fdurastante/psblas3
1
0
Fork 0
mirror of https://github.com/sfilippone/psblas3.git synced 2026-10-20 12:52:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

Rework cp_{from|to}_fmt for better OpenMP performance

Browse Source
This commit is contained in:
sfilippone
2023-06-28 11:16:53 +02:00
parent d378266f33
commit 2f403e0df7
15 changed files with 386 additions and 76 deletions
Download Patch File Download Diff File Expand all files Collapse all files
base/serial/impl/psb_s_csc_impl.F90
+43 -6
View File
@@ -2329,9 +2329,28 @@ subroutine psb_s_cp_csc_to_fmt(a,b,info)
b%psb_s_base_sparse_mat = a%psb_s_base_sparse_mat
nc = a%get_ncols()
nz = a%get_nzeros()
if (info == 0) call psb_safe_cpy( a%icp(1:nc+1), b%icp , info)
if (info == 0) call psb_safe_cpy( a%ia(1:nz), b%ia , info)
if (info == 0) call psb_safe_cpy( a%val(1:nz), b%val , info)
if (.false.) then
if (info == 0) call psb_safe_cpy( a%icp(1:nc+1), b%icp , info)
if (info == 0) call psb_safe_cpy( a%ia(1:nz), b%ia , info)
if (info == 0) call psb_safe_cpy( a%val(1:nz), b%val , info)
else
! Despite the implementation in safe_cpy, it seems better this way
call psb_realloc(nc+1,b%icp,info)
call psb_realloc(nz,b%ia,info)
call psb_realloc(nz,b%val,info)
!$omp parallel do private(i) schedule(static)
do i=1,nc+1
b%icp(i)=a%icp(i)
end do
!$omp end parallel do
!$omp parallel do private(j) schedule(static)
do j=1,nz
b%ia(j) = a%ia(j)
b%val(j) = a%val(j)
end do
!$omp end parallel do
end if
call b%set_host()
class default
@@ -2443,9 +2462,27 @@ subroutine psb_s_cp_csc_from_fmt(a,b,info)
a%psb_s_base_sparse_mat = b%psb_s_base_sparse_mat
nc = b%get_ncols()
nz = b%get_nzeros()
if (info == 0) call psb_safe_cpy( b%icp(1:nc+1), a%icp , info)
if (info == 0) call psb_safe_cpy( b%ia(1:nz), a%ia , info)
if (info == 0) call psb_safe_cpy( b%val(1:nz), a%val , info)
if (.false.) then
if (info == 0) call psb_safe_cpy( b%icp(1:nc+1), a%icp , info)
if (info == 0) call psb_safe_cpy( b%ia(1:nz), a%ia , info)
if (info == 0) call psb_safe_cpy( b%val(1:nz), a%val , info)
else
! Despite the implementation in safe_cpy, it seems better this way
call psb_realloc(nc+1,a%icp,info)
call psb_realloc(nz,a%ia,info)
call psb_realloc(nz,a%val,info)
!$omp parallel do private(i) schedule(static)
do i=1,nc+1
a%icp(i)=b%icp(i)
end do
!$omp end parallel do
!$omp parallel do private(j) schedule(static)
do j=1,nz
a%ia(j)=b%ia(j)
a%val(j)=b%val(j)
end do
!$omp end parallel do
end if
call a%set_host()
class default