 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE2), dccf_SSE2), idr6_SSE2);
1027 icf6_SSE3 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE3), dccf_SSE3), idr6_SSE3);
1029 _mm_store_ps(work+j, _mm_add_ps(_mm_load_ps(work+j),
1030 gmx_mm_sum4_ps(_mm_mul_ps(prod_ai_SSE0, icf4_SSE0),
1031 _mm_mul_ps(prod_ai_SSE1, icf4_SSE1),
1032 _mm_mul_ps(prod_ai_SSE2, icf4_SSE2),
1033 _mm_mul_ps(prod_ai_SSE3, icf4_SSE3))));
1035 gpi_SSE0 = _mm_add_ps(gpi_SSE0, _mm_mul_ps(prod_SSE, icf4_SSE0));
1036 gpi_SSE1 = _mm_add_ps(gpi_SSE1, _mm_mul_ps(prod_SSE, icf4_SSE1));
1037 gpi_SSE2 = _mm_add_ps(gpi_SSE2, _mm_mul_ps(prod_SSE, icf4_SSE2));
1038 gpi_SSE3 = _mm_add_ps(gpi_SSE3, _mm_mul_ps(prod_SSE, icf4_SSE3));
1040 /* Save ai->aj and aj->ai chain rule terms */
1041 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE0));
1043 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE1));
1045 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE2));
1047 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE3));
1050 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE0, icf6_SSE0));
1052 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE1, icf6_SSE1));
1054 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE2, icf6_SSE2));
1056 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE3, icf6_SSE3));
1060 /* Main part, no exclusions */
1061 for (j = nj1; j < nj2; j += UNROLLJ)
1063 /* load j atom coordinates */
1064 jx_SSE = _mm_load_ps(x_align+j);
1065 jy_SSE = _mm_load_ps(y_align+j);
1066 jz_SSE = _mm_load_ps(z_align+j);
1068 /* Calculate distance */
1069 dx_SSE0 = _mm_sub_ps(ix_SSE0, jx_SSE);
1070 dy_SSE0 = _mm_sub_ps(iy_SSE0, jy_SSE);
1071 dz_SSE0 = _mm_sub_ps(iz_SSE0, jz_SSE);
1072 dx_SSE1 = _mm_sub_ps(ix_SSE1, jx_SSE);
1073 dy_SSE1 = _mm_sub_ps(iy_SSE1, jy_SSE);
1074 dz_SSE1 = _mm_sub_ps(iz_SSE1, jz_SSE);
1075 dx_SSE2 = _mm_sub_ps(ix_SSE2, jx_SSE);
1076 dy_SSE2 = _mm_sub_ps(iy_SSE2, jy_SSE);
1077 dz_SSE2 = _mm_sub_ps(iz_SSE2, jz_SSE);
1078 dx_SSE3 = _mm_sub_ps(ix_SSE3, jx_SSE);
1079 dy_SSE3 = _mm_sub_ps(iy_SSE3, jy_SSE);
1080 dz_SSE3 = _mm_sub_ps(iz_SSE3, jz_SSE);
1082 /* rsq = dx*dx+dy*dy+dz*dz */
1083 rsq_SSE0 = gmx_mm_calc_rsq_ps(dx_SSE0, dy_SSE0, dz_SSE0);
1084 rsq_SSE1 = gmx_mm_calc_rsq_ps(dx_SSE1, dy_SSE1, dz_SSE1);
1085 rsq_SSE2 = gmx_mm_calc_rsq_ps(dx_SSE2, dy_SSE2, dz_SSE2);
1086 rsq_SSE3 = gmx_mm_calc_rsq_ps(dx_SSE3, dy_SSE3, dz_SSE3);
1088 /* Calculate 1/r and 1/r2 */
1089 rinv_SSE0 = gmx_mm_invsqrt_ps(rsq_SSE0);
1090 rinv_SSE1 = gmx_mm_invsqrt_ps(rsq_SSE1);
1091 rinv_SSE2 = gmx_mm_invsqrt_ps(rsq_SSE2);
1092 rinv_SSE3 = gmx_mm_invsqrt_ps(rsq_SSE3);
1095 rinv_SSE0 = _mm_and_ps(rinv_SSE0, imask_SSE0);
1096 rinv_SSE1 = _mm_and_ps(rinv_SSE1, imask_SSE1);
1097 rinv_SSE2 = _mm_and_ps(rinv_SSE2, imask_SSE2);
1098 rinv_SSE3 = _mm_and_ps(rinv_SSE3, imask_SSE3);
1100 irsq_SSE0 = _mm_mul_ps(rinv_SSE0, rinv_SSE0);
1101 irsq_SSE1 = _mm_mul_ps(rinv_SSE1, rinv_SSE1);
1102 irsq_SSE2 = _mm_mul_ps(rinv_SSE2, rinv_SSE2);
1103 irsq_SSE3 = _mm_mul_ps(rinv_SSE3, rinv_SSE3);
1104 idr4_SSE0 = _mm_mul_ps(irsq_SSE0, irsq_SSE0);
1105 idr4_SSE1 = _mm_mul_ps(irsq_SSE1, irsq_SSE1);
1106 idr4_SSE2 = _mm_mul_ps(irsq_SSE2, irsq_SSE2);
1107 idr4_SSE3 = _mm_mul_ps(irsq_SSE3, irsq_SSE3);
1108 idr6_SSE0 = _mm_mul_ps(idr4_SSE0, irsq_SSE0);
1109 idr6_SSE1 = _mm_mul_ps(idr4_SSE1, irsq_SSE1);
1110 idr6_SSE2 = _mm_mul_ps(idr4_SSE2, irsq_SSE2);
1111 idr6_SSE3 = _mm_mul_ps(idr4_SSE3, irsq_SSE3);
1113 raj_SSE = _mm_load_ps(gb_radius+j);
1115 rvdw_SSE0 = _mm_add_ps(rai_SSE0, raj_SSE);
1116 rvdw_SSE1 = _mm_add_ps(rai_SSE1, raj_SSE);
1117 rvdw_SSE2 = _mm_add_ps(rai_SSE2, raj_SSE);
1118 rvdw_SSE3 = _mm_add_ps(rai_SSE3, raj_SSE);
1119 vaj_SSE = _mm_load_ps(vsolv+j);
1121 ratio_SSE0 = _mm_mul_ps(rsq_SSE0, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE0, rvdw_SSE0)));
1122 ratio_SSE1 = _mm_mul_ps(rsq_SSE1, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE1, rvdw_SSE1)));
1123 ratio_SSE2 = _mm_mul_ps(rsq_SSE2, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE2, rvdw_SSE2)));
1124 ratio_SSE3 = _mm_mul_ps(rsq_SSE3, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE3, rvdw_SSE3)));
1126 ratio_SSE0 = _mm_min_ps(ratio_SSE0, still_p5inv_SSE);
1127 ratio_SSE1 = _mm_min_ps(ratio_SSE1, still_p5inv_SSE);
1128 ratio_SSE2 = _mm_min_ps(ratio_SSE2, still_p5inv_SSE);
1129 ratio_SSE3 = _mm_min_ps(ratio_SSE3, still_p5inv_SSE);
1130 theta_SSE0 = _mm_mul_ps(ratio_SSE0, still_pip5_SSE);
1131 theta_SSE1 = _mm_mul_ps(ratio_SSE1, still_pip5_SSE);
1132 theta_SSE2 = _mm_mul_ps(ratio_SSE2, still_pip5_SSE);
1133 theta_SSE3 = _mm_mul_ps(ratio_SSE3, still_pip5_SSE);
1134 gmx_mm_sincos_ps(theta_SSE0, &sinq_SSE0, &cosq_SSE0);
1135 gmx_mm_sincos_ps(theta_SSE1, &sinq_SSE1, &cosq_SSE1);
1136 gmx_mm_sincos_ps(theta_SSE2, &sinq_SSE2, &cosq_SSE2);
1137 gmx_mm_sincos_ps(theta_SSE3, &sinq_SSE3, &cosq_SSE3);
1138 term_SSE0 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE0));
1139 term_SSE1 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE1));
1140 term_SSE2 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE2));
1141 term_SSE3 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE3));
1142 ccf_SSE0 = _mm_mul_ps(term_SSE0, term_SSE0);
1143 ccf_SSE1 = _mm_mul_ps(term_SSE1, term_SSE1);
1144 ccf_SSE2 = _mm_mul_ps(term_SSE2, term_SSE2);
1145 ccf_SSE3 = _mm_mul_ps(term_SSE3, term_SSE3);
1146 dccf_SSE0 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE0),
1147 _mm_mul_ps(sinq_SSE0, theta_SSE0));
1148 dccf_SSE1 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE1),
1149 _mm_mul_ps(sinq_SSE1, theta_SSE1));
1150 dccf_SSE2 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE2),
1151 _mm_mul_ps(sinq_SSE2, theta_SSE2));
1152 dccf_SSE3 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE3),
1153 _mm_mul_ps(sinq_SSE3, theta_SSE3));
1155 prod_SSE = _mm_mul_ps(still_p4_SSE, vaj_SSE );
1156 icf4_SSE0 = _mm_mul_ps(ccf_SSE0, idr4_SSE0);
1157 icf4_SSE1 = _mm_mul_ps(ccf_SSE1, idr4_SSE1);
1158 icf4_SSE2 = _mm_mul_ps(ccf_SSE2, idr4_SSE2);
1159 icf4_SSE3 = _mm_mul_ps(ccf_SSE3, idr4_SSE3);
1160 icf6_SSE0 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE0), dccf_SSE0), idr6_SSE0);
1161 icf6_SSE1 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE1), dccf_SSE1), idr6_SSE1);
1162 icf6_SSE2 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE2), dccf_SSE2), idr6_SSE2);
1163 icf6_SSE3 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE3), dccf_SSE3), idr6_SSE3);
1165 _mm_store_ps(work+j, _mm_add_ps(_mm_load_ps(work+j),
1166 gmx_mm_sum4_ps(_mm_mul_ps(prod_ai_SSE0, icf4_SSE0),
1167 _mm_mul_ps(prod_ai_SSE1, icf4_SSE1),
1168 _mm_mul_ps(prod_ai_SSE2, icf4_SSE2),
1169 _mm_mul_ps(prod_ai_SSE3, icf4_SSE3))));
1171 gpi_SSE0 = _mm_add_ps(gpi_SSE0, _mm_mul_ps(prod_SSE, icf4_SSE0));
1172 gpi_SSE1 = _mm_add_ps(gpi_SSE1, _mm_mul_ps(prod_SSE, icf4_SSE1));
1173 gpi_SSE2 = _mm_add_ps(gpi_SSE2, _mm_mul_ps(prod_SSE, icf4_SSE2));
1174 gpi_SSE3 = _mm_add_ps(gpi_SSE3, _mm_mul_ps(prod_SSE, icf4_SSE3));
1176 /* Save ai->aj and aj->ai chain rule terms */
1177 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE0));
1179 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE1));
1181 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE2));
1183 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE3));
1186 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE0, icf6_SSE0));
1188 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE1, icf6_SSE1));
1190 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE2, icf6_SSE2));
1192 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE3, icf6_SSE3));
1195 /* Epilogue part, including exclusion mask */
1196 for (j = nj2; j < nj3; j += UNROLLJ)
1198 jmask_SSE0 = _mm_load_ps((real *)emask0);
1199 jmask_SSE1 = _mm_load_ps((real *)emask1);
1200 jmask_SSE2 = _mm_load_ps((real *)emask2);
1201 jmask_SSE3 = _mm_load_ps((real *)emask3);
1207 /* load j atom coordinates */
1208 jx_SSE = _mm_load_ps(x_align+j);
1209 jy_SSE = _mm_load_ps(y_align+j);
1210 jz_SSE = _mm_load_ps(z_align+j);
1212 /* Calculate distance */
1213 dx_SSE0 = _mm_sub_ps(ix_SSE0, jx_SSE);
1214 dy_SSE0 = _mm_sub_ps(iy_SSE0, jy_SSE);
1215 dz_SSE0 = _mm_sub_ps(iz_SSE0, jz_SSE);
1216 dx_SSE1 = _mm_sub_ps(ix_SSE1, jx_SSE);
1217 dy_SSE1 = _mm_sub_ps(iy_SSE1, jy_SSE);
1218 dz_SSE1 = _mm_sub_ps(iz_SSE1, jz_SSE);
1219 dx_SSE2 = _mm_sub_ps(ix_SSE2, jx_SSE);
1220 dy_SSE2 = _mm_sub_ps(iy_SSE2, jy_SSE);
1221 dz_SSE2 = _mm_sub_ps(iz_SSE2, jz_SSE);
1222 dx_SSE3 = _mm_sub_ps(ix_SSE3, jx_SSE);
1223 dy_SSE3 = _mm_sub_ps(iy_SSE3, jy_SSE);
1224 dz_SSE3 = _mm_sub_ps(iz_SSE3, jz_SSE);
1226 /* rsq = dx*dx+dy*dy+dz*dz */
1227 rsq_SSE0 = gmx_mm_calc_rsq_ps(dx_SSE0, dy_SSE0, dz_SSE0);
1228 rsq_SSE1 = gmx_mm_calc_rsq_ps(dx_SSE1, dy_SSE1, dz_SSE1);
1229 rsq_SSE2 = gmx_mm_calc_rsq_ps(dx_SSE2, dy_SSE2, dz_SSE2);
1230 rsq_SSE3 = gmx_mm_calc_rsq_ps(dx_SSE3, dy_SSE3, dz_SSE3);
1233 jmask_SSE0 = _mm_and_ps(jmask_SSE0, imask_SSE0);
1234 jmask_SSE1 = _mm_and_ps(jmask_SSE1, imask_SSE1);
1235 jmask_SSE2 = _mm_and_ps(jmask_SSE2, imask_SSE2);
1236 jmask_SSE3 = _mm_and_ps(jmask_SSE3, imask_SSE3);
1238 /* Calculate 1/r and 1/r2 */
1239 rinv_SSE0 = gmx_mm_invsqrt_ps(rsq_SSE0);
1240 rinv_SSE1 = gmx_mm_invsqrt_ps(rsq_SSE1);
1241 rinv_SSE2 = gmx_mm_invsqrt_ps(rsq_SSE2);
1242 rinv_SSE3 = gmx_mm_invsqrt_ps(rsq_SSE3);
1245 rinv_SSE0 = _mm_and_ps(rinv_SSE0, jmask_SSE0);
1246 rinv_SSE1 = _mm_and_ps(rinv_SSE1, jmask_SSE1);
1247 rinv_SSE2 = _mm_and_ps(rinv_SSE2, jmask_SSE2);
1248 rinv_SSE3 = _mm_and_ps(rinv_SSE3, jmask_SSE3);
1250 irsq_SSE0 = _mm_mul_ps(rinv_SSE0, rinv_SSE0);
1251 irsq_SSE1 = _mm_mul_ps(rinv_SSE1, rinv_SSE1);
1252 irsq_SSE2 = _mm_mul_ps(rinv_SSE2, rinv_SSE2);
1253 irsq_SSE3 = _mm_mul_ps(rinv_SSE3, rinv_SSE3);
1254 idr4_SSE0 = _mm_mul_ps(irsq_SSE0, irsq_SSE0);
1255 idr4_SSE1 = _mm_mul_ps(irsq_SSE1, irsq_SSE1);
1256 idr4_SSE2 = _mm_mul_ps(irsq_SSE2, irsq_SSE2);
1257 idr4_SSE3 = _mm_mul_ps(irsq_SSE3, irsq_SSE3);
1258 idr6_SSE0 = _mm_mul_ps(idr4_SSE0, irsq_SSE0);
1259 idr6_SSE1 = _mm_mul_ps(idr4_SSE1, irsq_SSE1);
1260 idr6_SSE2 = _mm_mul_ps(idr4_SSE2, irsq_SSE2);
1261 idr6_SSE3 = _mm_mul_ps(idr4_SSE3, irsq_SSE3);
1263 raj_SSE = _mm_load_ps(gb_radius+j);
1264 vaj_SSE = _mm_load_ps(vsolv+j);
1266 rvdw_SSE0 = _mm_add_ps(rai_SSE0, raj_SSE);
1267 rvdw_SSE1 = _mm_add_ps(rai_SSE1, raj_SSE);
1268 rvdw_SSE2 = _mm_add_ps(rai_SSE2, raj_SSE);
1269 rvdw_SSE3 = _mm_add_ps(rai_SSE3, raj_SSE);
1271 ratio_SSE0 = _mm_mul_ps(rsq_SSE0, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE0, rvdw_SSE0)));
1272 ratio_SSE1 = _mm_mul_ps(rsq_SSE1, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE1, rvdw_SSE1)));
1273 ratio_SSE2 = _mm_mul_ps(rsq_SSE2, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE2, rvdw_SSE2)));
1274 ratio_SSE3 = _mm_mul_ps(rsq_SSE3, gmx_mm_inv_ps( _mm_mul_ps(rvdw_SSE3, rvdw_SSE3)));
1276 ratio_SSE0 = _mm_min_ps(ratio_SSE0, still_p5inv_SSE);
1277 ratio_SSE1 = _mm_min_ps(ratio_SSE1, still_p5inv_SSE);
1278 ratio_SSE2 = _mm_min_ps(ratio_SSE2, still_p5inv_SSE);
1279 ratio_SSE3 = _mm_min_ps(ratio_SSE3, still_p5inv_SSE);
1280 theta_SSE0 = _mm_mul_ps(ratio_SSE0, still_pip5_SSE);
1281 theta_SSE1 = _mm_mul_ps(ratio_SSE1, still_pip5_SSE);
1282 theta_SSE2 = _mm_mul_ps(ratio_SSE2, still_pip5_SSE);
1283 theta_SSE3 = _mm_mul_ps(ratio_SSE3, still_pip5_SSE);
1284 gmx_mm_sincos_ps(theta_SSE0, &sinq_SSE0, &cosq_SSE0);
1285 gmx_mm_sincos_ps(theta_SSE1, &sinq_SSE1, &cosq_SSE1);
1286 gmx_mm_sincos_ps(theta_SSE2, &sinq_SSE2, &cosq_SSE2);
1287 gmx_mm_sincos_ps(theta_SSE3, &sinq_SSE3, &cosq_SSE3);
1288 term_SSE0 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE0));
1289 term_SSE1 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE1));
1290 term_SSE2 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE2));
1291 term_SSE3 = _mm_mul_ps(half_SSE, _mm_sub_ps(one_SSE, cosq_SSE3));
1292 ccf_SSE0 = _mm_mul_ps(term_SSE0, term_SSE0);
1293 ccf_SSE1 = _mm_mul_ps(term_SSE1, term_SSE1);
1294 ccf_SSE2 = _mm_mul_ps(term_SSE2, term_SSE2);
1295 ccf_SSE3 = _mm_mul_ps(term_SSE3, term_SSE3);
1296 dccf_SSE0 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE0),
1297 _mm_mul_ps(sinq_SSE0, theta_SSE0));
1298 dccf_SSE1 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE1),
1299 _mm_mul_ps(sinq_SSE1, theta_SSE1));
1300 dccf_SSE2 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE2),
1301 _mm_mul_ps(sinq_SSE2, theta_SSE2));
1302 dccf_SSE3 = _mm_mul_ps(_mm_mul_ps(two_SSE, term_SSE3),
1303 _mm_mul_ps(sinq_SSE3, theta_SSE3));
1305 prod_SSE = _mm_mul_ps(still_p4_SSE, vaj_SSE);
1306 icf4_SSE0 = _mm_mul_ps(ccf_SSE0, idr4_SSE0);
1307 icf4_SSE1 = _mm_mul_ps(ccf_SSE1, idr4_SSE1);
1308 icf4_SSE2 = _mm_mul_ps(ccf_SSE2, idr4_SSE2);
1309 icf4_SSE3 = _mm_mul_ps(ccf_SSE3, idr4_SSE3);
1310 icf6_SSE0 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE0), dccf_SSE0), idr6_SSE0);
1311 icf6_SSE1 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE1), dccf_SSE1), idr6_SSE1);
1312 icf6_SSE2 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE2), dccf_SSE2), idr6_SSE2);
1313 icf6_SSE3 = _mm_mul_ps( _mm_sub_ps( _mm_mul_ps(four_SSE, ccf_SSE3), dccf_SSE3), idr6_SSE3);
1315 _mm_store_ps(work+j, _mm_add_ps(_mm_load_ps(work+j),
1316 gmx_mm_sum4_ps(_mm_mul_ps(prod_ai_SSE0, icf4_SSE0),
1317 _mm_mul_ps(prod_ai_SSE1, icf4_SSE1),
1318 _mm_mul_ps(prod_ai_SSE2, icf4_SSE2),
1319 _mm_mul_ps(prod_ai_SSE3, icf4_SSE3))));
1321 gpi_SSE0 = _mm_add_ps(gpi_SSE0, _mm_mul_ps(prod_SSE, icf4_SSE0));
1322 gpi_SSE1 = _mm_add_ps(gpi_SSE1, _mm_mul_ps(prod_SSE, icf4_SSE1));
1323 gpi_SSE2 = _mm_add_ps(gpi_SSE2, _mm_mul_ps(prod_SSE, icf4_SSE2));
1324 gpi_SSE3 = _mm_add_ps(gpi_SSE3, _mm_mul_ps(prod_SSE, icf4_SSE3));
1326 /* Save ai->aj and aj->ai chain rule terms */
1327 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE0));
1329 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE1));
1331 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE2));
1333 _mm_store_ps(dadx, _mm_mul_ps(prod_SSE, icf6_SSE3));
1336 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE0, icf6_SSE0));
1338 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE1, icf6_SSE1));
1340 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE2, icf6_SSE2));
1342 _mm_store_ps(dadx, _mm_mul_ps(prod_ai_SSE3, icf6_SSE3));
1345 _MM_TRANSPOSE4_PS(gpi_SSE0, gpi_SSE1, gpi_SSE2, gpi_SSE3);
1346 gpi_SSE0 = _mm_add_ps(gpi_SSE0, gpi_SSE1);
1347 gpi_SSE2 = _mm_add_ps(gpi_SSE2, gpi_SSE3);
1348 gpi_SSE0 = _mm_add_ps(gpi_SSE0, gpi_SSE2);
1349 _mm_store_ps(work+i, _mm_add_ps(gpi_SSE0, _mm_load_ps(work+i)));
1352 /* In case we have written anything beyond natoms, move it back.
1353 * Never mind that we leave stuff above natoms; that will not
1354 * be accessed later in the routine.
1355 * In principle this should be a move rather than sum, but this
1356 * way we dont have to worry about even/odd offsets...
1358 for (i = natoms; i < ni1+1+natoms/2; i++)
1360 work[i-natoms] += work[i];
1363 /* Parallel summations would go here if ever implemented with DD */
1365 factor = 0.5 * ONE_4PI_EPS0;
1366 /* Calculate the radii - should we do all atoms, or just our local ones? */
1367 for (i = 0; i < natoms; i++)
1369 if (born->use[i] != 0)
1371 gpi = born->gpol[i]+work[i];
1373 born->bRad[i] = factor*gmx_invsqrt(gpi2);
1374 fr->invsqrta[i] = gmx_invsqrt(born->bRad[i]);
1384 genborn_allvsall_calc_hct_obc_radii_sse2_single(t_forcerec * fr,
1385 t_mdatoms * mdatoms,
1386 gmx_genborn_t * born,
1388 gmx_localtop_t * top,
1393 gmx_allvsallgb2_data_t *aadata;
1396 int nj0, nj1, nj2, nj3;
1418 real rai, rai_inv, rai_inv2, sum_ai, sum_ai2, sum_ai3, tsum, tchain;
1420 __m128 ix_SSE0, iy_SSE0, iz_SSE0;
1421 __m128 ix_SSE1, iy_SSE1, iz_SSE1;
1422 __m128 ix_SSE2, iy_SSE2, iz_SSE2;
1423 __m128 ix_SSE3, iy_SSE3, iz_SSE3;
1424 __m128 gpi_SSE0, rai_SSE0, prod_ai_SSE0;
1425 __m128 gpi_SSE1, rai_SSE1, prod_ai_SSE1;
1426 __m128 gpi_SSE2, rai_SSE2, prod_ai_SSE2;
1427 __m128 gpi_SSE3, rai_SSE3, prod_ai_SSE3;
1428 __m128 imask_SSE0, jmask_SSE0;
1429 __m128 imask_SSE1, jmask_SSE1;
1430 __m128 imask_SSE2, jmask_SSE2;
1431 __m128 imask_SSE3, jmask_SSE3;
1432 __m128 jx_SSE, jy_SSE, jz_SSE;
1433 __m128 dx_SSE0, dy_SSE0, dz_SSE0;
1434 __m128 dx_SSE1, dy_SSE1, dz_SSE1;
1435 __m128 dx_SSE2, dy_SSE2, dz_SSE2;
1436 __m128 dx_SSE3, dy_SSE3, dz_SSE3;
1437 __m128 rsq_SSE0, rinv_SSE0, irsq_SSE0, idr4_SSE0, idr6_SSE0;
1438 __m128 rsq_SSE1, rinv_SSE1, irsq_SSE1, idr4_SSE1, idr6_SSE1;
1439 __m128 rsq_SSE2, rinv_SSE2, irsq_SSE2, idr4_SSE2, idr6_SSE2;
1440 __m128 rsq_SSE3, rinv_SSE3, irsq_SSE3, idr4_SSE3, idr6_SSE3;
1441 __m128 raj_SSE, raj_inv_SSE, sk_aj_SSE, sk2_aj_SSE;
1442 __m128 ccf_SSE0, dccf_SSE0, prod_SSE0;
1443 __m128 ccf_SSE1, dccf_SSE1, prod_SSE1;
1444 __m128 ccf_SSE2, dccf_SSE2, prod_SSE2;
1445 __m128 ccf_SSE3, dccf_SSE3, prod_SSE3;
1446 __m128 icf4_SSE0, icf6_SSE0;
1447 __m128 icf4_SSE1, icf6_SSE1;
1448 __m128 icf4_SSE2, icf6_SSE2;
1449 __m128 icf4_SSE3, icf6_SSE3;
1450 __m128 oneeighth_SSE, onefourth_SSE, half_SSE, one_SSE, two_SSE, four_SSE;
1451 __m128 still_p4_SSE, still_p5inv_SSE, still_pip5_SSE;
1452 __m128 rai_inv_SSE0;
1453 __m128 rai_inv_SSE1;
1454 __m128 rai_inv_SSE2;
1455 __m128 rai_inv_SSE3;
1456 __m128 sk_ai_SSE0, sk2_ai_SSE0, sum_ai_SSE0;
1457 __m128 sk_ai_SSE1, sk2_ai_SSE1, sum_ai_SSE1;
1458 __m128 sk_ai_SSE2, sk2_ai_SSE2, sum_ai_SSE2;
1459 __m128 sk_ai_SSE3, sk2_ai_SSE3, sum_ai_SSE3;
1460 __m128 lij_inv_SSE0, sk2_rinv_SSE0;
1461 __m128 lij_inv_SSE1, sk2_rinv_SSE1;
1462 __m128 lij_inv_SSE2, sk2_rinv_SSE2;
1463 __m128 lij_inv_SSE3, sk2_rinv_SSE3;
1468 __m128 t1_SSE0, t2_SSE0, t3_SSE0, t4_SSE0;
1469 __m128 t1_SSE1, t2_SSE1, t3_SSE1, t4_SSE1;
1470 __m128 t1_SSE2, t2_SSE2, t3_SSE2, t4_SSE2;
1471 __m128 t1_SSE3, t2_SSE3, t3_SSE3, t4_SSE3;
1472 __m128 obc_mask1_SSE0, obc_mask2_SSE0, obc_mask3_SSE0;
1473 __m128 obc_mask1_SSE1, obc_mask2_SSE1, obc_mask3_SSE1;
1474 __m128 obc_mask1_SSE2, obc_mask2_SSE2, obc_mask3_SSE2;
1475 __m128 obc_mask1_SSE3, obc_mask2_SSE3, obc_mask3_SSE3;
1476 __m128 uij_SSE0, uij2_SSE0, uij3_SSE0;
1477 __m128 uij_SSE1, uij2_SSE1, uij3_SSE1;
1478 __m128 uij_SSE2, uij2_SSE2, uij3_SSE2;
1479 __m128 uij_SSE3, uij2_SSE3, uij3_SSE3;
1480 __m128 lij_SSE0, lij2_SSE0, lij3_SSE0;
1481 __m128 lij_SSE1, lij2_SSE1, lij3_SSE1;
1482 __m128 lij_SSE2, lij2_SSE2, lij3_SSE2;
1483 __m128 lij_SSE3, lij2_SSE3, lij3_SSE3;
1484 __m128 dlij_SSE0, diff2_SSE0, logterm_SSE0;
1485 __m128 dlij_SSE1, diff2_SSE1, logterm_SSE1;
1486 __m128 dlij_SSE2, diff2_SSE2, logterm_SSE2;
1487 __m128 dlij_SSE3, diff2_SSE3, logterm_SSE3;
1490 natoms = mdatoms->nr;
1492 ni1 = mdatoms->homenr;
1496 aadata = *((gmx_allvsallgb2_data_t **)paadata);
1501 genborn_allvsall_setup(&aadata, top, born, mdatoms, born->gb_doffset,
1502 egbOBC, TRUE, TRUE, TRUE);
1503 *((gmx_allvsallgb2_data_t **)paadata) = aadata;
1506 x_align = aadata->x_align;
1507 y_align = aadata->y_align;
1508 z_align = aadata->z_align;
1510 gb_radius = aadata->gb_radius;
1511 work = aadata->work;
1512 jindex = aadata->jindex_gb;
1514 obc_param = aadata->workparam;
1516 oneeighth_SSE = _mm_set1_ps(0.125);
1517 onefourth_SSE = _mm_set1_ps(0.25);
1518 half_SSE = _mm_set1_ps(0.5);
1519 one_SSE = _mm_set1_ps(1.0);
1520 two_SSE = _mm_set1_ps(2.0);
1521 four_SSE = _mm_set1_ps(4.0);
1522 doffset_SSE = _mm_set1_ps(born->gb_doffset);
1524 for (i = 0; i < natoms; i++)
1526 x_align[i] = x[3*i];
1527 y_align[i] = x[3*i+1];
1528 z_align[i] = x[3*i+2];
1532 for (i = 0; i < natoms/2+1; i++)
1534 x_align[natoms+i] = x_align[i];
1535 y_align[natoms+i] = y_align[i];
1536 z_align[natoms+i] = z_align[i];
1539 for (i = 0; i < natoms+natoms/2+1; i++)
1544 for (i = ni0; i < ni1; i += UNROLLI)
1546 /* We assume shifts are NOT used for all-vs-all interactions */
1548 /* Load i atom data */
1549 ix_SSE0 = _mm_load1_ps(x_align+i);
1550 iy_SSE0 = _mm_load1_ps(y_align+i);
1551 iz_SSE0 = _mm_load1_ps(z_align+i);
1552 ix_SSE1 = _mm_load1_ps(x_align+i+1);
1553 iy_SSE1 = _mm_load1_ps(y_align+i+1);
1554 iz_SSE1 = _mm_load1_ps(z_align+i+1);
1555 ix_SSE2 = _mm_load1_ps(x_align+i+2);
1556 iy_SSE2 = _mm_load1_ps(y_align+i+2);
1557 iz_SSE2 = _mm_load1_ps(z_align+i+2);
1558 ix_SSE3 = _mm_load1_ps(x_align+i+3);
1559 iy_SSE3 = _mm_load1_ps(y_align+i+3);
1560 iz_SSE3 = _mm_load1_ps(z_align+i+3);
1562 rai_SSE0 = _mm_load1_ps(gb_radius+i);
1563 rai_SSE1 = _mm_load1_ps(gb_radius+i+1);
1564 rai_SSE2 = _mm_load1_ps(gb_radius+i+2);
1565 rai_SSE3 = _mm_load1_ps(gb_radius+i+3);
1566 rai_inv_SSE0 = gmx_mm_inv_ps(rai_SSE0);
1567 rai_inv_SSE1 = gmx_mm_inv_ps(rai_SSE1);
1568 rai_inv_SSE2 = gmx_mm_inv_ps(rai_SSE2);
1569 rai_inv_SSE3 = gmx_mm_inv_ps(rai_SSE3);
1571 sk_ai_SSE0 = _mm_load1_ps(obc_param+i);
1572 sk_ai_SSE1 = _mm_load1_ps(obc_param+i+1);
1573 sk_ai_SSE2 = _mm_load1_ps(obc_param+i+2);
1574 sk_ai_SSE3 = _mm_load1_ps(obc_param+i+3);
1575 sk2_ai_SSE0 = _mm_mul_ps(sk_ai_SSE0, sk_ai_SSE0);
1576 sk2_ai_SSE1 = _mm_mul_ps(sk_ai_SSE1, sk_ai_SSE1);
1577 sk2_ai_SSE2 = _mm_mul_ps(sk_ai_SSE2, sk_ai_SSE2);
1578 sk2_ai_SSE3 = _mm_mul_ps(sk_ai_SSE3, sk_ai_SSE3);
1580 sum_ai_SSE0 = _mm_setzero_ps();
1581 sum_ai_SSE1 = _mm_setzero_ps();
1582 sum_ai_SSE2 = _mm_setzero_ps();
1583 sum_ai_SSE3 = _mm_setzero_ps();
1585 /* Load limits for loop over neighbors */
1587 nj1 = jindex[4*i+1];
1588 nj2 = jindex[4*i+2];
1589 nj3 = jindex[4*i+3];
1591 pmask0 = aadata->prologue_mask_gb[i];
1592 pmask1 = aadata->prologue_mask_gb[i+1];
1593 pmask2 = aadata->prologue_mask_gb[i+2];
1594 pmask3 = aadata->prologue_mask_gb[i+3];
1595 emask0 = aadata->epilogue_mask[i];
1596 emask1 = aadata->epilogue_mask[i+1];
1597 emask2 = aadata->epilogue_mask[i+2];
1598 emask3 = aadata->epilogue_mask[i+3];
1600 imask_SSE0 = _mm_load1_ps((real *)(aadata->imask+i));
1601 imask_SSE1 = _mm_load1_ps((real *)(aadata->imask+i+1));
1602 imask_SSE2 = _mm_load1_ps((real *)(aadata->imask+i+2));
1603 imask_SSE3 = _mm_load1_ps((real *)(aadata->imask+i+3));
1605 /* Prologue part, including exclusion mask */
1606 for (j = nj0; j < nj1; j += UNROLLJ)
1608 jmask_SSE0 = _mm_load_ps((real *)pmask0);
1609 jmask_SSE1 = _mm_load_ps((real *)pmask1);
1610 jmask_SSE2 = _mm_load_ps((real *)pmask2);
1611 jmask_SSE3 = _mm_load_ps((real *)pmask3);
1617 /* load j atom coordinates */
1618 jx_SSE = _mm_load_ps(x_align+j);
1619 jy_SSE = _mm_load_ps(y_align+j);
1620 jz_SSE = _mm_load_ps(z_align+j);
1622 /* Calculate distance */
1623 dx_SSE0 = _mm_sub_ps(ix_SSE0, jx_SSE);
1624 dy_SSE0 = _mm_sub_ps(iy_SSE0, jy_SSE);
1625 dz_SSE0 = _mm_sub_ps(iz_SSE0, jz_SSE);
1626 dx_SSE1 = _mm_sub_ps(ix_SSE1, jx_SSE);
1627 dy_SSE1 = _mm_sub_ps(iy_SSE1, jy_SSE);
1628 dz_SSE1 = _mm_sub_ps(iz_SSE1, jz_SSE);
1629 dx_SSE2 = _mm_sub_ps(ix_SSE2, jx_SSE);
1630 dy_SSE2 = _mm_sub_ps(iy_SSE2, jy_SSE);
1631 dz_SSE2 = _mm_sub_ps(iz_SSE2, jz_SSE);
1632 dx_SSE3 = _mm_sub_ps(ix_SSE3, jx_SSE);
1633 dy_SSE3 = _mm_sub_ps(iy_SSE3, jy_SSE);
1634 dz_SSE3 = _mm_sub_ps(iz_SSE3, jz_SSE);
1636 /* rsq = dx*dx+dy*dy+dz*dz */
1637 rsq_SSE0 = gmx_mm_calc_rsq_ps(dx_SSE0, dy_SSE0, dz_SSE0);
1638 rsq_SSE1 = gmx_mm_calc_rsq_ps(dx_SSE1, dy_SSE1, dz_SSE1);
1639 rsq_SSE2 = gmx_mm_calc_rsq_ps(dx_SSE2, dy_SSE2, dz_SSE2);
1640 rsq_SSE3 = gmx_mm_calc_rsq_ps(dx_SSE3, dy_SSE3, dz_SSE3);
1643 jmask_SSE0 = _mm_and_ps(jmask_SSE0, imask_SSE0);
1644 jmask_SSE1 = _mm_and_ps(jmask_SSE1, imask_SSE1);
1645 jmask_SSE2 = _mm_and_ps(jmask_SSE2, imask_SSE2);
1646 jmask_SSE3 = _mm_and_ps(jmask_SSE3, imask_SSE3);
1648 /* Calculate 1/r and 1/r2 */
1649 rinv_SSE0 = gmx_mm_invsqrt_ps(rsq_SSE0);
1650 rinv_SSE1 = gmx_mm_invsqrt_ps(rsq_SSE1);
1651 rinv_SSE2 = gmx_mm_invsqrt_ps(rsq_SSE2);
1652 rinv_SSE3 = gmx_mm_invsqrt_ps(rsq_SSE3);
1655 rinv_SSE0 = _mm_and_ps(rinv_SSE0, jmask_SSE0);
1656 rinv_SSE1 = _mm_and_ps(rinv_SSE1, jmask_SSE1);
1657 rinv_SSE2 = _mm_and_ps(rinv_SSE2, jmask_SSE2);
1658 rinv_SSE3 = _mm_and_ps(rinv_SSE3, jmask_SSE3);
1660 dr_SSE0 = _mm_mul_ps(rsq_SSE0, rinv_SSE0);
1661 dr_SSE1 = _mm_mul_ps(rsq_SSE1, rinv_SSE1);
1662 dr_SSE2 = _mm_mul_ps(rsq_SSE2, rinv_SSE2);
1663 dr_SSE3 = _mm_mul_ps(rsq_SSE3, rinv_SSE3);
1665 sk_aj_SSE = _mm_load_ps(obc_param+j);
1666 raj_SSE = _mm_load_ps(gb_radius+j);
1667 raj_inv_SSE = gmx_mm_inv_ps(raj_SSE);
1669 /* Evaluate influence of atom aj -> ai */
1670 t1_SSE0 = _mm_add_ps(dr_SSE0, sk_aj_SSE);
1671 t1_SSE1 = _mm_add_ps(dr_SSE1, sk_aj_SSE);
1672 t1_SSE2 = _mm_add_ps(dr_SSE2, sk_aj_SSE);
1673 t1_SSE3 = _mm_add_ps(dr_SSE3, sk_aj_SSE);
1674 t2_SSE0 = _mm_sub_ps(dr_SSE0, sk_aj_SSE);
1675 t2_SSE1 = _mm_sub_ps(dr_SSE1, sk_aj_SSE);
1676 t2_SSE2 = _mm_sub_ps(dr_SSE2, sk_aj_SSE);
1677 t2_SSE3 = _mm_sub_ps(dr_SSE3, sk_aj_SSE);
1678 t3_SSE0 = _mm_sub_ps(sk_aj_SSE, dr_SSE0);
1679 t3_SSE1 = _mm_sub_ps(sk_aj_SSE, dr_SSE1);
1680 t3_SSE2 = _mm_sub_ps(sk_aj_SSE, dr_SSE2);
1681 t3_SSE3 = _mm_sub_ps(sk_aj_SSE, dr_SSE3);
1683 obc_mask1_SSE0 = _mm_cmplt_ps(rai_SSE0, t1_SSE0);
1684 obc_mask1_SSE1 = _mm_cmplt_ps(rai_SSE1, t1_SSE1);
1685 obc_mask1_SSE2 = _mm_cmplt_ps(rai_SSE2, t1_SSE2);
1686 obc_mask1_SSE3 = _mm_cmplt_ps(rai_SSE3, t1_SSE3);
1687 obc_mask2_SSE0 = _mm_cmplt_ps(rai_SSE0, t2_SSE0);
1688 obc_mask2_SSE1 = _mm_cmplt_ps(rai_SSE1, t2_SSE1);
1689 obc_mask2_SSE2 = _mm_cmplt_ps(rai_SSE2, t2_SSE2);
1690 obc_mask2_SSE3 = _mm_cmplt_ps(rai_SSE3, t2_SSE3);
1691 obc_mask3_SSE0 = _mm_cmplt_ps(rai_SSE0, t3_SSE0);
1692 obc_mask3_SSE1 = _mm_cmplt_ps(rai_SSE1, t3_SSE1);
1693 obc_mask3_SSE2 = _mm_cmplt_ps(rai_SSE2, t3_SSE2);
1694 obc_mask3_SSE3 = _mm_cmplt_ps(rai_SSE3, t3_SSE3);
1695 obc_mask1_SSE0 = _mm_and_ps(obc_mask1_SSE0, jmask_SSE0);
1696 obc_mask1_SSE1 = _mm_and_ps(obc_mask1_SSE1, jmask_SSE1);
1697 obc_mask1_SSE2 = _mm_and_ps(obc_mask1_SSE2, jmask_SSE2);
1698 obc_mask1_SSE3 = _mm_and_ps(obc_mask1_SSE3, jmask_SSE3);
1700 uij_SSE0 = gmx_mm_inv_ps(t1_SSE0);
1701 uij_SSE1 = gmx_mm_inv_ps(t1_SSE1);
1702 uij_SSE2 = gmx_mm_inv_ps(t1_SSE2);
1703 uij_SSE3 = gmx_mm_inv_ps(t1_SSE3);
1704 lij_SSE0 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE0, gmx_mm_inv_ps(t2_SSE0)),
1705 _mm_andnot_ps(obc_mask2_SSE0, rai_inv_SSE0));
1706 lij_SSE1 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE1, gmx_mm_inv_ps(t2_SSE1)),
1707 _mm_andnot_ps(obc_mask2_SSE1, rai_inv_SSE1));
1708 lij_SSE2 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE2, gmx_mm_inv_ps(t2_SSE2)),
1709 _mm_andnot_ps(obc_mask2_SSE2, rai_inv_SSE2));
1710 lij_SSE3 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE3, gmx_mm_inv_ps(t2_SSE3)),
1711 _mm_andnot_ps(obc_mask2_SSE3, rai_inv_SSE3));
1712 dlij_SSE0 = _mm_and_ps(one_SSE, obc_mask2_SSE0);
1713 dlij_SSE1 = _mm_and_ps(one_SSE, obc_mask2_SSE1);
1714 dlij_SSE2 = _mm_and_ps(one_SSE, obc_mask2_SSE2);
1715 dlij_SSE3 = _mm_and_ps(one_SSE, obc_mask2_SSE3);
1717 uij2_SSE0 = _mm_mul_ps(uij_SSE0, uij_SSE0);
1718 uij2_SSE1 = _mm_mul_ps(uij_SSE1, uij_SSE1);
1719 uij2_SSE2 = _mm_mul_ps(uij_SSE2, uij_SSE2);
1720 uij2_SSE3 = _mm_mul_ps(uij_SSE3, uij_SSE3);
1721 uij3_SSE0 = _mm_mul_ps(uij2_SSE0, uij_SSE0);
1722 uij3_SSE1 = _mm_mul_ps(uij2_SSE1, uij_SSE1);
1723 uij3_SSE2 = _mm_mul_ps(uij2_SSE2, uij_SSE2);
1724 uij3_SSE3 = _mm_mul_ps(uij2_SSE3, uij_SSE3);
1725 lij2_SSE0 = _mm_mul_ps(lij_SSE0, lij_SSE0);
1726 lij2_SSE1 = _mm_mul_ps(lij_SSE1, lij_SSE1);
1727 lij2_SSE2 = _mm_mul_ps(lij_SSE2, lij_SSE2);
1728 lij2_SSE3 = _mm_mul_ps(lij_SSE3, lij_SSE3);
1729 lij3_SSE0 = _mm_mul_ps(lij2_SSE0, lij_SSE0);
1730 lij3_SSE1 = _mm_mul_ps(lij2_SSE1, lij_SSE1);
1731 lij3_SSE2 = _mm_mul_ps(lij2_SSE2, lij_SSE2);
1732 lij3_SSE3 = _mm_mul_ps(lij2_SSE3, lij_SSE3);
1734 diff2_SSE0 = _mm_sub_ps(uij2_SSE0, lij2_SSE0);
1735 diff2_SSE1 = _mm_sub_ps(uij2_SSE1, lij2_SSE1);
1736 diff2_SSE2 = _mm_sub_ps(uij2_SSE2, lij2_SSE2);
1737 diff2_SSE3 = _mm_sub_ps(uij2_SSE3, lij2_SSE3);
1738 lij_inv_SSE0 = gmx_mm_invsqrt_ps(lij2_SSE0);
1739 lij_inv_SSE1 = gmx_mm_invsqrt_ps(lij2_SSE1);
1740 lij_inv_SSE2 = gmx_mm_invsqrt_ps(lij2_SSE2);
1741 lij_inv_SSE3 = gmx_mm_invsqrt_ps(lij2_SSE3);
1742 sk2_aj_SSE = _mm_mul_ps(sk_aj_SSE, sk_aj_SSE);
1743 sk2_rinv_SSE0 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE0);
1744 sk2_rinv_SSE1 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE1);
1745 sk2_rinv_SSE2 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE2);
1746 sk2_rinv_SSE3 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE3);
1747 prod_SSE0 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE0);
1748 prod_SSE1 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE1);
1749 prod_SSE2 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE2);
1750 prod_SSE3 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE3);
1752 logterm_SSE0 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE0, lij_inv_SSE0));
1753 logterm_SSE1 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE1, lij_inv_SSE1));
1754 logterm_SSE2 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE2, lij_inv_SSE2));
1755 logterm_SSE3 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE3, lij_inv_SSE3));
1757 t1_SSE0 = _mm_sub_ps(lij_SSE0, uij_SSE0);
1758 t1_SSE1 = _mm_sub_ps(lij_SSE1, uij_SSE1);
1759 t1_SSE2 = _mm_sub_ps(lij_SSE2, uij_SSE2);
1760 t1_SSE3 = _mm_sub_ps(lij_SSE3, uij_SSE3);
1761 t2_SSE0 = _mm_mul_ps(diff2_SSE0,
1762 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE0),
1764 t2_SSE1 = _mm_mul_ps(diff2_SSE1,
1765 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE1),
1767 t2_SSE2 = _mm_mul_ps(diff2_SSE2,
1768 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE2),
1770 t2_SSE3 = _mm_mul_ps(diff2_SSE3,
1771 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE3),
1774 t3_SSE0 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE0, logterm_SSE0));
1775 t3_SSE1 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE1, logterm_SSE1));
1776 t3_SSE2 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE2, logterm_SSE2));
1777 t3_SSE3 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE3, logterm_SSE3));
1778 t1_SSE0 = _mm_add_ps(t1_SSE0, _mm_add_ps(t2_SSE0, t3_SSE0));
1779 t1_SSE1 = _mm_add_ps(t1_SSE1, _mm_add_ps(t2_SSE1, t3_SSE1));
1780 t1_SSE2 = _mm_add_ps(t1_SSE2, _mm_add_ps(t2_SSE2, t3_SSE2));
1781 t1_SSE3 = _mm_add_ps(t1_SSE3, _mm_add_ps(t2_SSE3, t3_SSE3));
1782 t4_SSE0 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE0, lij_SSE0));
1783 t4_SSE1 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE1, lij_SSE1));
1784 t4_SSE2 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE2, lij_SSE2));
1785 t4_SSE3 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE3, lij_SSE3));
1786 t4_SSE0 = _mm_and_ps(t4_SSE0, obc_mask3_SSE0);
1787 t4_SSE1 = _mm_and_ps(t4_SSE1, obc_mask3_SSE1);
1788 t4_SSE2 = _mm_and_ps(t4_SSE2, obc_mask3_SSE2);
1789 t4_SSE3 = _mm_and_ps(t4_SSE3, obc_mask3_SSE3);
1790 t1_SSE0 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE0, t4_SSE0));
1791 t1_SSE1 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE1, t4_SSE1));
1792 t1_SSE2 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE2, t4_SSE2));
1793 t1_SSE3 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE3, t4_SSE3));
1795 sum_ai_SSE0 = _mm_add_ps(sum_ai_SSE0, _mm_and_ps(t1_SSE0, obc_mask1_SSE0));
1796 sum_ai_SSE1 = _mm_add_ps(sum_ai_SSE1, _mm_and_ps(t1_SSE1, obc_mask1_SSE1));
1797 sum_ai_SSE2 = _mm_add_ps(sum_ai_SSE2, _mm_and_ps(t1_SSE2, obc_mask1_SSE2));
1798 sum_ai_SSE3 = _mm_add_ps(sum_ai_SSE3, _mm_and_ps(t1_SSE3, obc_mask1_SSE3));
1800 t1_SSE0 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE0),
1801 _mm_mul_ps(prod_SSE0, lij3_SSE0));
1802 t1_SSE1 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE1),
1803 _mm_mul_ps(prod_SSE1, lij3_SSE1));
1804 t1_SSE2 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE2),
1805 _mm_mul_ps(prod_SSE2, lij3_SSE2));
1806 t1_SSE3 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE3),
1807 _mm_mul_ps(prod_SSE3, lij3_SSE3));
1808 t1_SSE0 = _mm_sub_ps(t1_SSE0,
1809 _mm_mul_ps(onefourth_SSE,
1810 _mm_add_ps(_mm_mul_ps(lij_SSE0, rinv_SSE0),
1811 _mm_mul_ps(lij3_SSE0, dr_SSE0))));
1812 t1_SSE1 = _mm_sub_ps(t1_SSE1,
1813 _mm_mul_ps(onefourth_SSE,
1814 _mm_add_ps(_mm_mul_ps(lij_SSE1, rinv_SSE1),
1815 _mm_mul_ps(lij3_SSE1, dr_SSE1))));
1816 t1_SSE2 = _mm_sub_ps(t1_SSE2,
1817 _mm_mul_ps(onefourth_SSE,
1818 _mm_add_ps(_mm_mul_ps(lij_SSE2, rinv_SSE2),
1819 _mm_mul_ps(lij3_SSE2, dr_SSE2))));
1820 t1_SSE3 = _mm_sub_ps(t1_SSE3,
1821 _mm_mul_ps(onefourth_SSE,
1822 _mm_add_ps(_mm_mul_ps(lij_SSE3, rinv_SSE3),
1823 _mm_mul_ps(lij3_SSE3, dr_SSE3))));
1825 t2_SSE0 = _mm_mul_ps(onefourth_SSE,
1826 _mm_add_ps(_mm_mul_ps(uij_SSE0, rinv_SSE0),
1827 _mm_mul_ps(uij3_SSE0, dr_SSE0)));
1828 t2_SSE1 = _mm_mul_ps(onefourth_SSE,
1829 _mm_add_ps(_mm_mul_ps(uij_SSE1, rinv_SSE1),
1830 _mm_mul_ps(uij3_SSE1, dr_SSE1)));
1831 t2_SSE2 = _mm_mul_ps(onefourth_SSE,
1832 _mm_add_ps(_mm_mul_ps(uij_SSE2, rinv_SSE2),
1833 _mm_mul_ps(uij3_SSE2, dr_SSE2)));
1834 t2_SSE3 = _mm_mul_ps(onefourth_SSE,
1835 _mm_add_ps(_mm_mul_ps(uij_SSE3, rinv_SSE3),
1836 _mm_mul_ps(uij3_SSE3, dr_SSE3)));
1837 t2_SSE0 = _mm_sub_ps(t2_SSE0,
1838 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE0),
1839 _mm_mul_ps(prod_SSE0, uij3_SSE0)));
1840 t2_SSE1 = _mm_sub_ps(t2_SSE1,
1841 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE1),
1842 _mm_mul_ps(prod_SSE1, uij3_SSE1)));
1843 t2_SSE2 = _mm_sub_ps(t2_SSE2,
1844 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE2),
1845 _mm_mul_ps(prod_SSE2, uij3_SSE2)));
1846 t2_SSE3 = _mm_sub_ps(t2_SSE3,
1847 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE3),
1848 _mm_mul_ps(prod_SSE3, uij3_SSE3)));
1849 t3_SSE0 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE0),
1850 _mm_mul_ps(rinv_SSE0, rinv_SSE0));
1851 t3_SSE1 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE1),
1852 _mm_mul_ps(rinv_SSE1, rinv_SSE1));
1853 t3_SSE2 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE2),
1854 _mm_mul_ps(rinv_SSE2, rinv_SSE2));
1855 t3_SSE3 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE3),
1856 _mm_mul_ps(rinv_SSE3, rinv_SSE3));
1857 t3_SSE0 = _mm_sub_ps(t3_SSE0,
1858 _mm_mul_ps(_mm_mul_ps(diff2_SSE0, oneeighth_SSE),
1860 _mm_mul_ps(sk2_rinv_SSE0, rinv_SSE0))));
1861 t3_SSE1 = _mm_sub_ps(t3_SSE1,
1862 _mm_mul_ps(_mm_mul_ps(diff2_SSE1, oneeighth_SSE),
1864 _mm_mul_ps(sk2_rinv_SSE1, rinv_SSE1))));
1865 t3_SSE2 = _mm_sub_ps(t3_SSE2,
1866 _mm_mul_ps(_mm_mul_ps(diff2_SSE2, oneeighth_SSE),
1868 _mm_mul_ps(sk2_rinv_SSE2, rinv_SSE2))));
1869 t3_SSE3 = _mm_sub_ps(t3_SSE3,
1870 _mm_mul_ps(_mm_mul_ps(diff2_SSE3, oneeighth_SSE),
1872 _mm_mul_ps(sk2_rinv_SSE3, rinv_SSE3))));
1874 t1_SSE0 = _mm_mul_ps(rinv_SSE0,
1875 _mm_add_ps(_mm_mul_ps(dlij_SSE0, t1_SSE0),
1876 _mm_add_ps(t2_SSE0, t3_SSE0)));
1877 t1_SSE1 = _mm_mul_ps(rinv_SSE1,
1878 _mm_add_ps(_mm_mul_ps(dlij_SSE1, t1_SSE1),
1879 _mm_add_ps(t2_SSE1, t3_SSE1)));
1880 t1_SSE2 = _mm_mul_ps(rinv_SSE2,
1881 _mm_add_ps(_mm_mul_ps(dlij_SSE2, t1_SSE2),
1882 _mm_add_ps(t2_SSE2, t3_SSE2)));
1883 t1_SSE3 = _mm_mul_ps(rinv_SSE3,
1884 _mm_add_ps(_mm_mul_ps(dlij_SSE3, t1_SSE3),
1885 _mm_add_ps(t2_SSE3, t3_SSE3)));
1887 _mm_store_ps(dadx, _mm_and_ps(t1_SSE0, obc_mask1_SSE0));
1889 _mm_store_ps(dadx, _mm_and_ps(t1_SSE1, obc_mask1_SSE1));
1891 _mm_store_ps(dadx, _mm_and_ps(t1_SSE2, obc_mask1_SSE2));
1893 _mm_store_ps(dadx, _mm_and_ps(t1_SSE3, obc_mask1_SSE3));
1896 /* Evaluate influence of atom ai -> aj */
1897 t1_SSE0 = _mm_add_ps(dr_SSE0, sk_ai_SSE0);
1898 t1_SSE1 = _mm_add_ps(dr_SSE1, sk_ai_SSE1);
1899 t1_SSE2 = _mm_add_ps(dr_SSE2, sk_ai_SSE2);
1900 t1_SSE3 = _mm_add_ps(dr_SSE3, sk_ai_SSE3);
1901 t2_SSE0 = _mm_sub_ps(dr_SSE0, sk_ai_SSE0);
1902 t2_SSE1 = _mm_sub_ps(dr_SSE1, sk_ai_SSE1);
1903 t2_SSE2 = _mm_sub_ps(dr_SSE2, sk_ai_SSE2);
1904 t2_SSE3 = _mm_sub_ps(dr_SSE3, sk_ai_SSE3);
1905 t3_SSE0 = _mm_sub_ps(sk_ai_SSE0, dr_SSE0);
1906 t3_SSE1 = _mm_sub_ps(sk_ai_SSE1, dr_SSE1);
1907 t3_SSE2 = _mm_sub_ps(sk_ai_SSE2, dr_SSE2);
1908 t3_SSE3 = _mm_sub_ps(sk_ai_SSE3, dr_SSE3);
1910 obc_mask1_SSE0 = _mm_cmplt_ps(raj_SSE, t1_SSE0);
1911 obc_mask1_SSE1 = _mm_cmplt_ps(raj_SSE, t1_SSE1);
1912 obc_mask1_SSE2 = _mm_cmplt_ps(raj_SSE, t1_SSE2);
1913 obc_mask1_SSE3 = _mm_cmplt_ps(raj_SSE, t1_SSE3);
1914 obc_mask2_SSE0 = _mm_cmplt_ps(raj_SSE, t2_SSE0);
1915 obc_mask2_SSE1 = _mm_cmplt_ps(raj_SSE, t2_SSE1);
1916 obc_mask2_SSE2 = _mm_cmplt_ps(raj_SSE, t2_SSE2);
1917 obc_mask2_SSE3 = _mm_cmplt_ps(raj_SSE, t2_SSE3);
1918 obc_mask3_SSE0 = _mm_cmplt_ps(raj_SSE, t3_SSE0);
1919 obc_mask3_SSE1 = _mm_cmplt_ps(raj_SSE, t3_SSE1);
1920 obc_mask3_SSE2 = _mm_cmplt_ps(raj_SSE, t3_SSE2);
1921 obc_mask3_SSE3 = _mm_cmplt_ps(raj_SSE, t3_SSE3);
1922 obc_mask1_SSE0 = _mm_and_ps(obc_mask1_SSE0, jmask_SSE0);
1923 obc_mask1_SSE1 = _mm_and_ps(obc_mask1_SSE1, jmask_SSE1);
1924 obc_mask1_SSE2 = _mm_and_ps(obc_mask1_SSE2, jmask_SSE2);
1925 obc_mask1_SSE3 = _mm_and_ps(obc_mask1_SSE3, jmask_SSE3);
1927 uij_SSE0 = gmx_mm_inv_ps(t1_SSE0);
1928 uij_SSE1 = gmx_mm_inv_ps(t1_SSE1);
1929 uij_SSE2 = gmx_mm_inv_ps(t1_SSE2);
1930 uij_SSE3 = gmx_mm_inv_ps(t1_SSE3);
1931 lij_SSE0 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE0, gmx_mm_inv_ps(t2_SSE0)),
1932 _mm_andnot_ps(obc_mask2_SSE0, raj_inv_SSE));
1933 lij_SSE1 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE1, gmx_mm_inv_ps(t2_SSE1)),
1934 _mm_andnot_ps(obc_mask2_SSE1, raj_inv_SSE));
1935 lij_SSE2 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE2, gmx_mm_inv_ps(t2_SSE2)),
1936 _mm_andnot_ps(obc_mask2_SSE2, raj_inv_SSE));
1937 lij_SSE3 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE3, gmx_mm_inv_ps(t2_SSE3)),
1938 _mm_andnot_ps(obc_mask2_SSE3, raj_inv_SSE));
1939 dlij_SSE0 = _mm_and_ps(one_SSE, obc_mask2_SSE0);
1940 dlij_SSE1 = _mm_and_ps(one_SSE, obc_mask2_SSE1);
1941 dlij_SSE2 = _mm_and_ps(one_SSE, obc_mask2_SSE2);
1942 dlij_SSE3 = _mm_and_ps(one_SSE, obc_mask2_SSE3);
1944 uij2_SSE0 = _mm_mul_ps(uij_SSE0, uij_SSE0);
1945 uij2_SSE1 = _mm_mul_ps(uij_SSE1, uij_SSE1);
1946 uij2_SSE2 = _mm_mul_ps(uij_SSE2, uij_SSE2);
1947 uij2_SSE3 = _mm_mul_ps(uij_SSE3, uij_SSE3);
1948 uij3_SSE0 = _mm_mul_ps(uij2_SSE0, uij_SSE0);
1949 uij3_SSE1 = _mm_mul_ps(uij2_SSE1, uij_SSE1);
1950 uij3_SSE2 = _mm_mul_ps(uij2_SSE2, uij_SSE2);
1951 uij3_SSE3 = _mm_mul_ps(uij2_SSE3, uij_SSE3);
1952 lij2_SSE0 = _mm_mul_ps(lij_SSE0, lij_SSE0);
1953 lij2_SSE1 = _mm_mul_ps(lij_SSE1, lij_SSE1);
1954 lij2_SSE2 = _mm_mul_ps(lij_SSE2, lij_SSE2);
1955 lij2_SSE3 = _mm_mul_ps(lij_SSE3, lij_SSE3);
1956 lij3_SSE0 = _mm_mul_ps(lij2_SSE0, lij_SSE0);
1957 lij3_SSE1 = _mm_mul_ps(lij2_SSE1, lij_SSE1);
1958 lij3_SSE2 = _mm_mul_ps(lij2_SSE2, lij_SSE2);
1959 lij3_SSE3 = _mm_mul_ps(lij2_SSE3, lij_SSE3);
1961 diff2_SSE0 = _mm_sub_ps(uij2_SSE0, lij2_SSE0);
1962 diff2_SSE1 = _mm_sub_ps(uij2_SSE1, lij2_SSE1);
1963 diff2_SSE2 = _mm_sub_ps(uij2_SSE2, lij2_SSE2);
1964 diff2_SSE3 = _mm_sub_ps(uij2_SSE3, lij2_SSE3);
1965 lij_inv_SSE0 = gmx_mm_invsqrt_ps(lij2_SSE0);
1966 lij_inv_SSE1 = gmx_mm_invsqrt_ps(lij2_SSE1);
1967 lij_inv_SSE2 = gmx_mm_invsqrt_ps(lij2_SSE2);
1968 lij_inv_SSE3 = gmx_mm_invsqrt_ps(lij2_SSE3);
1969 sk2_rinv_SSE0 = _mm_mul_ps(sk2_ai_SSE0, rinv_SSE0);
1970 sk2_rinv_SSE1 = _mm_mul_ps(sk2_ai_SSE1, rinv_SSE1);
1971 sk2_rinv_SSE2 = _mm_mul_ps(sk2_ai_SSE2, rinv_SSE2);
1972 sk2_rinv_SSE3 = _mm_mul_ps(sk2_ai_SSE3, rinv_SSE3);
1973 prod_SSE0 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE0);
1974 prod_SSE1 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE1);
1975 prod_SSE2 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE2);
1976 prod_SSE3 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE3);
1978 logterm_SSE0 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE0, lij_inv_SSE0));
1979 logterm_SSE1 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE1, lij_inv_SSE1));
1980 logterm_SSE2 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE2, lij_inv_SSE2));
1981 logterm_SSE3 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE3, lij_inv_SSE3));
1982 t1_SSE0 = _mm_sub_ps(lij_SSE0, uij_SSE0);
1983 t1_SSE1 = _mm_sub_ps(lij_SSE1, uij_SSE1);
1984 t1_SSE2 = _mm_sub_ps(lij_SSE2, uij_SSE2);
1985 t1_SSE3 = _mm_sub_ps(lij_SSE3, uij_SSE3);
1986 t2_SSE0 = _mm_mul_ps(diff2_SSE0,
1987 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE0),
1989 t2_SSE1 = _mm_mul_ps(diff2_SSE1,
1990 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE1),
1992 t2_SSE2 = _mm_mul_ps(diff2_SSE2,
1993 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE2),
1995 t2_SSE3 = _mm_mul_ps(diff2_SSE3,
1996 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE3),
1998 t3_SSE0 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE0, logterm_SSE0));
1999 t3_SSE1 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE1, logterm_SSE1));
2000 t3_SSE2 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE2, logterm_SSE2));
2001 t3_SSE3 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE3, logterm_SSE3));
2002 t1_SSE0 = _mm_add_ps(t1_SSE0, _mm_add_ps(t2_SSE0, t3_SSE0));
2003 t1_SSE1 = _mm_add_ps(t1_SSE1, _mm_add_ps(t2_SSE1, t3_SSE1));
2004 t1_SSE2 = _mm_add_ps(t1_SSE2, _mm_add_ps(t2_SSE2, t3_SSE2));
2005 t1_SSE3 = _mm_add_ps(t1_SSE3, _mm_add_ps(t2_SSE3, t3_SSE3));
2006 t4_SSE0 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE0));
2007 t4_SSE1 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE1));
2008 t4_SSE2 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE2));
2009 t4_SSE3 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE3));
2010 t4_SSE0 = _mm_and_ps(t4_SSE0, obc_mask3_SSE0);
2011 t4_SSE1 = _mm_and_ps(t4_SSE1, obc_mask3_SSE1);
2012 t4_SSE2 = _mm_and_ps(t4_SSE2, obc_mask3_SSE2);
2013 t4_SSE3 = _mm_and_ps(t4_SSE3, obc_mask3_SSE3);
2014 t1_SSE0 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE0, t4_SSE0));
2015 t1_SSE1 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE1, t4_SSE1));
2016 t1_SSE2 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE2, t4_SSE2));
2017 t1_SSE3 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE3, t4_SSE3));
2019 _mm_store_ps(work+j, _mm_add_ps(_mm_load_ps(work+j),
2020 gmx_mm_sum4_ps(_mm_and_ps(t1_SSE0, obc_mask1_SSE0),
2021 _mm_and_ps(t1_SSE1, obc_mask1_SSE1),
2022 _mm_and_ps(t1_SSE2, obc_mask1_SSE2),
2023 _mm_and_ps(t1_SSE3, obc_mask1_SSE3))));
2025 t1_SSE0 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE0),
2026 _mm_mul_ps(prod_SSE0, lij3_SSE0));
2027 t1_SSE1 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE1),
2028 _mm_mul_ps(prod_SSE1, lij3_SSE1));
2029 t1_SSE2 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE2),
2030 _mm_mul_ps(prod_SSE2, lij3_SSE2));
2031 t1_SSE3 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE3),
2032 _mm_mul_ps(prod_SSE3, lij3_SSE3));
2033 t1_SSE0 = _mm_sub_ps(t1_SSE0,
2034 _mm_mul_ps(onefourth_SSE,
2035 _mm_add_ps(_mm_mul_ps(lij_SSE0, rinv_SSE0),
2036 _mm_mul_ps(lij3_SSE0, dr_SSE0))));
2037 t1_SSE1 = _mm_sub_ps(t1_SSE1,
2038 _mm_mul_ps(onefourth_SSE,
2039 _mm_add_ps(_mm_mul_ps(lij_SSE1, rinv_SSE1),
2040 _mm_mul_ps(lij3_SSE1, dr_SSE1))));
2041 t1_SSE2 = _mm_sub_ps(t1_SSE2,
2042 _mm_mul_ps(onefourth_SSE,
2043 _mm_add_ps(_mm_mul_ps(lij_SSE2, rinv_SSE2),
2044 _mm_mul_ps(lij3_SSE2, dr_SSE2))));
2045 t1_SSE3 = _mm_sub_ps(t1_SSE3,
2046 _mm_mul_ps(onefourth_SSE,
2047 _mm_add_ps(_mm_mul_ps(lij_SSE3, rinv_SSE3),
2048 _mm_mul_ps(lij3_SSE3, dr_SSE3))));
2049 t2_SSE0 = _mm_mul_ps(onefourth_SSE,
2050 _mm_add_ps(_mm_mul_ps(uij_SSE0, rinv_SSE0),
2051 _mm_mul_ps(uij3_SSE0, dr_SSE0)));
2052 t2_SSE1 = _mm_mul_ps(onefourth_SSE,
2053 _mm_add_ps(_mm_mul_ps(uij_SSE1, rinv_SSE1),
2054 _mm_mul_ps(uij3_SSE1, dr_SSE1)));
2055 t2_SSE2 = _mm_mul_ps(onefourth_SSE,
2056 _mm_add_ps(_mm_mul_ps(uij_SSE2, rinv_SSE2),
2057 _mm_mul_ps(uij3_SSE2, dr_SSE2)));
2058 t2_SSE3 = _mm_mul_ps(onefourth_SSE,
2059 _mm_add_ps(_mm_mul_ps(uij_SSE3, rinv_SSE3),
2060 _mm_mul_ps(uij3_SSE3, dr_SSE3)));
2061 t2_SSE0 = _mm_sub_ps(t2_SSE0,
2062 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE0),
2063 _mm_mul_ps(prod_SSE0, uij3_SSE0)));
2064 t2_SSE1 = _mm_sub_ps(t2_SSE1,
2065 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE1),
2066 _mm_mul_ps(prod_SSE1, uij3_SSE1)));
2067 t2_SSE2 = _mm_sub_ps(t2_SSE2,
2068 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE2),
2069 _mm_mul_ps(prod_SSE2, uij3_SSE2)));
2070 t2_SSE3 = _mm_sub_ps(t2_SSE3,
2071 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE3),
2072 _mm_mul_ps(prod_SSE3, uij3_SSE3)));
2074 t3_SSE0 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE0),
2075 _mm_mul_ps(rinv_SSE0, rinv_SSE0));
2076 t3_SSE1 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE1),
2077 _mm_mul_ps(rinv_SSE1, rinv_SSE1));
2078 t3_SSE2 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE2),
2079 _mm_mul_ps(rinv_SSE2, rinv_SSE2));
2080 t3_SSE3 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE3),
2081 _mm_mul_ps(rinv_SSE3, rinv_SSE3));
2083 t3_SSE0 = _mm_sub_ps(t3_SSE0,
2084 _mm_mul_ps(_mm_mul_ps(diff2_SSE0, oneeighth_SSE),
2086 _mm_mul_ps(sk2_rinv_SSE0, rinv_SSE0))));
2087 t3_SSE1 = _mm_sub_ps(t3_SSE1,
2088 _mm_mul_ps(_mm_mul_ps(diff2_SSE1, oneeighth_SSE),
2090 _mm_mul_ps(sk2_rinv_SSE1, rinv_SSE1))));
2091 t3_SSE2 = _mm_sub_ps(t3_SSE2,
2092 _mm_mul_ps(_mm_mul_ps(diff2_SSE2, oneeighth_SSE),
2094 _mm_mul_ps(sk2_rinv_SSE2, rinv_SSE2))));
2095 t3_SSE3 = _mm_sub_ps(t3_SSE3,
2096 _mm_mul_ps(_mm_mul_ps(diff2_SSE3, oneeighth_SSE),
2098 _mm_mul_ps(sk2_rinv_SSE3, rinv_SSE3))));
2101 t1_SSE0 = _mm_mul_ps(rinv_SSE0,
2102 _mm_add_ps(_mm_mul_ps(dlij_SSE0, t1_SSE0),
2103 _mm_add_ps(t2_SSE0, t3_SSE0)));
2104 t1_SSE1 = _mm_mul_ps(rinv_SSE1,
2105 _mm_add_ps(_mm_mul_ps(dlij_SSE1, t1_SSE1),
2106 _mm_add_ps(t2_SSE1, t3_SSE1)));
2107 t1_SSE2 = _mm_mul_ps(rinv_SSE2,
2108 _mm_add_ps(_mm_mul_ps(dlij_SSE2, t1_SSE2),
2109 _mm_add_ps(t2_SSE2, t3_SSE2)));
2110 t1_SSE3 = _mm_mul_ps(rinv_SSE3,
2111 _mm_add_ps(_mm_mul_ps(dlij_SSE3, t1_SSE3),
2112 _mm_add_ps(t2_SSE3, t3_SSE3)));
2114 _mm_store_ps(dadx, _mm_and_ps(t1_SSE0, obc_mask1_SSE0));
2116 _mm_store_ps(dadx, _mm_and_ps(t1_SSE1, obc_mask1_SSE1));
2118 _mm_store_ps(dadx, _mm_and_ps(t1_SSE2, obc_mask1_SSE2));
2120 _mm_store_ps(dadx, _mm_and_ps(t1_SSE3, obc_mask1_SSE3));
2125 /* Main part, no exclusions */
2126 for (j = nj1; j < nj2; j += UNROLLJ)
2128 /* load j atom coordinates */
2129 jx_SSE = _mm_load_ps(x_align+j);
2130 jy_SSE = _mm_load_ps(y_align+j);
2131 jz_SSE = _mm_load_ps(z_align+j);
2133 /* Calculate distance */
2134 dx_SSE0 = _mm_sub_ps(ix_SSE0, jx_SSE);
2135 dy_SSE0 = _mm_sub_ps(iy_SSE0, jy_SSE);
2136 dz_SSE0 = _mm_sub_ps(iz_SSE0, jz_SSE);
2137 dx_SSE1 = _mm_sub_ps(ix_SSE1, jx_SSE);
2138 dy_SSE1 = _mm_sub_ps(iy_SSE1, jy_SSE);
2139 dz_SSE1 = _mm_sub_ps(iz_SSE1, jz_SSE);
2140 dx_SSE2 = _mm_sub_ps(ix_SSE2, jx_SSE);
2141 dy_SSE2 = _mm_sub_ps(iy_SSE2, jy_SSE);
2142 dz_SSE2 = _mm_sub_ps(iz_SSE2, jz_SSE);
2143 dx_SSE3 = _mm_sub_ps(ix_SSE3, jx_SSE);
2144 dy_SSE3 = _mm_sub_ps(iy_SSE3, jy_SSE);
2145 dz_SSE3 = _mm_sub_ps(iz_SSE3, jz_SSE);
2147 /* rsq = dx*dx+dy*dy+dz*dz */
2148 rsq_SSE0 = gmx_mm_calc_rsq_ps(dx_SSE0, dy_SSE0, dz_SSE0);
2149 rsq_SSE1 = gmx_mm_calc_rsq_ps(dx_SSE1, dy_SSE1, dz_SSE1);
2150 rsq_SSE2 = gmx_mm_calc_rsq_ps(dx_SSE2, dy_SSE2, dz_SSE2);
2151 rsq_SSE3 = gmx_mm_calc_rsq_ps(dx_SSE3, dy_SSE3, dz_SSE3);
2153 /* Calculate 1/r and 1/r2 */
2154 rinv_SSE0 = gmx_mm_invsqrt_ps(rsq_SSE0);
2155 rinv_SSE1 = gmx_mm_invsqrt_ps(rsq_SSE1);
2156 rinv_SSE2 = gmx_mm_invsqrt_ps(rsq_SSE2);
2157 rinv_SSE3 = gmx_mm_invsqrt_ps(rsq_SSE3);
2160 rinv_SSE0 = _mm_and_ps(rinv_SSE0, imask_SSE0);
2161 rinv_SSE1 = _mm_and_ps(rinv_SSE1, imask_SSE1);
2162 rinv_SSE2 = _mm_and_ps(rinv_SSE2, imask_SSE2);
2163 rinv_SSE3 = _mm_and_ps(rinv_SSE3, imask_SSE3);
2165 dr_SSE0 = _mm_mul_ps(rsq_SSE0, rinv_SSE0);
2166 dr_SSE1 = _mm_mul_ps(rsq_SSE1, rinv_SSE1);
2167 dr_SSE2 = _mm_mul_ps(rsq_SSE2, rinv_SSE2);
2168 dr_SSE3 = _mm_mul_ps(rsq_SSE3, rinv_SSE3);
2170 sk_aj_SSE = _mm_load_ps(obc_param+j);
2171 raj_SSE = _mm_load_ps(gb_radius+j);
2173 raj_inv_SSE = gmx_mm_inv_ps(raj_SSE);
2175 /* Evaluate influence of atom aj -> ai */
2176 t1_SSE0 = _mm_add_ps(dr_SSE0, sk_aj_SSE);
2177 t1_SSE1 = _mm_add_ps(dr_SSE1, sk_aj_SSE);
2178 t1_SSE2 = _mm_add_ps(dr_SSE2, sk_aj_SSE);
2179 t1_SSE3 = _mm_add_ps(dr_SSE3, sk_aj_SSE);
2180 t2_SSE0 = _mm_sub_ps(dr_SSE0, sk_aj_SSE);
2181 t2_SSE1 = _mm_sub_ps(dr_SSE1, sk_aj_SSE);
2182 t2_SSE2 = _mm_sub_ps(dr_SSE2, sk_aj_SSE);
2183 t2_SSE3 = _mm_sub_ps(dr_SSE3, sk_aj_SSE);
2184 t3_SSE0 = _mm_sub_ps(sk_aj_SSE, dr_SSE0);
2185 t3_SSE1 = _mm_sub_ps(sk_aj_SSE, dr_SSE1);
2186 t3_SSE2 = _mm_sub_ps(sk_aj_SSE, dr_SSE2);
2187 t3_SSE3 = _mm_sub_ps(sk_aj_SSE, dr_SSE3);
2189 obc_mask1_SSE0 = _mm_cmplt_ps(rai_SSE0, t1_SSE0);
2190 obc_mask1_SSE1 = _mm_cmplt_ps(rai_SSE1, t1_SSE1);
2191 obc_mask1_SSE2 = _mm_cmplt_ps(rai_SSE2, t1_SSE2);
2192 obc_mask1_SSE3 = _mm_cmplt_ps(rai_SSE3, t1_SSE3);
2193 obc_mask2_SSE0 = _mm_cmplt_ps(rai_SSE0, t2_SSE0);
2194 obc_mask2_SSE1 = _mm_cmplt_ps(rai_SSE1, t2_SSE1);
2195 obc_mask2_SSE2 = _mm_cmplt_ps(rai_SSE2, t2_SSE2);
2196 obc_mask2_SSE3 = _mm_cmplt_ps(rai_SSE3, t2_SSE3);
2197 obc_mask3_SSE0 = _mm_cmplt_ps(rai_SSE0, t3_SSE0);
2198 obc_mask3_SSE1 = _mm_cmplt_ps(rai_SSE1, t3_SSE1);
2199 obc_mask3_SSE2 = _mm_cmplt_ps(rai_SSE2, t3_SSE2);
2200 obc_mask3_SSE3 = _mm_cmplt_ps(rai_SSE3, t3_SSE3);
2201 obc_mask1_SSE0 = _mm_and_ps(obc_mask1_SSE0, imask_SSE0);
2202 obc_mask1_SSE1 = _mm_and_ps(obc_mask1_SSE1, imask_SSE1);
2203 obc_mask1_SSE2 = _mm_and_ps(obc_mask1_SSE2, imask_SSE2);
2204 obc_mask1_SSE3 = _mm_and_ps(obc_mask1_SSE3, imask_SSE3);
2206 uij_SSE0 = gmx_mm_inv_ps(t1_SSE0);
2207 uij_SSE1 = gmx_mm_inv_ps(t1_SSE1);
2208 uij_SSE2 = gmx_mm_inv_ps(t1_SSE2);
2209 uij_SSE3 = gmx_mm_inv_ps(t1_SSE3);
2210 lij_SSE0 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE0, gmx_mm_inv_ps(t2_SSE0)),
2211 _mm_andnot_ps(obc_mask2_SSE0, rai_inv_SSE0));
2212 lij_SSE1 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE1, gmx_mm_inv_ps(t2_SSE1)),
2213 _mm_andnot_ps(obc_mask2_SSE1, rai_inv_SSE1));
2214 lij_SSE2 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE2, gmx_mm_inv_ps(t2_SSE2)),
2215 _mm_andnot_ps(obc_mask2_SSE2, rai_inv_SSE2));
2216 lij_SSE3 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE3, gmx_mm_inv_ps(t2_SSE3)),
2217 _mm_andnot_ps(obc_mask2_SSE3, rai_inv_SSE3));
2218 dlij_SSE0 = _mm_and_ps(one_SSE, obc_mask2_SSE0);
2219 dlij_SSE1 = _mm_and_ps(one_SSE, obc_mask2_SSE1);
2220 dlij_SSE2 = _mm_and_ps(one_SSE, obc_mask2_SSE2);
2221 dlij_SSE3 = _mm_and_ps(one_SSE, obc_mask2_SSE3);
2223 uij2_SSE0 = _mm_mul_ps(uij_SSE0, uij_SSE0);
2224 uij2_SSE1 = _mm_mul_ps(uij_SSE1, uij_SSE1);
2225 uij2_SSE2 = _mm_mul_ps(uij_SSE2, uij_SSE2);
2226 uij2_SSE3 = _mm_mul_ps(uij_SSE3, uij_SSE3);
2227 uij3_SSE0 = _mm_mul_ps(uij2_SSE0, uij_SSE0);
2228 uij3_SSE1 = _mm_mul_ps(uij2_SSE1, uij_SSE1);
2229 uij3_SSE2 = _mm_mul_ps(uij2_SSE2, uij_SSE2);
2230 uij3_SSE3 = _mm_mul_ps(uij2_SSE3, uij_SSE3);
2231 lij2_SSE0 = _mm_mul_ps(lij_SSE0, lij_SSE0);
2232 lij2_SSE1 = _mm_mul_ps(lij_SSE1, lij_SSE1);
2233 lij2_SSE2 = _mm_mul_ps(lij_SSE2, lij_SSE2);
2234 lij2_SSE3 = _mm_mul_ps(lij_SSE3, lij_SSE3);
2235 lij3_SSE0 = _mm_mul_ps(lij2_SSE0, lij_SSE0);
2236 lij3_SSE1 = _mm_mul_ps(lij2_SSE1, lij_SSE1);
2237 lij3_SSE2 = _mm_mul_ps(lij2_SSE2, lij_SSE2);
2238 lij3_SSE3 = _mm_mul_ps(lij2_SSE3, lij_SSE3);
2240 diff2_SSE0 = _mm_sub_ps(uij2_SSE0, lij2_SSE0);
2241 diff2_SSE1 = _mm_sub_ps(uij2_SSE1, lij2_SSE1);
2242 diff2_SSE2 = _mm_sub_ps(uij2_SSE2, lij2_SSE2);
2243 diff2_SSE3 = _mm_sub_ps(uij2_SSE3, lij2_SSE3);
2244 lij_inv_SSE0 = gmx_mm_invsqrt_ps(lij2_SSE0);
2245 lij_inv_SSE1 = gmx_mm_invsqrt_ps(lij2_SSE1);
2246 lij_inv_SSE2 = gmx_mm_invsqrt_ps(lij2_SSE2);
2247 lij_inv_SSE3 = gmx_mm_invsqrt_ps(lij2_SSE3);
2248 sk2_aj_SSE = _mm_mul_ps(sk_aj_SSE, sk_aj_SSE);
2249 sk2_rinv_SSE0 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE0);
2250 sk2_rinv_SSE1 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE1);
2251 sk2_rinv_SSE2 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE2);
2252 sk2_rinv_SSE3 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE3);
2253 prod_SSE0 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE0);
2254 prod_SSE1 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE1);
2255 prod_SSE2 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE2);
2256 prod_SSE3 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE3);
2258 logterm_SSE0 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE0, lij_inv_SSE0));
2259 logterm_SSE1 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE1, lij_inv_SSE1));
2260 logterm_SSE2 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE2, lij_inv_SSE2));
2261 logterm_SSE3 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE3, lij_inv_SSE3));
2263 t1_SSE0 = _mm_sub_ps(lij_SSE0, uij_SSE0);
2264 t1_SSE1 = _mm_sub_ps(lij_SSE1, uij_SSE1);
2265 t1_SSE2 = _mm_sub_ps(lij_SSE2, uij_SSE2);
2266 t1_SSE3 = _mm_sub_ps(lij_SSE3, uij_SSE3);
2267 t2_SSE0 = _mm_mul_ps(diff2_SSE0,
2268 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE0),
2270 t2_SSE1 = _mm_mul_ps(diff2_SSE1,
2271 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE1),
2273 t2_SSE2 = _mm_mul_ps(diff2_SSE2,
2274 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE2),
2276 t2_SSE3 = _mm_mul_ps(diff2_SSE3,
2277 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE3),
2280 t3_SSE0 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE0, logterm_SSE0));
2281 t3_SSE1 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE1, logterm_SSE1));
2282 t3_SSE2 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE2, logterm_SSE2));
2283 t3_SSE3 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE3, logterm_SSE3));
2284 t1_SSE0 = _mm_add_ps(t1_SSE0, _mm_add_ps(t2_SSE0, t3_SSE0));
2285 t1_SSE1 = _mm_add_ps(t1_SSE1, _mm_add_ps(t2_SSE1, t3_SSE1));
2286 t1_SSE2 = _mm_add_ps(t1_SSE2, _mm_add_ps(t2_SSE2, t3_SSE2));
2287 t1_SSE3 = _mm_add_ps(t1_SSE3, _mm_add_ps(t2_SSE3, t3_SSE3));
2288 t4_SSE0 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE0, lij_SSE0));
2289 t4_SSE1 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE1, lij_SSE1));
2290 t4_SSE2 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE2, lij_SSE2));
2291 t4_SSE3 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE3, lij_SSE3));
2292 t4_SSE0 = _mm_and_ps(t4_SSE0, obc_mask3_SSE0);
2293 t4_SSE1 = _mm_and_ps(t4_SSE1, obc_mask3_SSE1);
2294 t4_SSE2 = _mm_and_ps(t4_SSE2, obc_mask3_SSE2);
2295 t4_SSE3 = _mm_and_ps(t4_SSE3, obc_mask3_SSE3);
2296 t1_SSE0 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE0, t4_SSE0));
2297 t1_SSE1 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE1, t4_SSE1));
2298 t1_SSE2 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE2, t4_SSE2));
2299 t1_SSE3 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE3, t4_SSE3));
2301 sum_ai_SSE0 = _mm_add_ps(sum_ai_SSE0, _mm_and_ps(t1_SSE0, obc_mask1_SSE0));
2302 sum_ai_SSE1 = _mm_add_ps(sum_ai_SSE1, _mm_and_ps(t1_SSE1, obc_mask1_SSE1));
2303 sum_ai_SSE2 = _mm_add_ps(sum_ai_SSE2, _mm_and_ps(t1_SSE2, obc_mask1_SSE2));
2304 sum_ai_SSE3 = _mm_add_ps(sum_ai_SSE3, _mm_and_ps(t1_SSE3, obc_mask1_SSE3));
2306 t1_SSE0 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE0),
2307 _mm_mul_ps(prod_SSE0, lij3_SSE0));
2308 t1_SSE1 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE1),
2309 _mm_mul_ps(prod_SSE1, lij3_SSE1));
2310 t1_SSE2 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE2),
2311 _mm_mul_ps(prod_SSE2, lij3_SSE2));
2312 t1_SSE3 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE3),
2313 _mm_mul_ps(prod_SSE3, lij3_SSE3));
2314 t1_SSE0 = _mm_sub_ps(t1_SSE0,
2315 _mm_mul_ps(onefourth_SSE,
2316 _mm_add_ps(_mm_mul_ps(lij_SSE0, rinv_SSE0),
2317 _mm_mul_ps(lij3_SSE0, dr_SSE0))));
2318 t1_SSE1 = _mm_sub_ps(t1_SSE1,
2319 _mm_mul_ps(onefourth_SSE,
2320 _mm_add_ps(_mm_mul_ps(lij_SSE1, rinv_SSE1),
2321 _mm_mul_ps(lij3_SSE1, dr_SSE1))));
2322 t1_SSE2 = _mm_sub_ps(t1_SSE2,
2323 _mm_mul_ps(onefourth_SSE,
2324 _mm_add_ps(_mm_mul_ps(lij_SSE2, rinv_SSE2),
2325 _mm_mul_ps(lij3_SSE2, dr_SSE2))));
2326 t1_SSE3 = _mm_sub_ps(t1_SSE3,
2327 _mm_mul_ps(onefourth_SSE,
2328 _mm_add_ps(_mm_mul_ps(lij_SSE3, rinv_SSE3),
2329 _mm_mul_ps(lij3_SSE3, dr_SSE3))));
2331 t2_SSE0 = _mm_mul_ps(onefourth_SSE,
2332 _mm_add_ps(_mm_mul_ps(uij_SSE0, rinv_SSE0),
2333 _mm_mul_ps(uij3_SSE0, dr_SSE0)));
2334 t2_SSE1 = _mm_mul_ps(onefourth_SSE,
2335 _mm_add_ps(_mm_mul_ps(uij_SSE1, rinv_SSE1),
2336 _mm_mul_ps(uij3_SSE1, dr_SSE1)));
2337 t2_SSE2 = _mm_mul_ps(onefourth_SSE,
2338 _mm_add_ps(_mm_mul_ps(uij_SSE2, rinv_SSE2),
2339 _mm_mul_ps(uij3_SSE2, dr_SSE2)));
2340 t2_SSE3 = _mm_mul_ps(onefourth_SSE,
2341 _mm_add_ps(_mm_mul_ps(uij_SSE3, rinv_SSE3),
2342 _mm_mul_ps(uij3_SSE3, dr_SSE3)));
2343 t2_SSE0 = _mm_sub_ps(t2_SSE0,
2344 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE0),
2345 _mm_mul_ps(prod_SSE0, uij3_SSE0)));
2346 t2_SSE1 = _mm_sub_ps(t2_SSE1,
2347 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE1),
2348 _mm_mul_ps(prod_SSE1, uij3_SSE1)));
2349 t2_SSE2 = _mm_sub_ps(t2_SSE2,
2350 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE2),
2351 _mm_mul_ps(prod_SSE2, uij3_SSE2)));
2352 t2_SSE3 = _mm_sub_ps(t2_SSE3,
2353 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE3),
2354 _mm_mul_ps(prod_SSE3, uij3_SSE3)));
2355 t3_SSE0 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE0),
2356 _mm_mul_ps(rinv_SSE0, rinv_SSE0));
2357 t3_SSE1 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE1),
2358 _mm_mul_ps(rinv_SSE1, rinv_SSE1));
2359 t3_SSE2 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE2),
2360 _mm_mul_ps(rinv_SSE2, rinv_SSE2));
2361 t3_SSE3 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE3),
2362 _mm_mul_ps(rinv_SSE3, rinv_SSE3));
2363 t3_SSE0 = _mm_sub_ps(t3_SSE0,
2364 _mm_mul_ps(_mm_mul_ps(diff2_SSE0, oneeighth_SSE),
2366 _mm_mul_ps(sk2_rinv_SSE0, rinv_SSE0))));
2367 t3_SSE1 = _mm_sub_ps(t3_SSE1,
2368 _mm_mul_ps(_mm_mul_ps(diff2_SSE1, oneeighth_SSE),
2370 _mm_mul_ps(sk2_rinv_SSE1, rinv_SSE1))));
2371 t3_SSE2 = _mm_sub_ps(t3_SSE2,
2372 _mm_mul_ps(_mm_mul_ps(diff2_SSE2, oneeighth_SSE),
2374 _mm_mul_ps(sk2_rinv_SSE2, rinv_SSE2))));
2375 t3_SSE3 = _mm_sub_ps(t3_SSE3,
2376 _mm_mul_ps(_mm_mul_ps(diff2_SSE3, oneeighth_SSE),
2378 _mm_mul_ps(sk2_rinv_SSE3, rinv_SSE3))));
2380 t1_SSE0 = _mm_mul_ps(rinv_SSE0,
2381 _mm_add_ps(_mm_mul_ps(dlij_SSE0, t1_SSE0),
2382 _mm_add_ps(t2_SSE0, t3_SSE0)));
2383 t1_SSE1 = _mm_mul_ps(rinv_SSE1,
2384 _mm_add_ps(_mm_mul_ps(dlij_SSE1, t1_SSE1),
2385 _mm_add_ps(t2_SSE1, t3_SSE1)));
2386 t1_SSE2 = _mm_mul_ps(rinv_SSE2,
2387 _mm_add_ps(_mm_mul_ps(dlij_SSE2, t1_SSE2),
2388 _mm_add_ps(t2_SSE2, t3_SSE2)));
2389 t1_SSE3 = _mm_mul_ps(rinv_SSE3,
2390 _mm_add_ps(_mm_mul_ps(dlij_SSE3, t1_SSE3),
2391 _mm_add_ps(t2_SSE3, t3_SSE3)));
2393 _mm_store_ps(dadx, _mm_and_ps(t1_SSE0, obc_mask1_SSE0));
2395 _mm_store_ps(dadx, _mm_and_ps(t1_SSE1, obc_mask1_SSE1));
2397 _mm_store_ps(dadx, _mm_and_ps(t1_SSE2, obc_mask1_SSE2));
2399 _mm_store_ps(dadx, _mm_and_ps(t1_SSE3, obc_mask1_SSE3));
2402 /* Evaluate influence of atom ai -> aj */
2403 t1_SSE0 = _mm_add_ps(dr_SSE0, sk_ai_SSE0);
2404 t1_SSE1 = _mm_add_ps(dr_SSE1, sk_ai_SSE1);
2405 t1_SSE2 = _mm_add_ps(dr_SSE2, sk_ai_SSE2);
2406 t1_SSE3 = _mm_add_ps(dr_SSE3, sk_ai_SSE3);
2407 t2_SSE0 = _mm_sub_ps(dr_SSE0, sk_ai_SSE0);
2408 t2_SSE1 = _mm_sub_ps(dr_SSE1, sk_ai_SSE1);
2409 t2_SSE2 = _mm_sub_ps(dr_SSE2, sk_ai_SSE2);
2410 t2_SSE3 = _mm_sub_ps(dr_SSE3, sk_ai_SSE3);
2411 t3_SSE0 = _mm_sub_ps(sk_ai_SSE0, dr_SSE0);
2412 t3_SSE1 = _mm_sub_ps(sk_ai_SSE1, dr_SSE1);
2413 t3_SSE2 = _mm_sub_ps(sk_ai_SSE2, dr_SSE2);
2414 t3_SSE3 = _mm_sub_ps(sk_ai_SSE3, dr_SSE3);
2416 obc_mask1_SSE0 = _mm_cmplt_ps(raj_SSE, t1_SSE0);
2417 obc_mask1_SSE1 = _mm_cmplt_ps(raj_SSE, t1_SSE1);
2418 obc_mask1_SSE2 = _mm_cmplt_ps(raj_SSE, t1_SSE2);
2419 obc_mask1_SSE3 = _mm_cmplt_ps(raj_SSE, t1_SSE3);
2420 obc_mask2_SSE0 = _mm_cmplt_ps(raj_SSE, t2_SSE0);
2421 obc_mask2_SSE1 = _mm_cmplt_ps(raj_SSE, t2_SSE1);
2422 obc_mask2_SSE2 = _mm_cmplt_ps(raj_SSE, t2_SSE2);
2423 obc_mask2_SSE3 = _mm_cmplt_ps(raj_SSE, t2_SSE3);
2424 obc_mask3_SSE0 = _mm_cmplt_ps(raj_SSE, t3_SSE0);
2425 obc_mask3_SSE1 = _mm_cmplt_ps(raj_SSE, t3_SSE1);
2426 obc_mask3_SSE2 = _mm_cmplt_ps(raj_SSE, t3_SSE2);
2427 obc_mask3_SSE3 = _mm_cmplt_ps(raj_SSE, t3_SSE3);
2428 obc_mask1_SSE0 = _mm_and_ps(obc_mask1_SSE0, imask_SSE0);
2429 obc_mask1_SSE1 = _mm_and_ps(obc_mask1_SSE1, imask_SSE1);
2430 obc_mask1_SSE2 = _mm_and_ps(obc_mask1_SSE2, imask_SSE2);
2431 obc_mask1_SSE3 = _mm_and_ps(obc_mask1_SSE3, imask_SSE3);
2433 uij_SSE0 = gmx_mm_inv_ps(t1_SSE0);
2434 uij_SSE1 = gmx_mm_inv_ps(t1_SSE1);
2435 uij_SSE2 = gmx_mm_inv_ps(t1_SSE2);
2436 uij_SSE3 = gmx_mm_inv_ps(t1_SSE3);
2437 lij_SSE0 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE0, gmx_mm_inv_ps(t2_SSE0)),
2438 _mm_andnot_ps(obc_mask2_SSE0, raj_inv_SSE));
2439 lij_SSE1 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE1, gmx_mm_inv_ps(t2_SSE1)),
2440 _mm_andnot_ps(obc_mask2_SSE1, raj_inv_SSE));
2441 lij_SSE2 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE2, gmx_mm_inv_ps(t2_SSE2)),
2442 _mm_andnot_ps(obc_mask2_SSE2, raj_inv_SSE));
2443 lij_SSE3 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE3, gmx_mm_inv_ps(t2_SSE3)),
2444 _mm_andnot_ps(obc_mask2_SSE3, raj_inv_SSE));
2445 dlij_SSE0 = _mm_and_ps(one_SSE, obc_mask2_SSE0);
2446 dlij_SSE1 = _mm_and_ps(one_SSE, obc_mask2_SSE1);
2447 dlij_SSE2 = _mm_and_ps(one_SSE, obc_mask2_SSE2);
2448 dlij_SSE3 = _mm_and_ps(one_SSE, obc_mask2_SSE3);
2450 uij2_SSE0 = _mm_mul_ps(uij_SSE0, uij_SSE0);
2451 uij2_SSE1 = _mm_mul_ps(uij_SSE1, uij_SSE1);
2452 uij2_SSE2 = _mm_mul_ps(uij_SSE2, uij_SSE2);
2453 uij2_SSE3 = _mm_mul_ps(uij_SSE3, uij_SSE3);
2454 uij3_SSE0 = _mm_mul_ps(uij2_SSE0, uij_SSE0);
2455 uij3_SSE1 = _mm_mul_ps(uij2_SSE1, uij_SSE1);
2456 uij3_SSE2 = _mm_mul_ps(uij2_SSE2, uij_SSE2);
2457 uij3_SSE3 = _mm_mul_ps(uij2_SSE3, uij_SSE3);
2458 lij2_SSE0 = _mm_mul_ps(lij_SSE0, lij_SSE0);
2459 lij2_SSE1 = _mm_mul_ps(lij_SSE1, lij_SSE1);
2460 lij2_SSE2 = _mm_mul_ps(lij_SSE2, lij_SSE2);
2461 lij2_SSE3 = _mm_mul_ps(lij_SSE3, lij_SSE3);
2462 lij3_SSE0 = _mm_mul_ps(lij2_SSE0, lij_SSE0);
2463 lij3_SSE1 = _mm_mul_ps(lij2_SSE1, lij_SSE1);
2464 lij3_SSE2 = _mm_mul_ps(lij2_SSE2, lij_SSE2);
2465 lij3_SSE3 = _mm_mul_ps(lij2_SSE3, lij_SSE3);
2467 diff2_SSE0 = _mm_sub_ps(uij2_SSE0, lij2_SSE0);
2468 diff2_SSE1 = _mm_sub_ps(uij2_SSE1, lij2_SSE1);
2469 diff2_SSE2 = _mm_sub_ps(uij2_SSE2, lij2_SSE2);
2470 diff2_SSE3 = _mm_sub_ps(uij2_SSE3, lij2_SSE3);
2471 lij_inv_SSE0 = gmx_mm_invsqrt_ps(lij2_SSE0);
2472 lij_inv_SSE1 = gmx_mm_invsqrt_ps(lij2_SSE1);
2473 lij_inv_SSE2 = gmx_mm_invsqrt_ps(lij2_SSE2);
2474 lij_inv_SSE3 = gmx_mm_invsqrt_ps(lij2_SSE3);
2475 sk2_rinv_SSE0 = _mm_mul_ps(sk2_ai_SSE0, rinv_SSE0);
2476 sk2_rinv_SSE1 = _mm_mul_ps(sk2_ai_SSE1, rinv_SSE1);
2477 sk2_rinv_SSE2 = _mm_mul_ps(sk2_ai_SSE2, rinv_SSE2);
2478 sk2_rinv_SSE3 = _mm_mul_ps(sk2_ai_SSE3, rinv_SSE3);
2479 prod_SSE0 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE0);
2480 prod_SSE1 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE1);
2481 prod_SSE2 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE2);
2482 prod_SSE3 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE3);
2484 logterm_SSE0 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE0, lij_inv_SSE0));
2485 logterm_SSE1 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE1, lij_inv_SSE1));
2486 logterm_SSE2 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE2, lij_inv_SSE2));
2487 logterm_SSE3 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE3, lij_inv_SSE3));
2488 t1_SSE0 = _mm_sub_ps(lij_SSE0, uij_SSE0);
2489 t1_SSE1 = _mm_sub_ps(lij_SSE1, uij_SSE1);
2490 t1_SSE2 = _mm_sub_ps(lij_SSE2, uij_SSE2);
2491 t1_SSE3 = _mm_sub_ps(lij_SSE3, uij_SSE3);
2492 t2_SSE0 = _mm_mul_ps(diff2_SSE0,
2493 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE0),
2495 t2_SSE1 = _mm_mul_ps(diff2_SSE1,
2496 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE1),
2498 t2_SSE2 = _mm_mul_ps(diff2_SSE2,
2499 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE2),
2501 t2_SSE3 = _mm_mul_ps(diff2_SSE3,
2502 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE3),
2504 t3_SSE0 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE0, logterm_SSE0));
2505 t3_SSE1 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE1, logterm_SSE1));
2506 t3_SSE2 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE2, logterm_SSE2));
2507 t3_SSE3 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE3, logterm_SSE3));
2508 t1_SSE0 = _mm_add_ps(t1_SSE0, _mm_add_ps(t2_SSE0, t3_SSE0));
2509 t1_SSE1 = _mm_add_ps(t1_SSE1, _mm_add_ps(t2_SSE1, t3_SSE1));
2510 t1_SSE2 = _mm_add_ps(t1_SSE2, _mm_add_ps(t2_SSE2, t3_SSE2));
2511 t1_SSE3 = _mm_add_ps(t1_SSE3, _mm_add_ps(t2_SSE3, t3_SSE3));
2512 t4_SSE0 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE0));
2513 t4_SSE1 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE1));
2514 t4_SSE2 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE2));
2515 t4_SSE3 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE3));
2516 t4_SSE0 = _mm_and_ps(t4_SSE0, obc_mask3_SSE0);
2517 t4_SSE1 = _mm_and_ps(t4_SSE1, obc_mask3_SSE1);
2518 t4_SSE2 = _mm_and_ps(t4_SSE2, obc_mask3_SSE2);
2519 t4_SSE3 = _mm_and_ps(t4_SSE3, obc_mask3_SSE3);
2520 t1_SSE0 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE0, t4_SSE0));
2521 t1_SSE1 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE1, t4_SSE1));
2522 t1_SSE2 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE2, t4_SSE2));
2523 t1_SSE3 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE3, t4_SSE3));
2525 _mm_store_ps(work+j, _mm_add_ps(_mm_load_ps(work+j),
2526 gmx_mm_sum4_ps(_mm_and_ps(t1_SSE0, obc_mask1_SSE0),
2527 _mm_and_ps(t1_SSE1, obc_mask1_SSE1),
2528 _mm_and_ps(t1_SSE2, obc_mask1_SSE2),
2529 _mm_and_ps(t1_SSE3, obc_mask1_SSE3))));
2531 t1_SSE0 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE0),
2532 _mm_mul_ps(prod_SSE0, lij3_SSE0));
2533 t1_SSE1 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE1),
2534 _mm_mul_ps(prod_SSE1, lij3_SSE1));
2535 t1_SSE2 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE2),
2536 _mm_mul_ps(prod_SSE2, lij3_SSE2));
2537 t1_SSE3 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE3),
2538 _mm_mul_ps(prod_SSE3, lij3_SSE3));
2539 t1_SSE0 = _mm_sub_ps(t1_SSE0,
2540 _mm_mul_ps(onefourth_SSE,
2541 _mm_add_ps(_mm_mul_ps(lij_SSE0, rinv_SSE0),
2542 _mm_mul_ps(lij3_SSE0, dr_SSE0))));
2543 t1_SSE1 = _mm_sub_ps(t1_SSE1,
2544 _mm_mul_ps(onefourth_SSE,
2545 _mm_add_ps(_mm_mul_ps(lij_SSE1, rinv_SSE1),
2546 _mm_mul_ps(lij3_SSE1, dr_SSE1))));
2547 t1_SSE2 = _mm_sub_ps(t1_SSE2,
2548 _mm_mul_ps(onefourth_SSE,
2549 _mm_add_ps(_mm_mul_ps(lij_SSE2, rinv_SSE2),
2550 _mm_mul_ps(lij3_SSE2, dr_SSE2))));
2551 t1_SSE3 = _mm_sub_ps(t1_SSE3,
2552 _mm_mul_ps(onefourth_SSE,
2553 _mm_add_ps(_mm_mul_ps(lij_SSE3, rinv_SSE3),
2554 _mm_mul_ps(lij3_SSE3, dr_SSE3))));
2555 t2_SSE0 = _mm_mul_ps(onefourth_SSE,
2556 _mm_add_ps(_mm_mul_ps(uij_SSE0, rinv_SSE0),
2557 _mm_mul_ps(uij3_SSE0, dr_SSE0)));
2558 t2_SSE1 = _mm_mul_ps(onefourth_SSE,
2559 _mm_add_ps(_mm_mul_ps(uij_SSE1, rinv_SSE1),
2560 _mm_mul_ps(uij3_SSE1, dr_SSE1)));
2561 t2_SSE2 = _mm_mul_ps(onefourth_SSE,
2562 _mm_add_ps(_mm_mul_ps(uij_SSE2, rinv_SSE2),
2563 _mm_mul_ps(uij3_SSE2, dr_SSE2)));
2564 t2_SSE3 = _mm_mul_ps(onefourth_SSE,
2565 _mm_add_ps(_mm_mul_ps(uij_SSE3, rinv_SSE3),
2566 _mm_mul_ps(uij3_SSE3, dr_SSE3)));
2567 t2_SSE0 = _mm_sub_ps(t2_SSE0,
2568 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE0),
2569 _mm_mul_ps(prod_SSE0, uij3_SSE0)));
2570 t2_SSE1 = _mm_sub_ps(t2_SSE1,
2571 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE1),
2572 _mm_mul_ps(prod_SSE1, uij3_SSE1)));
2573 t2_SSE2 = _mm_sub_ps(t2_SSE2,
2574 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE2),
2575 _mm_mul_ps(prod_SSE2, uij3_SSE2)));
2576 t2_SSE3 = _mm_sub_ps(t2_SSE3,
2577 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE3),
2578 _mm_mul_ps(prod_SSE3, uij3_SSE3)));
2580 t3_SSE0 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE0),
2581 _mm_mul_ps(rinv_SSE0, rinv_SSE0));
2582 t3_SSE1 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE1),
2583 _mm_mul_ps(rinv_SSE1, rinv_SSE1));
2584 t3_SSE2 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE2),
2585 _mm_mul_ps(rinv_SSE2, rinv_SSE2));
2586 t3_SSE3 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE3),
2587 _mm_mul_ps(rinv_SSE3, rinv_SSE3));
2589 t3_SSE0 = _mm_sub_ps(t3_SSE0,
2590 _mm_mul_ps(_mm_mul_ps(diff2_SSE0, oneeighth_SSE),
2592 _mm_mul_ps(sk2_rinv_SSE0, rinv_SSE0))));
2593 t3_SSE1 = _mm_sub_ps(t3_SSE1,
2594 _mm_mul_ps(_mm_mul_ps(diff2_SSE1, oneeighth_SSE),
2596 _mm_mul_ps(sk2_rinv_SSE1, rinv_SSE1))));
2597 t3_SSE2 = _mm_sub_ps(t3_SSE2,
2598 _mm_mul_ps(_mm_mul_ps(diff2_SSE2, oneeighth_SSE),
2600 _mm_mul_ps(sk2_rinv_SSE2, rinv_SSE2))));
2601 t3_SSE3 = _mm_sub_ps(t3_SSE3,
2602 _mm_mul_ps(_mm_mul_ps(diff2_SSE3, oneeighth_SSE),
2604 _mm_mul_ps(sk2_rinv_SSE3, rinv_SSE3))));
2606 t1_SSE0 = _mm_mul_ps(rinv_SSE0,
2607 _mm_add_ps(_mm_mul_ps(dlij_SSE0, t1_SSE0),
2608 _mm_add_ps(t2_SSE0, t3_SSE0)));
2609 t1_SSE1 = _mm_mul_ps(rinv_SSE1,
2610 _mm_add_ps(_mm_mul_ps(dlij_SSE1, t1_SSE1),
2611 _mm_add_ps(t2_SSE1, t3_SSE1)));
2612 t1_SSE2 = _mm_mul_ps(rinv_SSE2,
2613 _mm_add_ps(_mm_mul_ps(dlij_SSE2, t1_SSE2),
2614 _mm_add_ps(t2_SSE2, t3_SSE2)));
2615 t1_SSE3 = _mm_mul_ps(rinv_SSE3,
2616 _mm_add_ps(_mm_mul_ps(dlij_SSE3, t1_SSE3),
2617 _mm_add_ps(t2_SSE3, t3_SSE3)));
2619 _mm_store_ps(dadx, _mm_and_ps(t1_SSE0, obc_mask1_SSE0));
2621 _mm_store_ps(dadx, _mm_and_ps(t1_SSE1, obc_mask1_SSE1));
2623 _mm_store_ps(dadx, _mm_and_ps(t1_SSE2, obc_mask1_SSE2));
2625 _mm_store_ps(dadx, _mm_and_ps(t1_SSE3, obc_mask1_SSE3));
2629 /* Epilogue part, including exclusion mask */
2630 for (j = nj2; j < nj3; j += UNROLLJ)
2632 jmask_SSE0 = _mm_load_ps((real *)emask0);
2633 jmask_SSE1 = _mm_load_ps((real *)emask1);
2634 jmask_SSE2 = _mm_load_ps((real *)emask2);
2635 jmask_SSE3 = _mm_load_ps((real *)emask3);
2641 /* load j atom coordinates */
2642 jx_SSE = _mm_load_ps(x_align+j);
2643 jy_SSE = _mm_load_ps(y_align+j);
2644 jz_SSE = _mm_load_ps(z_align+j);
2646 /* Calculate distance */
2647 dx_SSE0 = _mm_sub_ps(ix_SSE0, jx_SSE);
2648 dy_SSE0 = _mm_sub_ps(iy_SSE0, jy_SSE);
2649 dz_SSE0 = _mm_sub_ps(iz_SSE0, jz_SSE);
2650 dx_SSE1 = _mm_sub_ps(ix_SSE1, jx_SSE);
2651 dy_SSE1 = _mm_sub_ps(iy_SSE1, jy_SSE);
2652 dz_SSE1 = _mm_sub_ps(iz_SSE1, jz_SSE);
2653 dx_SSE2 = _mm_sub_ps(ix_SSE2, jx_SSE);
2654 dy_SSE2 = _mm_sub_ps(iy_SSE2, jy_SSE);
2655 dz_SSE2 = _mm_sub_ps(iz_SSE2, jz_SSE);
2656 dx_SSE3 = _mm_sub_ps(ix_SSE3, jx_SSE);
2657 dy_SSE3 = _mm_sub_ps(iy_SSE3, jy_SSE);
2658 dz_SSE3 = _mm_sub_ps(iz_SSE3, jz_SSE);
2660 /* rsq = dx*dx+dy*dy+dz*dz */
2661 rsq_SSE0 = gmx_mm_calc_rsq_ps(dx_SSE0, dy_SSE0, dz_SSE0);
2662 rsq_SSE1 = gmx_mm_calc_rsq_ps(dx_SSE1, dy_SSE1, dz_SSE1);
2663 rsq_SSE2 = gmx_mm_calc_rsq_ps(dx_SSE2, dy_SSE2, dz_SSE2);
2664 rsq_SSE3 = gmx_mm_calc_rsq_ps(dx_SSE3, dy_SSE3, dz_SSE3);
2667 jmask_SSE0 = _mm_and_ps(jmask_SSE0, imask_SSE0);
2668 jmask_SSE1 = _mm_and_ps(jmask_SSE1, imask_SSE1);
2669 jmask_SSE2 = _mm_and_ps(jmask_SSE2, imask_SSE2);
2670 jmask_SSE3 = _mm_and_ps(jmask_SSE3, imask_SSE3);
2672 /* Calculate 1/r and 1/r2 */
2673 rinv_SSE0 = gmx_mm_invsqrt_ps(rsq_SSE0);
2674 rinv_SSE1 = gmx_mm_invsqrt_ps(rsq_SSE1);
2675 rinv_SSE2 = gmx_mm_invsqrt_ps(rsq_SSE2);
2676 rinv_SSE3 = gmx_mm_invsqrt_ps(rsq_SSE3);
2679 rinv_SSE0 = _mm_and_ps(rinv_SSE0, jmask_SSE0);
2680 rinv_SSE1 = _mm_and_ps(rinv_SSE1, jmask_SSE1);
2681 rinv_SSE2 = _mm_and_ps(rinv_SSE2, jmask_SSE2);
2682 rinv_SSE3 = _mm_and_ps(rinv_SSE3, jmask_SSE3);
2684 dr_SSE0 = _mm_mul_ps(rsq_SSE0, rinv_SSE0);
2685 dr_SSE1 = _mm_mul_ps(rsq_SSE1, rinv_SSE1);
2686 dr_SSE2 = _mm_mul_ps(rsq_SSE2, rinv_SSE2);
2687 dr_SSE3 = _mm_mul_ps(rsq_SSE3, rinv_SSE3);
2689 sk_aj_SSE = _mm_load_ps(obc_param+j);
2690 raj_SSE = _mm_load_ps(gb_radius+j);
2692 raj_inv_SSE = gmx_mm_inv_ps(raj_SSE);
2694 /* Evaluate influence of atom aj -> ai */
2695 t1_SSE0 = _mm_add_ps(dr_SSE0, sk_aj_SSE);
2696 t1_SSE1 = _mm_add_ps(dr_SSE1, sk_aj_SSE);
2697 t1_SSE2 = _mm_add_ps(dr_SSE2, sk_aj_SSE);
2698 t1_SSE3 = _mm_add_ps(dr_SSE3, sk_aj_SSE);
2699 t2_SSE0 = _mm_sub_ps(dr_SSE0, sk_aj_SSE);
2700 t2_SSE1 = _mm_sub_ps(dr_SSE1, sk_aj_SSE);
2701 t2_SSE2 = _mm_sub_ps(dr_SSE2, sk_aj_SSE);
2702 t2_SSE3 = _mm_sub_ps(dr_SSE3, sk_aj_SSE);
2703 t3_SSE0 = _mm_sub_ps(sk_aj_SSE, dr_SSE0);
2704 t3_SSE1 = _mm_sub_ps(sk_aj_SSE, dr_SSE1);
2705 t3_SSE2 = _mm_sub_ps(sk_aj_SSE, dr_SSE2);
2706 t3_SSE3 = _mm_sub_ps(sk_aj_SSE, dr_SSE3);
2708 obc_mask1_SSE0 = _mm_cmplt_ps(rai_SSE0, t1_SSE0);
2709 obc_mask1_SSE1 = _mm_cmplt_ps(rai_SSE1, t1_SSE1);
2710 obc_mask1_SSE2 = _mm_cmplt_ps(rai_SSE2, t1_SSE2);
2711 obc_mask1_SSE3 = _mm_cmplt_ps(rai_SSE3, t1_SSE3);
2712 obc_mask2_SSE0 = _mm_cmplt_ps(rai_SSE0, t2_SSE0);
2713 obc_mask2_SSE1 = _mm_cmplt_ps(rai_SSE1, t2_SSE1);
2714 obc_mask2_SSE2 = _mm_cmplt_ps(rai_SSE2, t2_SSE2);
2715 obc_mask2_SSE3 = _mm_cmplt_ps(rai_SSE3, t2_SSE3);
2716 obc_mask3_SSE0 = _mm_cmplt_ps(rai_SSE0, t3_SSE0);
2717 obc_mask3_SSE1 = _mm_cmplt_ps(rai_SSE1, t3_SSE1);
2718 obc_mask3_SSE2 = _mm_cmplt_ps(rai_SSE2, t3_SSE2);
2719 obc_mask3_SSE3 = _mm_cmplt_ps(rai_SSE3, t3_SSE3);
2720 obc_mask1_SSE0 = _mm_and_ps(obc_mask1_SSE0, jmask_SSE0);
2721 obc_mask1_SSE1 = _mm_and_ps(obc_mask1_SSE1, jmask_SSE1);
2722 obc_mask1_SSE2 = _mm_and_ps(obc_mask1_SSE2, jmask_SSE2);
2723 obc_mask1_SSE3 = _mm_and_ps(obc_mask1_SSE3, jmask_SSE3);
2725 uij_SSE0 = gmx_mm_inv_ps(t1_SSE0);
2726 uij_SSE1 = gmx_mm_inv_ps(t1_SSE1);
2727 uij_SSE2 = gmx_mm_inv_ps(t1_SSE2);
2728 uij_SSE3 = gmx_mm_inv_ps(t1_SSE3);
2729 lij_SSE0 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE0, gmx_mm_inv_ps(t2_SSE0)),
2730 _mm_andnot_ps(obc_mask2_SSE0, rai_inv_SSE0));
2731 lij_SSE1 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE1, gmx_mm_inv_ps(t2_SSE1)),
2732 _mm_andnot_ps(obc_mask2_SSE1, rai_inv_SSE1));
2733 lij_SSE2 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE2, gmx_mm_inv_ps(t2_SSE2)),
2734 _mm_andnot_ps(obc_mask2_SSE2, rai_inv_SSE2));
2735 lij_SSE3 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE3, gmx_mm_inv_ps(t2_SSE3)),
2736 _mm_andnot_ps(obc_mask2_SSE3, rai_inv_SSE3));
2737 dlij_SSE0 = _mm_and_ps(one_SSE, obc_mask2_SSE0);
2738 dlij_SSE1 = _mm_and_ps(one_SSE, obc_mask2_SSE1);
2739 dlij_SSE2 = _mm_and_ps(one_SSE, obc_mask2_SSE2);
2740 dlij_SSE3 = _mm_and_ps(one_SSE, obc_mask2_SSE3);
2742 uij2_SSE0 = _mm_mul_ps(uij_SSE0, uij_SSE0);
2743 uij2_SSE1 = _mm_mul_ps(uij_SSE1, uij_SSE1);
2744 uij2_SSE2 = _mm_mul_ps(uij_SSE2, uij_SSE2);
2745 uij2_SSE3 = _mm_mul_ps(uij_SSE3, uij_SSE3);
2746 uij3_SSE0 = _mm_mul_ps(uij2_SSE0, uij_SSE0);
2747 uij3_SSE1 = _mm_mul_ps(uij2_SSE1, uij_SSE1);
2748 uij3_SSE2 = _mm_mul_ps(uij2_SSE2, uij_SSE2);
2749 uij3_SSE3 = _mm_mul_ps(uij2_SSE3, uij_SSE3);
2750 lij2_SSE0 = _mm_mul_ps(lij_SSE0, lij_SSE0);
2751 lij2_SSE1 = _mm_mul_ps(lij_SSE1, lij_SSE1);
2752 lij2_SSE2 = _mm_mul_ps(lij_SSE2, lij_SSE2);
2753 lij2_SSE3 = _mm_mul_ps(lij_SSE3, lij_SSE3);
2754 lij3_SSE0 = _mm_mul_ps(lij2_SSE0, lij_SSE0);
2755 lij3_SSE1 = _mm_mul_ps(lij2_SSE1, lij_SSE1);
2756 lij3_SSE2 = _mm_mul_ps(lij2_SSE2, lij_SSE2);
2757 lij3_SSE3 = _mm_mul_ps(lij2_SSE3, lij_SSE3);
2759 diff2_SSE0 = _mm_sub_ps(uij2_SSE0, lij2_SSE0);
2760 diff2_SSE1 = _mm_sub_ps(uij2_SSE1, lij2_SSE1);
2761 diff2_SSE2 = _mm_sub_ps(uij2_SSE2, lij2_SSE2);
2762 diff2_SSE3 = _mm_sub_ps(uij2_SSE3, lij2_SSE3);
2763 lij_inv_SSE0 = gmx_mm_invsqrt_ps(lij2_SSE0);
2764 lij_inv_SSE1 = gmx_mm_invsqrt_ps(lij2_SSE1);
2765 lij_inv_SSE2 = gmx_mm_invsqrt_ps(lij2_SSE2);
2766 lij_inv_SSE3 = gmx_mm_invsqrt_ps(lij2_SSE3);
2767 sk2_aj_SSE = _mm_mul_ps(sk_aj_SSE, sk_aj_SSE);
2768 sk2_rinv_SSE0 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE0);
2769 sk2_rinv_SSE1 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE1);
2770 sk2_rinv_SSE2 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE2);
2771 sk2_rinv_SSE3 = _mm_mul_ps(sk2_aj_SSE, rinv_SSE3);
2772 prod_SSE0 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE0);
2773 prod_SSE1 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE1);
2774 prod_SSE2 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE2);
2775 prod_SSE3 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE3);
2777 logterm_SSE0 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE0, lij_inv_SSE0));
2778 logterm_SSE1 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE1, lij_inv_SSE1));
2779 logterm_SSE2 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE2, lij_inv_SSE2));
2780 logterm_SSE3 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE3, lij_inv_SSE3));
2782 t1_SSE0 = _mm_sub_ps(lij_SSE0, uij_SSE0);
2783 t1_SSE1 = _mm_sub_ps(lij_SSE1, uij_SSE1);
2784 t1_SSE2 = _mm_sub_ps(lij_SSE2, uij_SSE2);
2785 t1_SSE3 = _mm_sub_ps(lij_SSE3, uij_SSE3);
2786 t2_SSE0 = _mm_mul_ps(diff2_SSE0,
2787 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE0),
2789 t2_SSE1 = _mm_mul_ps(diff2_SSE1,
2790 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE1),
2792 t2_SSE2 = _mm_mul_ps(diff2_SSE2,
2793 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE2),
2795 t2_SSE3 = _mm_mul_ps(diff2_SSE3,
2796 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE3),
2799 t3_SSE0 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE0, logterm_SSE0));
2800 t3_SSE1 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE1, logterm_SSE1));
2801 t3_SSE2 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE2, logterm_SSE2));
2802 t3_SSE3 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE3, logterm_SSE3));
2803 t1_SSE0 = _mm_add_ps(t1_SSE0, _mm_add_ps(t2_SSE0, t3_SSE0));
2804 t1_SSE1 = _mm_add_ps(t1_SSE1, _mm_add_ps(t2_SSE1, t3_SSE1));
2805 t1_SSE2 = _mm_add_ps(t1_SSE2, _mm_add_ps(t2_SSE2, t3_SSE2));
2806 t1_SSE3 = _mm_add_ps(t1_SSE3, _mm_add_ps(t2_SSE3, t3_SSE3));
2807 t4_SSE0 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE0, lij_SSE0));
2808 t4_SSE1 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE1, lij_SSE1));
2809 t4_SSE2 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE2, lij_SSE2));
2810 t4_SSE3 = _mm_mul_ps(two_SSE, _mm_sub_ps(rai_inv_SSE3, lij_SSE3));
2811 t4_SSE0 = _mm_and_ps(t4_SSE0, obc_mask3_SSE0);
2812 t4_SSE1 = _mm_and_ps(t4_SSE1, obc_mask3_SSE1);
2813 t4_SSE2 = _mm_and_ps(t4_SSE2, obc_mask3_SSE2);
2814 t4_SSE3 = _mm_and_ps(t4_SSE3, obc_mask3_SSE3);
2815 t1_SSE0 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE0, t4_SSE0));
2816 t1_SSE1 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE1, t4_SSE1));
2817 t1_SSE2 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE2, t4_SSE2));
2818 t1_SSE3 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE3, t4_SSE3));
2820 sum_ai_SSE0 = _mm_add_ps(sum_ai_SSE0, _mm_and_ps(t1_SSE0, obc_mask1_SSE0));
2821 sum_ai_SSE1 = _mm_add_ps(sum_ai_SSE1, _mm_and_ps(t1_SSE1, obc_mask1_SSE1));
2822 sum_ai_SSE2 = _mm_add_ps(sum_ai_SSE2, _mm_and_ps(t1_SSE2, obc_mask1_SSE2));
2823 sum_ai_SSE3 = _mm_add_ps(sum_ai_SSE3, _mm_and_ps(t1_SSE3, obc_mask1_SSE3));
2825 t1_SSE0 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE0),
2826 _mm_mul_ps(prod_SSE0, lij3_SSE0));
2827 t1_SSE1 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE1),
2828 _mm_mul_ps(prod_SSE1, lij3_SSE1));
2829 t1_SSE2 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE2),
2830 _mm_mul_ps(prod_SSE2, lij3_SSE2));
2831 t1_SSE3 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE3),
2832 _mm_mul_ps(prod_SSE3, lij3_SSE3));
2833 t1_SSE0 = _mm_sub_ps(t1_SSE0,
2834 _mm_mul_ps(onefourth_SSE,
2835 _mm_add_ps(_mm_mul_ps(lij_SSE0, rinv_SSE0),
2836 _mm_mul_ps(lij3_SSE0, dr_SSE0))));
2837 t1_SSE1 = _mm_sub_ps(t1_SSE1,
2838 _mm_mul_ps(onefourth_SSE,
2839 _mm_add_ps(_mm_mul_ps(lij_SSE1, rinv_SSE1),
2840 _mm_mul_ps(lij3_SSE1, dr_SSE1))));
2841 t1_SSE2 = _mm_sub_ps(t1_SSE2,
2842 _mm_mul_ps(onefourth_SSE,
2843 _mm_add_ps(_mm_mul_ps(lij_SSE2, rinv_SSE2),
2844 _mm_mul_ps(lij3_SSE2, dr_SSE2))));
2845 t1_SSE3 = _mm_sub_ps(t1_SSE3,
2846 _mm_mul_ps(onefourth_SSE,
2847 _mm_add_ps(_mm_mul_ps(lij_SSE3, rinv_SSE3),
2848 _mm_mul_ps(lij3_SSE3, dr_SSE3))));
2850 t2_SSE0 = _mm_mul_ps(onefourth_SSE,
2851 _mm_add_ps(_mm_mul_ps(uij_SSE0, rinv_SSE0),
2852 _mm_mul_ps(uij3_SSE0, dr_SSE0)));
2853 t2_SSE1 = _mm_mul_ps(onefourth_SSE,
2854 _mm_add_ps(_mm_mul_ps(uij_SSE1, rinv_SSE1),
2855 _mm_mul_ps(uij3_SSE1, dr_SSE1)));
2856 t2_SSE2 = _mm_mul_ps(onefourth_SSE,
2857 _mm_add_ps(_mm_mul_ps(uij_SSE2, rinv_SSE2),
2858 _mm_mul_ps(uij3_SSE2, dr_SSE2)));
2859 t2_SSE3 = _mm_mul_ps(onefourth_SSE,
2860 _mm_add_ps(_mm_mul_ps(uij_SSE3, rinv_SSE3),
2861 _mm_mul_ps(uij3_SSE3, dr_SSE3)));
2862 t2_SSE0 = _mm_sub_ps(t2_SSE0,
2863 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE0),
2864 _mm_mul_ps(prod_SSE0, uij3_SSE0)));
2865 t2_SSE1 = _mm_sub_ps(t2_SSE1,
2866 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE1),
2867 _mm_mul_ps(prod_SSE1, uij3_SSE1)));
2868 t2_SSE2 = _mm_sub_ps(t2_SSE2,
2869 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE2),
2870 _mm_mul_ps(prod_SSE2, uij3_SSE2)));
2871 t2_SSE3 = _mm_sub_ps(t2_SSE3,
2872 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE3),
2873 _mm_mul_ps(prod_SSE3, uij3_SSE3)));
2874 t3_SSE0 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE0),
2875 _mm_mul_ps(rinv_SSE0, rinv_SSE0));
2876 t3_SSE1 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE1),
2877 _mm_mul_ps(rinv_SSE1, rinv_SSE1));
2878 t3_SSE2 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE2),
2879 _mm_mul_ps(rinv_SSE2, rinv_SSE2));
2880 t3_SSE3 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE3),
2881 _mm_mul_ps(rinv_SSE3, rinv_SSE3));
2882 t3_SSE0 = _mm_sub_ps(t3_SSE0,
2883 _mm_mul_ps(_mm_mul_ps(diff2_SSE0, oneeighth_SSE),
2885 _mm_mul_ps(sk2_rinv_SSE0, rinv_SSE0))));
2886 t3_SSE1 = _mm_sub_ps(t3_SSE1,
2887 _mm_mul_ps(_mm_mul_ps(diff2_SSE1, oneeighth_SSE),
2889 _mm_mul_ps(sk2_rinv_SSE1, rinv_SSE1))));
2890 t3_SSE2 = _mm_sub_ps(t3_SSE2,
2891 _mm_mul_ps(_mm_mul_ps(diff2_SSE2, oneeighth_SSE),
2893 _mm_mul_ps(sk2_rinv_SSE2, rinv_SSE2))));
2894 t3_SSE3 = _mm_sub_ps(t3_SSE3,
2895 _mm_mul_ps(_mm_mul_ps(diff2_SSE3, oneeighth_SSE),
2897 _mm_mul_ps(sk2_rinv_SSE3, rinv_SSE3))));
2899 t1_SSE0 = _mm_mul_ps(rinv_SSE0,
2900 _mm_add_ps(_mm_mul_ps(dlij_SSE0, t1_SSE0),
2901 _mm_add_ps(t2_SSE0, t3_SSE0)));
2902 t1_SSE1 = _mm_mul_ps(rinv_SSE1,
2903 _mm_add_ps(_mm_mul_ps(dlij_SSE1, t1_SSE1),
2904 _mm_add_ps(t2_SSE1, t3_SSE1)));
2905 t1_SSE2 = _mm_mul_ps(rinv_SSE2,
2906 _mm_add_ps(_mm_mul_ps(dlij_SSE2, t1_SSE2),
2907 _mm_add_ps(t2_SSE2, t3_SSE2)));
2908 t1_SSE3 = _mm_mul_ps(rinv_SSE3,
2909 _mm_add_ps(_mm_mul_ps(dlij_SSE3, t1_SSE3),
2910 _mm_add_ps(t2_SSE3, t3_SSE3)));
2912 _mm_store_ps(dadx, _mm_and_ps(t1_SSE0, obc_mask1_SSE0));
2914 _mm_store_ps(dadx, _mm_and_ps(t1_SSE1, obc_mask1_SSE1));
2916 _mm_store_ps(dadx, _mm_and_ps(t1_SSE2, obc_mask1_SSE2));
2918 _mm_store_ps(dadx, _mm_and_ps(t1_SSE3, obc_mask1_SSE3));
2921 /* Evaluate influence of atom ai -> aj */
2922 t1_SSE0 = _mm_add_ps(dr_SSE0, sk_ai_SSE0);
2923 t1_SSE1 = _mm_add_ps(dr_SSE1, sk_ai_SSE1);
2924 t1_SSE2 = _mm_add_ps(dr_SSE2, sk_ai_SSE2);
2925 t1_SSE3 = _mm_add_ps(dr_SSE3, sk_ai_SSE3);
2926 t2_SSE0 = _mm_sub_ps(dr_SSE0, sk_ai_SSE0);
2927 t2_SSE1 = _mm_sub_ps(dr_SSE1, sk_ai_SSE1);
2928 t2_SSE2 = _mm_sub_ps(dr_SSE2, sk_ai_SSE2);
2929 t2_SSE3 = _mm_sub_ps(dr_SSE3, sk_ai_SSE3);
2930 t3_SSE0 = _mm_sub_ps(sk_ai_SSE0, dr_SSE0);
2931 t3_SSE1 = _mm_sub_ps(sk_ai_SSE1, dr_SSE1);
2932 t3_SSE2 = _mm_sub_ps(sk_ai_SSE2, dr_SSE2);
2933 t3_SSE3 = _mm_sub_ps(sk_ai_SSE3, dr_SSE3);
2935 obc_mask1_SSE0 = _mm_cmplt_ps(raj_SSE, t1_SSE0);
2936 obc_mask1_SSE1 = _mm_cmplt_ps(raj_SSE, t1_SSE1);
2937 obc_mask1_SSE2 = _mm_cmplt_ps(raj_SSE, t1_SSE2);
2938 obc_mask1_SSE3 = _mm_cmplt_ps(raj_SSE, t1_SSE3);
2939 obc_mask2_SSE0 = _mm_cmplt_ps(raj_SSE, t2_SSE0);
2940 obc_mask2_SSE1 = _mm_cmplt_ps(raj_SSE, t2_SSE1);
2941 obc_mask2_SSE2 = _mm_cmplt_ps(raj_SSE, t2_SSE2);
2942 obc_mask2_SSE3 = _mm_cmplt_ps(raj_SSE, t2_SSE3);
2943 obc_mask3_SSE0 = _mm_cmplt_ps(raj_SSE, t3_SSE0);
2944 obc_mask3_SSE1 = _mm_cmplt_ps(raj_SSE, t3_SSE1);
2945 obc_mask3_SSE2 = _mm_cmplt_ps(raj_SSE, t3_SSE2);
2946 obc_mask3_SSE3 = _mm_cmplt_ps(raj_SSE, t3_SSE3);
2947 obc_mask1_SSE0 = _mm_and_ps(obc_mask1_SSE0, jmask_SSE0);
2948 obc_mask1_SSE1 = _mm_and_ps(obc_mask1_SSE1, jmask_SSE1);
2949 obc_mask1_SSE2 = _mm_and_ps(obc_mask1_SSE2, jmask_SSE2);
2950 obc_mask1_SSE3 = _mm_and_ps(obc_mask1_SSE3, jmask_SSE3);
2952 uij_SSE0 = gmx_mm_inv_ps(t1_SSE0);
2953 uij_SSE1 = gmx_mm_inv_ps(t1_SSE1);
2954 uij_SSE2 = gmx_mm_inv_ps(t1_SSE2);
2955 uij_SSE3 = gmx_mm_inv_ps(t1_SSE3);
2956 lij_SSE0 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE0, gmx_mm_inv_ps(t2_SSE0)),
2957 _mm_andnot_ps(obc_mask2_SSE0, raj_inv_SSE));
2958 lij_SSE1 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE1, gmx_mm_inv_ps(t2_SSE1)),
2959 _mm_andnot_ps(obc_mask2_SSE1, raj_inv_SSE));
2960 lij_SSE2 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE2, gmx_mm_inv_ps(t2_SSE2)),
2961 _mm_andnot_ps(obc_mask2_SSE2, raj_inv_SSE));
2962 lij_SSE3 = _mm_or_ps( _mm_and_ps(obc_mask2_SSE3, gmx_mm_inv_ps(t2_SSE3)),
2963 _mm_andnot_ps(obc_mask2_SSE3, raj_inv_SSE));
2964 dlij_SSE0 = _mm_and_ps(one_SSE, obc_mask2_SSE0);
2965 dlij_SSE1 = _mm_and_ps(one_SSE, obc_mask2_SSE1);
2966 dlij_SSE2 = _mm_and_ps(one_SSE, obc_mask2_SSE2);
2967 dlij_SSE3 = _mm_and_ps(one_SSE, obc_mask2_SSE3);
2969 uij2_SSE0 = _mm_mul_ps(uij_SSE0, uij_SSE0);
2970 uij2_SSE1 = _mm_mul_ps(uij_SSE1, uij_SSE1);
2971 uij2_SSE2 = _mm_mul_ps(uij_SSE2, uij_SSE2);
2972 uij2_SSE3 = _mm_mul_ps(uij_SSE3, uij_SSE3);
2973 uij3_SSE0 = _mm_mul_ps(uij2_SSE0, uij_SSE0);
2974 uij3_SSE1 = _mm_mul_ps(uij2_SSE1, uij_SSE1);
2975 uij3_SSE2 = _mm_mul_ps(uij2_SSE2, uij_SSE2);
2976 uij3_SSE3 = _mm_mul_ps(uij2_SSE3, uij_SSE3);
2977 lij2_SSE0 = _mm_mul_ps(lij_SSE0, lij_SSE0);
2978 lij2_SSE1 = _mm_mul_ps(lij_SSE1, lij_SSE1);
2979 lij2_SSE2 = _mm_mul_ps(lij_SSE2, lij_SSE2);
2980 lij2_SSE3 = _mm_mul_ps(lij_SSE3, lij_SSE3);
2981 lij3_SSE0 = _mm_mul_ps(lij2_SSE0, lij_SSE0);
2982 lij3_SSE1 = _mm_mul_ps(lij2_SSE1, lij_SSE1);
2983 lij3_SSE2 = _mm_mul_ps(lij2_SSE2, lij_SSE2);
2984 lij3_SSE3 = _mm_mul_ps(lij2_SSE3, lij_SSE3);
2986 diff2_SSE0 = _mm_sub_ps(uij2_SSE0, lij2_SSE0);
2987 diff2_SSE1 = _mm_sub_ps(uij2_SSE1, lij2_SSE1);
2988 diff2_SSE2 = _mm_sub_ps(uij2_SSE2, lij2_SSE2);
2989 diff2_SSE3 = _mm_sub_ps(uij2_SSE3, lij2_SSE3);
2990 lij_inv_SSE0 = gmx_mm_invsqrt_ps(lij2_SSE0);
2991 lij_inv_SSE1 = gmx_mm_invsqrt_ps(lij2_SSE1);
2992 lij_inv_SSE2 = gmx_mm_invsqrt_ps(lij2_SSE2);
2993 lij_inv_SSE3 = gmx_mm_invsqrt_ps(lij2_SSE3);
2994 sk2_rinv_SSE0 = _mm_mul_ps(sk2_ai_SSE0, rinv_SSE0);
2995 sk2_rinv_SSE1 = _mm_mul_ps(sk2_ai_SSE1, rinv_SSE1);
2996 sk2_rinv_SSE2 = _mm_mul_ps(sk2_ai_SSE2, rinv_SSE2);
2997 sk2_rinv_SSE3 = _mm_mul_ps(sk2_ai_SSE3, rinv_SSE3);
2998 prod_SSE0 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE0);
2999 prod_SSE1 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE1);
3000 prod_SSE2 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE2);
3001 prod_SSE3 = _mm_mul_ps(onefourth_SSE, sk2_rinv_SSE3);
3003 logterm_SSE0 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE0, lij_inv_SSE0));
3004 logterm_SSE1 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE1, lij_inv_SSE1));
3005 logterm_SSE2 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE2, lij_inv_SSE2));
3006 logterm_SSE3 = gmx_mm_log_ps(_mm_mul_ps(uij_SSE3, lij_inv_SSE3));
3007 t1_SSE0 = _mm_sub_ps(lij_SSE0, uij_SSE0);
3008 t1_SSE1 = _mm_sub_ps(lij_SSE1, uij_SSE1);
3009 t1_SSE2 = _mm_sub_ps(lij_SSE2, uij_SSE2);
3010 t1_SSE3 = _mm_sub_ps(lij_SSE3, uij_SSE3);
3011 t2_SSE0 = _mm_mul_ps(diff2_SSE0,
3012 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE0),
3014 t2_SSE1 = _mm_mul_ps(diff2_SSE1,
3015 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE1),
3017 t2_SSE2 = _mm_mul_ps(diff2_SSE2,
3018 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE2),
3020 t2_SSE3 = _mm_mul_ps(diff2_SSE3,
3021 _mm_sub_ps(_mm_mul_ps(onefourth_SSE, dr_SSE3),
3023 t3_SSE0 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE0, logterm_SSE0));
3024 t3_SSE1 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE1, logterm_SSE1));
3025 t3_SSE2 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE2, logterm_SSE2));
3026 t3_SSE3 = _mm_mul_ps(half_SSE, _mm_mul_ps(rinv_SSE3, logterm_SSE3));
3027 t1_SSE0 = _mm_add_ps(t1_SSE0, _mm_add_ps(t2_SSE0, t3_SSE0));
3028 t1_SSE1 = _mm_add_ps(t1_SSE1, _mm_add_ps(t2_SSE1, t3_SSE1));
3029 t1_SSE2 = _mm_add_ps(t1_SSE2, _mm_add_ps(t2_SSE2, t3_SSE2));
3030 t1_SSE3 = _mm_add_ps(t1_SSE3, _mm_add_ps(t2_SSE3, t3_SSE3));
3031 t4_SSE0 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE0));
3032 t4_SSE1 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE1));
3033 t4_SSE2 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE2));
3034 t4_SSE3 = _mm_mul_ps(two_SSE, _mm_sub_ps(raj_inv_SSE, lij_SSE3));
3035 t4_SSE0 = _mm_and_ps(t4_SSE0, obc_mask3_SSE0);
3036 t4_SSE1 = _mm_and_ps(t4_SSE1, obc_mask3_SSE1);
3037 t4_SSE2 = _mm_and_ps(t4_SSE2, obc_mask3_SSE2);
3038 t4_SSE3 = _mm_and_ps(t4_SSE3, obc_mask3_SSE3);
3039 t1_SSE0 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE0, t4_SSE0));
3040 t1_SSE1 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE1, t4_SSE1));
3041 t1_SSE2 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE2, t4_SSE2));
3042 t1_SSE3 = _mm_mul_ps(half_SSE, _mm_add_ps(t1_SSE3, t4_SSE3));
3044 _mm_store_ps(work+j, _mm_add_ps(_mm_load_ps(work+j),
3045 gmx_mm_sum4_ps(_mm_and_ps(t1_SSE0, obc_mask1_SSE0),
3046 _mm_and_ps(t1_SSE1, obc_mask1_SSE1),
3047 _mm_and_ps(t1_SSE2, obc_mask1_SSE2),
3048 _mm_and_ps(t1_SSE3, obc_mask1_SSE3))));
3050 t1_SSE0 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE0),
3051 _mm_mul_ps(prod_SSE0, lij3_SSE0));
3052 t1_SSE1 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE1),
3053 _mm_mul_ps(prod_SSE1, lij3_SSE1));
3054 t1_SSE2 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE2),
3055 _mm_mul_ps(prod_SSE2, lij3_SSE2));
3056 t1_SSE3 = _mm_add_ps(_mm_mul_ps(half_SSE, lij2_SSE3),
3057 _mm_mul_ps(prod_SSE3, lij3_SSE3));
3058 t1_SSE0 = _mm_sub_ps(t1_SSE0,
3059 _mm_mul_ps(onefourth_SSE,
3060 _mm_add_ps(_mm_mul_ps(lij_SSE0, rinv_SSE0),
3061 _mm_mul_ps(lij3_SSE0, dr_SSE0))));
3062 t1_SSE1 = _mm_sub_ps(t1_SSE1,
3063 _mm_mul_ps(onefourth_SSE,
3064 _mm_add_ps(_mm_mul_ps(lij_SSE1, rinv_SSE1),
3065 _mm_mul_ps(lij3_SSE1, dr_SSE1))));
3066 t1_SSE2 = _mm_sub_ps(t1_SSE2,
3067 _mm_mul_ps(onefourth_SSE,
3068 _mm_add_ps(_mm_mul_ps(lij_SSE2, rinv_SSE2),
3069 _mm_mul_ps(lij3_SSE2, dr_SSE2))));
3070 t1_SSE3 = _mm_sub_ps(t1_SSE3,
3071 _mm_mul_ps(onefourth_SSE,
3072 _mm_add_ps(_mm_mul_ps(lij_SSE3, rinv_SSE3),
3073 _mm_mul_ps(lij3_SSE3, dr_SSE3))));
3074 t2_SSE0 = _mm_mul_ps(onefourth_SSE,
3075 _mm_add_ps(_mm_mul_ps(uij_SSE0, rinv_SSE0),
3076 _mm_mul_ps(uij3_SSE0, dr_SSE0)));
3077 t2_SSE1 = _mm_mul_ps(onefourth_SSE,
3078 _mm_add_ps(_mm_mul_ps(uij_SSE1, rinv_SSE1),
3079 _mm_mul_ps(uij3_SSE1, dr_SSE1)));
3080 t2_SSE2 = _mm_mul_ps(onefourth_SSE,
3081 _mm_add_ps(_mm_mul_ps(uij_SSE2, rinv_SSE2),
3082 _mm_mul_ps(uij3_SSE2, dr_SSE2)));
3083 t2_SSE3 = _mm_mul_ps(onefourth_SSE,
3084 _mm_add_ps(_mm_mul_ps(uij_SSE3, rinv_SSE3),
3085 _mm_mul_ps(uij3_SSE3, dr_SSE3)));
3086 t2_SSE0 = _mm_sub_ps(t2_SSE0,
3087 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE0),
3088 _mm_mul_ps(prod_SSE0, uij3_SSE0)));
3089 t2_SSE1 = _mm_sub_ps(t2_SSE1,
3090 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE1),
3091 _mm_mul_ps(prod_SSE1, uij3_SSE1)));
3092 t2_SSE2 = _mm_sub_ps(t2_SSE2,
3093 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE2),
3094 _mm_mul_ps(prod_SSE2, uij3_SSE2)));
3095 t2_SSE3 = _mm_sub_ps(t2_SSE3,
3096 _mm_add_ps(_mm_mul_ps(half_SSE, uij2_SSE3),
3097 _mm_mul_ps(prod_SSE3, uij3_SSE3)));
3099 t3_SSE0 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE0),
3100 _mm_mul_ps(rinv_SSE0, rinv_SSE0));
3101 t3_SSE1 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE1),
3102 _mm_mul_ps(rinv_SSE1, rinv_SSE1));
3103 t3_SSE2 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE2),
3104 _mm_mul_ps(rinv_SSE2, rinv_SSE2));
3105 t3_SSE3 = _mm_mul_ps(_mm_mul_ps(onefourth_SSE, logterm_SSE3),
3106 _mm_mul_ps(rinv_SSE3, rinv_SSE3));
3108 t3_SSE0 = _mm_sub_ps(t3_SSE0,
3109 _mm_mul_ps(_mm_mul_ps(diff2_SSE0, oneeighth_SSE),
3111 _mm_mul_ps(sk2_rinv_SSE0, rinv_SSE0))));
3112 t3_SSE1 = _mm_sub_ps(t3_SSE1,
3113 _mm_mul_ps(_mm_mul_ps(diff2_SSE1, oneeighth_SSE),
3115 _mm_mul_ps(sk2_rinv_SSE1, rinv_SSE1))));
3116 t3_SSE2 = _mm_sub_ps(t3_SSE2,
3117 _mm_mul_ps(_mm_mul_ps(diff2_SSE2, oneeighth_SSE),
3119 _mm_mul_ps(sk2_rinv_SSE2, rinv_SSE2))));
3120 t3_SSE3 = _mm_sub_ps(t3_SSE3,
3121 _mm_mul_ps(_mm_mul_ps(diff2_SSE3, oneeighth_SSE),
3123 _mm_mul_ps(sk2_rinv_SSE3, rinv_SSE3))));
3126 t1_SSE0 = _mm_mul_ps(rinv_SSE0,
3127 _mm_add_ps(_mm_mul_ps(dlij_SSE0, t1_SSE0),
3128 _mm_add_ps(t2_SSE0, t3_SSE0)));
3129 t1_SSE1 = _mm_mul_ps(rinv_SSE1,
3130 _mm_add_ps(_mm_mul_ps(dlij_SSE1, t1_SSE1),
3131 _mm_add_ps(t2_SSE1, t3_SSE1)));
3132 t1_SSE2 = _mm_mul_ps(rinv_SSE2,
3133 _mm_add_ps(_mm_mul_ps(dlij_SSE2, t1_SSE2),
3134 _mm_add_ps(t2_SSE2, t3_SSE2)));
3135 t1_SSE3 = _mm_mul_ps(rinv_SSE3,
3136 _mm_add_ps(_mm_mul_ps(dlij_SSE3, t1_SSE3),
3137 _mm_add_ps(t2_SSE3, t3_SSE3)));
3139 _mm_store_ps(dadx, _mm_and_ps(t1_SSE0, obc_mask1_SSE0));
3141 _mm_store_ps(dadx, _mm_and_ps(t1_SSE1, obc_mask1_SSE1));
3143 _mm_store_ps(dadx, _mm_and_ps(t1_SSE2, obc_mask1_SSE2));
3145 _mm_store_ps(dadx, _mm_and_ps(t1_SSE3, obc_mask1_SSE3));
3148 _MM_TRANSPOSE4_PS(sum_ai_SSE0, sum_ai_SSE1, sum_ai_SSE2, sum_ai_SSE3);
3149 sum_ai_SSE0 = _mm_add_ps(sum_ai_SSE0, sum_ai_SSE1);
3150 sum_ai_SSE2 = _mm_add_ps(sum_ai_SSE2, sum_ai_SSE3);
3151 sum_ai_SSE0 = _mm_add_ps(sum_ai_SSE0, sum_ai_SSE2);
3152 _mm_store_ps(work+i, _mm_add_ps(sum_ai_SSE0, _mm_load_ps(work+i)));
3156 for (i = 0; i < natoms/2+1; i++)
3158 work[i] += work[natoms+i];
3161 /* Parallel summations would go here if ever implemented with DD */
3163 if (gb_algorithm == egbHCT)
3166 for (i = 0; i < natoms; i++)
3168 if (born->use[i] != 0)
3170 rai = top->atomtypes.gb_radius[mdatoms->typeA[i]]-born->gb_doffset;
3171 sum_ai = 1.0/rai - work[i];
3172 min_rad = rai + born->gb_doffset;
3175 born->bRad[i] = rad > min_rad ? rad : min_rad;
3176 fr->invsqrta[i] = gmx_invsqrt(born->bRad[i]);
3185 /* Calculate the radii */
3186 for (i = 0; i < natoms; i++)
3189 if (born->use[i] != 0)
3191 rai = top->atomtypes.gb_radius[mdatoms->typeA[i]];
3193 rai = rai-born->gb_doffset;
3195 sum_ai = rai * work[i];
3196 sum_ai2 = sum_ai * sum_ai;
3197 sum_ai3 = sum_ai2 * sum_ai;
3199 tsum = tanh(born->obc_alpha*sum_ai-born->obc_beta*sum_ai2+born->obc_gamma*sum_ai3);
3200 born->bRad[i] = rai_inv - tsum*rai_inv2;
3201 born->bRad[i] = 1.0 / born->bRad[i];
3203 fr->invsqrta[i] = gmx_invsqrt(born->bRad[i]);
3205 tchain = rai * (born->obc_alpha-2*born->obc_beta*sum_ai+3*born->obc_gamma*sum_ai2);
3206 born->drobc[i] = (1.0-tsum*tsum)*tchain*rai_inv2;
3222 genborn_allvsall_calc_chainrule_sse2_single(t_forcerec * fr,
3223 t_mdatoms * mdatoms,
3224 gmx_genborn_t * born,
3230 gmx_allvsallgb2_data_t *aadata;
3233 int nj0, nj1, nj2, nj3;
3246 real rbai, rbaj, fgb, fgb_ai, rbi;
3257 __m128 jmask_SSE0, jmask_SSE1, jmask_SSE2, jmask_SSE3;
3258 __m128 ix_SSE0, iy_SSE0, iz_SSE0;
3259 __m128 ix_SSE1, iy_SSE1, iz_SSE1;
3260 __m128 ix_SSE2, iy_SSE2, iz_SSE2;
3261 __m128 ix_SSE3, iy_SSE3, iz_SSE3;
3262 __m128 fix_SSE0, fiy_SSE0, fiz_SSE0;
3263 __m128 fix_SSE1, fiy_SSE1, fiz_SSE1;
3264 __m128 fix_SSE2, fiy_SSE2, fiz_SSE2;
3265 __m128 fix_SSE3, fiy_SSE3, fiz_SSE3;
3266 __m128 rbai_SSE0, rbai_SSE1, rbai_SSE2, rbai_SSE3;
3267 __m128 imask_SSE0, imask_SSE1, imask_SSE2, imask_SSE3;
3268 __m128 jx_SSE, jy_SSE, jz_SSE, rbaj_SSE;
3269 __m128 dx_SSE0, dy_SSE0, dz_SSE0;
3270 __m128 dx_SSE1, dy_SSE1, dz_SSE1;
3271 __m128 dx_SSE2, dy_SSE2, dz_SSE2;
3272 __m128 dx_SSE3, dy_SSE3, dz_SSE3;
3273 __m128 fgb_SSE0, fgb_ai_SSE0;
3274 __m128 fgb_SSE1, fgb_ai_SSE1;
3275 __m128 fgb_SSE2, fgb_ai_SSE2;
3276 __m128 fgb_SSE3, fgb_ai_SSE3;
3277 __m128 tx_SSE0, ty_SSE0, tz_SSE0;
3278 __m128 tx_SSE1, ty_SSE1, tz_SSE1;
3279 __m128 tx_SSE2, ty_SSE2, tz_SSE2;
3280 __m128 tx_SSE3, ty_SSE3, tz_SSE3;
3283 natoms = mdatoms->nr;
3285 ni1 = mdatoms->homenr;
3288 aadata = (gmx_allvsallgb2_data_t *)paadata;
3290 x_align = aadata->x_align;
3291 y_align = aadata->y_align;
3292 z_align = aadata->z_align;
3293 fx_align = aadata->fx_align;
3294 fy_align = aadata->fy_align;
3295 fz_align = aadata->fz_align;
3297 jindex = aadata->jindex_gb;
3303 /* Loop to get the proper form for the Born radius term */
3304 if (gb_algorithm == egbSTILL)
3306 for (i = 0; i < natoms; i++)
3308 rbi = born->bRad[i];
3309 rb[i] = (2 * rbi * rbi * fr->dvda[i])/ONE_4PI_EPS0;
3312 else if (gb_algorithm == egbHCT)
3314 for (i = 0; i < natoms; i++)
3316 rbi = born->bRad[i];
3317 rb[i] = rbi * rbi * fr->dvda[i];
3320 else if (gb_algorithm == egbOBC)
3322 for (idx = 0; idx < natoms; idx++)
3324 rbi = born->bRad[idx];
3325 rb[idx] = rbi * rbi * born->drobc[idx] * fr->dvda[idx];
3329 for (i = 0; i < 2*natoms; i++)
3337 for (i = 0; i < natoms; i++)
3339 rb[i+natoms] = rb[i];
3342 for (i = ni0; i < ni1; i += UNROLLI)
3344 /* We assume shifts are NOT used for all-vs-all interactions */
3346 /* Load i atom data */
3347 ix_SSE0 = _mm_load1_ps(x_align+i);
3348 iy_SSE0 = _mm_load1_ps(y_align+i);
3349 iz_SSE0 = _mm_load1_ps(z_align+i);
3350 ix_SSE1 = _mm_load1_ps(x_align+i+1);
3351 iy_SSE1 = _mm_load1_ps(y_align+i+1);
3352 iz_SSE1 = _mm_load1_ps(z_align+i+1);
3353 ix_SSE2 = _mm_load1_ps(x_align+i+2);
3354 iy_SSE2 = _mm_load1_ps(y_align+i+2);
3355 iz_SSE2 = _mm_load1_ps(z_align+i+2);
3356 ix_SSE3 = _mm_load1_ps(x_align+i+3);
3357 iy_SSE3 = _mm_load1_ps(y_align+i+3);
3358 iz_SSE3 = _mm_load1_ps(z_align+i+3);
3360 fix_SSE0 = _mm_setzero_ps();
3361 fiy_SSE0 = _mm_setzero_ps();
3362 fiz_SSE0 = _mm_setzero_ps();
3363 fix_SSE1 = _mm_setzero_ps();
3364 fiy_SSE1 = _mm_setzero_ps();
3365 fiz_SSE1 = _mm_setzero_ps();
3366 fix_SSE2 = _mm_setzero_ps();
3367 fiy_SSE2 = _mm_setzero_ps();
3368 fiz_SSE2 = _mm_setzero_ps();
3369 fix_SSE3 = _mm_setzero_ps();
3370 fiy_SSE3 = _mm_setzero_ps();
3371 fiz_SSE3 = _mm_setzero_ps();
3373 rbai_SSE0 = _mm_load1_ps(rb+i);
3374 rbai_SSE1 = _mm_load1_ps(rb+i+1);
3375 rbai_SSE2 = _mm_load1_ps(rb+i+2);
3376 rbai_SSE3 = _mm_load1_ps(rb+i+3);
3378 /* Load limits for loop over neighbors */
3380 nj3 = jindex[4*i+3];
3382 /* No masks necessary, since the stored chain rule derivatives will be zero in those cases! */
3383 for (j = nj0; j < nj3; j += UNROLLJ)
3385 /* load j atom coordinates */
3386 jx_SSE = _mm_load_ps(x_align+j);
3387 jy_SSE = _mm_load_ps(y_align+j);
3388 jz_SSE = _mm_load_ps(z_align+j);
3390 /* Calculate distance */
3391 dx_SSE0 = _mm_sub_ps(ix_SSE0, jx_SSE);
3392 dy_SSE0 = _mm_sub_ps(iy_SSE0, jy_SSE);
3393 dz_SSE0 = _mm_sub_ps(iz_SSE0, jz_SSE);
3394 dx_SSE1 = _mm_sub_ps(ix_SSE1, jx_SSE);
3395 dy_SSE1 = _mm_sub_ps(iy_SSE1, jy_SSE);
3396 dz_SSE1 = _mm_sub_ps(iz_SSE1, jz_SSE);
3397 dx_SSE2 = _mm_sub_ps(ix_SSE2, jx_SSE);
3398 dy_SSE2 = _mm_sub_ps(iy_SSE2, jy_SSE);
3399 dz_SSE2 = _mm_sub_ps(iz_SSE2, jz_SSE);
3400 dx_SSE3 = _mm_sub_ps(ix_SSE3, jx_SSE);
3401 dy_SSE3 = _mm_sub_ps(iy_SSE3, jy_SSE);
3402 dz_SSE3 = _mm_sub_ps(iz_SSE3, jz_SSE);
3404 rbaj_SSE = _mm_load_ps(rb+j);
3406 fgb_SSE0 = _mm_mul_ps(rbai_SSE0, _mm_load_ps(dadx));
3408 fgb_SSE1 = _mm_mul_ps(rbai_SSE1, _mm_load_ps(dadx));
3410 fgb_SSE2 = _mm_mul_ps(rbai_SSE2, _mm_load_ps(dadx));
3412 fgb_SSE3 = _mm_mul_ps(rbai_SSE3, _mm_load_ps(dadx));
3415 fgb_ai_SSE0 = _mm_mul_ps(rbaj_SSE, _mm_load_ps(dadx));
3417 fgb_ai_SSE1 = _mm_mul_ps(rbaj_SSE, _mm_load_ps(dadx));
3419 fgb_ai_SSE2 = _mm_mul_ps(rbaj_SSE, _mm_load_ps(dadx));
3421 fgb_ai_SSE3 = _mm_mul_ps(rbaj_SSE, _mm_load_ps(dadx));
3424 /* Total force between ai and aj is the sum of ai->aj and aj->ai */
3425 fgb_SSE0 = _mm_add_ps(fgb_SSE0, fgb_ai_SSE0);
3426 fgb_SSE1 = _mm_add_ps(fgb_SSE1, fgb_ai_SSE1);
3427 fgb_SSE2 = _mm_add_ps(fgb_SSE2, fgb_ai_SSE2);
3428 fgb_SSE3 = _mm_add_ps(fgb_SSE3, fgb_ai_SSE3);
3430 /* Calculate temporary vectorial force */
3431 tx_SSE0 = _mm_mul_ps(fgb_SSE0, dx_SSE0);
3432 ty_SSE0 = _mm_mul_ps(fgb_SSE0, dy_SSE0);
3433 tz_SSE0 = _mm_mul_ps(fgb_SSE0, dz_SSE0);
3434 tx_SSE1 = _mm_mul_ps(fgb_SSE1, dx_SSE1);
3435 ty_SSE1 = _mm_mul_ps(fgb_SSE1, dy_SSE1);
3436 tz_SSE1 = _mm_mul_ps(fgb_SSE1, dz_SSE1);
3437 tx_SSE2 = _mm_mul_ps(fgb_SSE2, dx_SSE2);
3438 ty_SSE2 = _mm_mul_ps(fgb_SSE2, dy_SSE2);
3439 tz_SSE2 = _mm_mul_ps(fgb_SSE2, dz_SSE2);
3440 tx_SSE3 = _mm_mul_ps(fgb_SSE3, dx_SSE3);
3441 ty_SSE3 = _mm_mul_ps(fgb_SSE3, dy_SSE3);
3442 tz_SSE3 = _mm_mul_ps(fgb_SSE3, dz_SSE3);
3444 /* Increment i atom force */
3445 fix_SSE0 = _mm_add_ps(fix_SSE0, tx_SSE0);
3446 fiy_SSE0 = _mm_add_ps(fiy_SSE0, ty_SSE0);
3447 fiz_SSE0 = _mm_add_ps(fiz_SSE0, tz_SSE0);
3448 fix_SSE1 = _mm_add_ps(fix_SSE1, tx_SSE1);
3449 fiy_SSE1 = _mm_add_ps(fiy_SSE1, ty_SSE1);
3450 fiz_SSE1 = _mm_add_ps(fiz_SSE1, tz_SSE1);
3451 fix_SSE2 = _mm_add_ps(fix_SSE2, tx_SSE2);
3452 fiy_SSE2 = _mm_add_ps(fiy_SSE2, ty_SSE2);
3453 fiz_SSE2 = _mm_add_ps(fiz_SSE2, tz_SSE2);
3454 fix_SSE3 = _mm_add_ps(fix_SSE3, tx_SSE3);
3455 fiy_SSE3 = _mm_add_ps(fiy_SSE3, ty_SSE3);
3456 fiz_SSE3 = _mm_add_ps(fiz_SSE3, tz_SSE3);
3458 /* Decrement j atom force */
3459 _mm_store_ps(fx_align+j,
3460 _mm_sub_ps( _mm_load_ps(fx_align+j), gmx_mm_sum4_ps(tx_SSE0, tx_SSE1, tx_SSE2, tx_SSE3) ));
3461 _mm_store_ps(fy_align+j,
3462 _mm_sub_ps( _mm_load_ps(fy_align+j), gmx_mm_sum4_ps(ty_SSE0, ty_SSE1, ty_SSE2, ty_SSE3) ));
3463 _mm_store_ps(fz_align+j,
3464 _mm_sub_ps( _mm_load_ps(fz_align+j), gmx_mm_sum4_ps(tz_SSE0, tz_SSE1, tz_SSE2, tz_SSE3) ));
3466 /* Add i forces to mem and shifted force list */
3467 _MM_TRANSPOSE4_PS(fix_SSE0, fix_SSE1, fix_SSE2, fix_SSE3);
3468 fix_SSE0 = _mm_add_ps(fix_SSE0, fix_SSE1);
3469 fix_SSE2 = _mm_add_ps(fix_SSE2, fix_SSE3);
3470 fix_SSE0 = _mm_add_ps(fix_SSE0, fix_SSE2);
3471 _mm_store_ps(fx_align+i, _mm_add_ps(fix_SSE0, _mm_load_ps(fx_align+i)));
3473 _MM_TRANSPOSE4_PS(fiy_SSE0, fiy_SSE1, fiy_SSE2, fiy_SSE3);
3474 fiy_SSE0 = _mm_add_ps(fiy_SSE0, fiy_SSE1);
3475 fiy_SSE2 = _mm_add_ps(fiy_SSE2, fiy_SSE3);
3476 fiy_SSE0 = _mm_add_ps(fiy_SSE0, fiy_SSE2);
3477 _mm_store_ps(fy_align+i, _mm_add_ps(fiy_SSE0, _mm_load_ps(fy_align+i)));
3479 _MM_TRANSPOSE4_PS(fiz_SSE0, fiz_SSE1, fiz_SSE2, fiz_SSE3);
3480 fiz_SSE0 = _mm_add_ps(fiz_SSE0, fiz_SSE1);
3481 fiz_SSE2 = _mm_add_ps(fiz_SSE2, fiz_SSE3);
3482 fiz_SSE0 = _mm_add_ps(fiz_SSE0, fiz_SSE2);
3483 _mm_store_ps(fz_align+i, _mm_add_ps(fiz_SSE0, _mm_load_ps(fz_align+i)));
3486 for (i = 0; i < natoms; i++)
3488 f[3*i] += fx_align[i] + fx_align[natoms+i];
3489 f[3*i+1] += fy_align[i] + fy_align[natoms+i];
3490 f[3*i+2] += fz_align[i] + fz_align[natoms+i];
3497 /* dummy variable when not using SSE */
3498 int genborn_allvsall_sse2_single_dummy;