fdurastante/psblas3
1
0
Fork 0
mirror of https://github.com/sfilippone/psblas3.git synced 2026-10-20 04:42:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix DOT in OpenACC

Browse Source
This commit is contained in:
sfilippone
2024-10-08 17:41:42 +02:00
parent 4822861b73
commit 5903c0b272
4 changed files with 14 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files
openacc/psb_c_oacc_vect_mod.F90
+4 -8
View File
@@ -798,22 +798,18 @@ contains
class(psb_c_base_vect_type), intent(inout) :: y
integer(psb_ipk_), intent(in) :: n
complex(psb_spk_) :: res
complex(psb_spk_), external :: ddot
integer(psb_ipk_) :: info
res = czero
!!$ write(0,*) 'oacc_dot_v'
select type(yy => y)
type is (psb_c_base_vect_type)
if (x%is_dev()) call x%sync()
res = ddot(n, x%v, 1, yy%v, 1)
type is (psb_c_vect_oacc)
if (x%is_host()) call x%sync()
if (yy%is_host()) call yy%sync()
res = c_inner_oacc_dot(n, x%v, yy%v)
class default
call x%sync()
res = y%dot(n, x%v)
if (x%is_dev()) call x%sync()
res = y%dot(n, x%v)
end select
contains
function c_inner_oacc_dot(n, x, y) result(res)
@@ -838,10 +834,10 @@ contains
complex(psb_spk_), intent(in) :: y(:)
integer(psb_ipk_), intent(in) :: n
complex(psb_spk_) :: res
complex(psb_spk_), external :: ddot
complex(psb_spk_), external :: cdot
if (x%is_dev()) call x%sync()
res = ddot(n, y, 1, x%v, 1)
res = cdot(n, y, 1, x%v, 1)
end function c_oacc_dot_a
openacc/psb_d_oacc_vect_mod.F90
+2 -6
View File
@@ -798,22 +798,18 @@ contains
class(psb_d_base_vect_type), intent(inout) :: y
integer(psb_ipk_), intent(in) :: n
real(psb_dpk_) :: res
real(psb_dpk_), external :: ddot
integer(psb_ipk_) :: info
res = dzero
!!$ write(0,*) 'oacc_dot_v'
select type(yy => y)
type is (psb_d_base_vect_type)
if (x%is_dev()) call x%sync()
res = ddot(n, x%v, 1, yy%v, 1)
type is (psb_d_vect_oacc)
if (x%is_host()) call x%sync()
if (yy%is_host()) call yy%sync()
res = d_inner_oacc_dot(n, x%v, yy%v)
class default
call x%sync()
res = y%dot(n, x%v)
if (x%is_dev()) call x%sync()
res = y%dot(n, x%v)
end select
contains
function d_inner_oacc_dot(n, x, y) result(res)
openacc/psb_s_oacc_vect_mod.F90
+4 -8
View File
@@ -798,22 +798,18 @@ contains
class(psb_s_base_vect_type), intent(inout) :: y
integer(psb_ipk_), intent(in) :: n
real(psb_spk_) :: res
real(psb_spk_), external :: ddot
integer(psb_ipk_) :: info
res = szero
!!$ write(0,*) 'oacc_dot_v'
select type(yy => y)
type is (psb_s_base_vect_type)
if (x%is_dev()) call x%sync()
res = ddot(n, x%v, 1, yy%v, 1)
type is (psb_s_vect_oacc)
if (x%is_host()) call x%sync()
if (yy%is_host()) call yy%sync()
res = s_inner_oacc_dot(n, x%v, yy%v)
class default
call x%sync()
res = y%dot(n, x%v)
if (x%is_dev()) call x%sync()
res = y%dot(n, x%v)
end select
contains
function s_inner_oacc_dot(n, x, y) result(res)
@@ -838,10 +834,10 @@ contains
real(psb_spk_), intent(in) :: y(:)
integer(psb_ipk_), intent(in) :: n
real(psb_spk_) :: res
real(psb_spk_), external :: ddot
real(psb_spk_), external :: sdot
if (x%is_dev()) call x%sync()
res = ddot(n, y, 1, x%v, 1)
res = sdot(n, y, 1, x%v, 1)
end function s_oacc_dot_a
openacc/psb_z_oacc_vect_mod.F90
+4 -8
View File
@@ -798,22 +798,18 @@ contains
class(psb_z_base_vect_type), intent(inout) :: y
integer(psb_ipk_), intent(in) :: n
complex(psb_dpk_) :: res
complex(psb_dpk_), external :: ddot
integer(psb_ipk_) :: info
res = zzero
!!$ write(0,*) 'oacc_dot_v'
select type(yy => y)
type is (psb_z_base_vect_type)
if (x%is_dev()) call x%sync()
res = ddot(n, x%v, 1, yy%v, 1)
type is (psb_z_vect_oacc)
if (x%is_host()) call x%sync()
if (yy%is_host()) call yy%sync()
res = z_inner_oacc_dot(n, x%v, yy%v)
class default
call x%sync()
res = y%dot(n, x%v)
if (x%is_dev()) call x%sync()
res = y%dot(n, x%v)
end select
contains
function z_inner_oacc_dot(n, x, y) result(res)
@@ -838,10 +834,10 @@ contains
complex(psb_dpk_), intent(in) :: y(:)
integer(psb_ipk_), intent(in) :: n
complex(psb_dpk_) :: res
complex(psb_dpk_), external :: ddot
complex(psb_dpk_), external :: zdot
if (x%is_dev()) call x%sync()
res = ddot(n, y, 1, x%v, 1)
res = zdot(n, y, 1, x%v, 1)
end function z_oacc_dot_a