Skip to content
GitLab
Projects
Groups
Snippets
/
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
fleur
fleur
Commits
ded182a0
Commit
ded182a0
authored
Jun 03, 2016
by
Daniel Wortmann
Browse files
Fixed routine for parallel case. TODO: the dgemm was replaced by matmul, this should be
changed again for reasonable performance
parent
8c1f7140
Changes
1
Hide whitespace changes
Inline
Side-by-side
eigen/hsmt_nonsph.F90
View file @
ded182a0
...
...
@@ -73,6 +73,8 @@ CONTAINS
INTEGER
::
n_threads
,
thread
,
blocksize
,
maxloop
INTEGER
,
ALLOCATABLE
::
start_thread
(:),
stop_thread
(:)
lnonsphd
=
MAXVAL
(
atoms
%
lnonsph
)
*
(
MAXVAL
(
atoms
%
lnonsph
)
+2
)
ALLOCATE
(
dtd
(
0
:
lnonsphd
,
0
:
lnonsphd
),
utd
(
0
:
lnonsphd
,
0
:
lnonsphd
),
dtu
(
0
:
lnonsphd
,
0
:
lnonsphd
),
utu
(
0
:
lnonsphd
,
0
:
lnonsphd
))
!Decide how to distribute the work
...
...
@@ -209,13 +211,14 @@ CONTAINS
DO
WHILE
(
kii
<
lapw
%
nv_tot
)
!DO kii = n_rank, nv_tot-1, n_size
ki
=
MOD
(
kii
,
lapw
%
nv
(
iintsp
))
+
1
bsize
=
MIN
(
SIZE
(
aa_block
,
1
),
lapw
%
nv
(
iintsp
)
/
n_size
-
ki
*
n_size
+1
)
!Either use maximal blocksize or number of rows left to calculate
bsize
=
MIN
(
SIZE
(
aa_block
,
1
),
(
lapw
%
nv
(
iintsp
)
-
ki
)/
n_size
+1
)
!Either use maximal blocksize or number of rows left to calculate
IF
(
bsize
<
1
)
EXIT
!nothing more to do here
bsize2
=
bsize
*
n_size
!aa_block(:bsize,:ki+bsize2-1)=matmul(a(ki:ki+bsize-1:n_size,0:lmp,iintsp),conjg(transpose(ax(:ki+bsize2-1,0:lmp))))+ &
! matmul(b(ki:ki+bsize-1:n_size,0:lmp,iintsp),conjg(transpose(bx(:ki+bsize2-1,0:lmp))))
CALL
zgemm
(
"N"
,
"C"
,
bsize
,
ki
+
bsize2
-1
,
lmp
+1
,
one
,
a
(
ki
,
0
,
iintsp
),
SIZE
(
a
,
1
)
*
n_size
,
ax
(
1
,
0
),
SIZE
(
ax
,
1
),
zero
,
aa_block
,
SIZE
(
aa_block
,
1
))
CALL
zgemm
(
"N"
,
"C"
,
bsize
,
ki
+
bsize2
-1
,
lmp
+1
,
one
,
b
(
ki
,
0
,
iintsp
),
SIZE
(
a
,
1
)
*
n_size
,
bx
(
1
,
0
),
SIZE
(
ax
,
1
),
one
,
aa_block
,
SIZE
(
aa_block
,
1
))
bsize2
=
min
(
bsize2
,
lapw
%
nv
(
iintsp
)
-
ki
+1
)
aa_block
(:
bsize
,:
ki
+
bsize2
-1
)
=
matmul
(
a
(
ki
:
ki
+
bsize2
-1
:
n_size
,
0
:
lmp
,
iintsp
),
conjg
(
transpose
(
ax
(:
ki
+
bsize2
-1
,
0
:
lmp
))))
+
&
matmul
(
b
(
ki
:
ki
+
bsize2
-1
:
n_size
,
0
:
lmp
,
iintsp
),
conjg
(
transpose
(
bx
(:
ki
+
bsize2
-1
,
0
:
lmp
))))
!CALL zgemm("N","C",bsize,ki+bsize2-1,lmp+1,one,a(ki,0,iintsp),SIZE(a,1)*n_size,ax(1,0),SIZE(ax,1),zero,aa_block,SIZE(aa_block,1))
!CALL zgemm("N","C",bsize,ki+bsize2-1,lmp+1,one,b(ki,0,iintsp),SIZE(a,1)*n_size,bx(1,0),SIZE(ax,1),one,aa_block,SIZE(aa_block,1))
DO
kb
=
1
,
bsize
IF
(
noco
%
l_noco
.AND.
(
.NOT.
noco
%
l_ss
)
)
THEN
nc
=
1
+
kii
/
n_size
...
...
Write
Preview
Supports
Markdown
0%
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment