64-BitBrainstorm_2026/AMSS-NCKU
8
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add OpenMP parallelization to BSSN RHS hot-path stencil routines

Browse Source 
Enable OpenMP threading for the dominant computational kernels:
- makefile.inc: add -qopenmp to f90appflags
- diff_new.f90: split fderivs/fdderivs into OpenMP interior + serial boundary
- kodiss.f90: split kodis into OpenMP interior + serial boundary
- lopsidediff.f90: add OMP PARALLEL DO COLLAPSE(2) to lopsided
- fmisc.f90: parallelize symmetry_bd bulk array copy
- bssn_rhs.f90: add OMP WORKSHARE to array-syntax operations

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
CGH0S7
2026-02-07 13:58:55 +08:00
parent 09ffdb553d
commit c6e4d4ab71
6 changed files with 162 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
AMSS_NCKU_source/kodiss.f90
View File

@@ -159,36 +159,12 @@ integer, parameter :: NO_SYMM=0, OCTANT=2
call symmetry_bd(3,ex,f,fh,SoA)
do k=1,ex(3)
do j=1,ex(2)
do i=1,ex(1)
if(i-3 >= imin .and. i+3 <= imax .and. &
j-3 >= jmin .and. j+3 <= jmax .and. &
k-3 >= kmin .and. k+3 <= kmax) then
#if 0
! x direction
f_rhs(i,j,k) = f_rhs(i,j,k) + eps/dX/cof * ( &
(fh(i-3,j,k)+fh(i+3,j,k)) - &
SIX*(fh(i-2,j,k)+fh(i+2,j,k)) + &
FIT*(fh(i-1,j,k)+fh(i+1,j,k)) - &
TWT* fh(i,j,k) )
! y direction
f_rhs(i,j,k) = f_rhs(i,j,k) + eps/dY/cof * ( &
(fh(i,j-3,k)+fh(i,j+3,k)) - &
SIX*(fh(i,j-2,k)+fh(i,j+2,k)) + &
FIT*(fh(i,j-1,k)+fh(i,j+1,k)) - &
TWT* fh(i,j,k) )
! z direction
f_rhs(i,j,k) = f_rhs(i,j,k) + eps/dZ/cof * ( &
(fh(i,j,k-3)+fh(i,j,k+3)) - &
SIX*(fh(i,j,k-2)+fh(i,j,k+2)) + &
FIT*(fh(i,j,k-1)+fh(i,j,k+1)) - &
TWT* fh(i,j,k) )
#else
! calculation order if important ?
! Interior: all stencil points guaranteed in-bounds
!$OMP PARALLEL DO COLLAPSE(2) SCHEDULE(static) PRIVATE(i,j,k)
do k=4,ex(3)-3
do j=4,ex(2)-3
!DIR$ IVDEP
do i=4,ex(1)-3
f_rhs(i,j,k) = f_rhs(i,j,k) + eps/cof *( ( &
(fh(i-3,j,k)+fh(i+3,j,k)) - &
SIX*(fh(i-2,j,k)+fh(i+2,j,k)) + &
@@ -204,9 +180,37 @@ integer, parameter :: NO_SYMM=0, OCTANT=2
SIX*(fh(i,j,k-2)+fh(i,j,k+2)) + &
FIT*(fh(i,j,k-1)+fh(i,j,k+1)) - &
TWT* fh(i,j,k) )/dZ )
#endif
endif
enddo
enddo
enddo
!$OMP END PARALLEL DO
! Boundary shell: original branching logic for points near edges
do k=1,ex(3)
do j=1,ex(2)
do i=1,ex(1)
if(i >= 4 .and. i <= ex(1)-3 .and. &
j >= 4 .and. j <= ex(2)-3 .and. &
k >= 4 .and. k <= ex(3)-3) cycle
if(i-3 >= imin .and. i+3 <= imax .and. &
j-3 >= jmin .and. j+3 <= jmax .and. &
k-3 >= kmin .and. k+3 <= kmax) then
f_rhs(i,j,k) = f_rhs(i,j,k) + eps/cof *( ( &
(fh(i-3,j,k)+fh(i+3,j,k)) - &
SIX*(fh(i-2,j,k)+fh(i+2,j,k)) + &
FIT*(fh(i-1,j,k)+fh(i+1,j,k)) - &
TWT* fh(i,j,k) )/dX + &
( &
(fh(i,j-3,k)+fh(i,j+3,k)) - &
SIX*(fh(i,j-2,k)+fh(i,j+2,k)) + &
FIT*(fh(i,j-1,k)+fh(i,j+1,k)) - &
TWT* fh(i,j,k) )/dY + &
( &
(fh(i,j,k-3)+fh(i,j,k+3)) - &
SIX*(fh(i,j,k-2)+fh(i,j,k+2)) + &
FIT*(fh(i,j,k-1)+fh(i,j,k+1)) - &
TWT* fh(i,j,k) )/dZ )
endif
enddo
enddo
enddo