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
elpa
elpa
Commits
fbb0e6eb
Commit
fbb0e6eb
authored
Jan 04, 2013
by
Alexander Heinecke
Browse files
updated kernel names of real kernels
parent
ae81ebca
Changes
8
Hide whitespace changes
Inline
Side-by-side
ELPA_2011.12.Intrinsics/src/elpa2.f90
View file @
fbb0e6eb
...
...
@@ -1703,7 +1703,7 @@ contains
do
j
=
ncols
,
2
,
-2
w
(:,
1
)
=
bcast_buffer
(
1
:
nbw
,
j
+
off
)
w
(:,
2
)
=
bcast_buffer
(
1
:
nbw
,
j
+
off
-1
)
call
double_hh_trafo
_2hv
(
a
(
1
,
j
+
off
+
a_off
-1
,
istripe
),
w
,
nbw
,
nl
,
stripe_width
,
nbw
)
call
double_hh_trafo
(
a
(
1
,
j
+
off
+
a_off
-1
,
istripe
),
w
,
nbw
,
nl
,
stripe_width
,
nbw
)
enddo
if
(
j
==
1
)
call
single_hh_trafo
(
a
(
1
,
1
+
off
+
a_off
,
istripe
),
bcast_buffer
(
1
,
off
+1
),
nbw
,
nl
,
stripe_width
)
...
...
@@ -1713,12 +1713,12 @@ contains
! w(:,2) = bcast_buffer(1:nbw,j+off-1)
! w(:,3) = bcast_buffer(1:nbw,j+off-2)
! w(:,4) = bcast_buffer(1:nbw,j+off-3)
! call
double
_hh_trafo
_4hv
(a(1,j+off+a_off-3,istripe), w, nbw, nl, stripe_width, nbw)
! call
quad
_hh_trafo(a(1,j+off+a_off-3,istripe), w, nbw, nl, stripe_width, nbw)
!enddo
!do jj = j, 2, -2
! w(:,1) = bcast_buffer(1:nbw,jj+off)
! w(:,2) = bcast_buffer(1:nbw,jj+off-1)
! call double_hh_trafo
_2hv
(a(1,jj+off+a_off-1,istripe), w, nbw, nl, stripe_width, nbw)
! call double_hh_trafo(a(1,jj+off+a_off-1,istripe), w, nbw, nl, stripe_width, nbw)
!enddo
!if(jj==1) call single_hh_trafo(a(1,1+off+a_off,istripe),bcast_buffer(1,off+1), nbw, nl, stripe_width)
...
...
@@ -1730,19 +1730,19 @@ contains
! w(:,4) = bcast_buffer(1:nbw,j+off-3)
! w(:,5) = bcast_buffer(1:nbw,j+off-4)
! w(:,6) = bcast_buffer(1:nbw,j+off-5)
! call
double
_hh_trafo
_6hv
(a(1,j+off+a_off-5,istripe), w, nbw, nl, stripe_width, nbw)
! call
hexa
_hh_trafo(a(1,j+off+a_off-5,istripe), w, nbw, nl, stripe_width, nbw)
!enddo
!do jj = j, 4, -4
! w(:,1) = bcast_buffer(1:nbw,jj+off)
! w(:,2) = bcast_buffer(1:nbw,jj+off-1)
! w(:,3) = bcast_buffer(1:nbw,jj+off-2)
! w(:,4) = bcast_buffer(1:nbw,jj+off-3)
! call
double
_hh_trafo
_4hv
(a(1,jj+off+a_off-3,istripe), w, nbw, nl, stripe_width, nbw)
! call
quad
_hh_trafo(a(1,jj+off+a_off-3,istripe), w, nbw, nl, stripe_width, nbw)
!enddo
!do jjj = jj, 2, -2
! w(:,1) = bcast_buffer(1:nbw,jjj+off)
! w(:,2) = bcast_buffer(1:nbw,jjj+off-1)
! call double_hh_trafo
_2hv
(a(1,jjj+off+a_off-1,istripe), w, nbw, nl, stripe_width, nbw)
! call double_hh_trafo(a(1,jjj+off+a_off-1,istripe), w, nbw, nl, stripe_width, nbw)
!enddo
!if(jjj==1) call single_hh_trafo(a(1,1+off+a_off,istripe),bcast_buffer(1,off+1), nbw, nl, stripe_width)
...
...
ELPA_2011.12.Intrinsics/src/elpa2_kernels/elpa2_kernels_real.f90
View file @
fbb0e6eb
...
...
@@ -12,7 +12,7 @@
!
! --------------------------------------------------------------------------------------------------
subroutine
double_hh_trafo
_2hv
(
q
,
hh
,
nb
,
nq
,
ldq
,
ldh
)
subroutine
double_hh_trafo
(
q
,
hh
,
nb
,
nq
,
ldq
,
ldh
)
implicit
none
...
...
ELPA_2011.12.Intrinsics/src/elpa2_kernels/elpa2_kernels_real_bg.f90
View file @
fbb0e6eb
...
...
@@ -11,7 +11,7 @@
!
! --------------------------------------------------------------------------------------------------
subroutine
double_hh_trafo
_2hv
(
q
,
hh
,
nb
,
nq
,
ldq
,
ldh
)
subroutine
double_hh_trafo
(
q
,
hh
,
nb
,
nq
,
ldq
,
ldh
)
implicit
none
...
...
ELPA_2011.12.Intrinsics/src/elpa2_kernels/elpa2_kernels_real_bgq.f90
View file @
fbb0e6eb
...
...
@@ -11,7 +11,7 @@
!
! --------------------------------------------------------------------------------------------------
subroutine
double_hh_trafo
_2hv
(
q
,
hh
,
nb
,
nq
,
ldq
,
ldh
)
subroutine
double_hh_trafo
(
q
,
hh
,
nb
,
nq
,
ldq
,
ldh
)
implicit
none
...
...
ELPA_2011.12.Intrinsics/src/elpa2_kernels/elpa2_kernels_real_simple.f90
View file @
fbb0e6eb
...
...
@@ -14,7 +14,7 @@
!
! --------------------------------------------------------------------------------------------------
subroutine
double_hh_trafo
_2hv
(
q
,
hh
,
nb
,
nq
,
ldq
,
ldh
)
subroutine
double_hh_trafo
(
q
,
hh
,
nb
,
nq
,
ldq
,
ldh
)
implicit
none
...
...
ELPA_2011.12.Intrinsics/src/elpa2_kernels/elpa2_tum_kernels_real_sse-avx_2hv.c
View file @
fbb0e6eb
...
...
@@ -34,12 +34,12 @@ void hh_trafo_kernel_8_SSE_2hv(double* q, double* hh, int nb, int ldq, int ldh,
void
hh_trafo_kernel_12_SSE_2hv
(
double
*
q
,
double
*
hh
,
int
nb
,
int
ldq
,
int
ldh
,
double
s
);
#endif
void
double_hh_trafo_
2hv_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
);
void
double_hh_trafo_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
);
#if 0
void double_hh_trafo_
2hv_
fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh);
void double_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh);
#endif
void
double_hh_trafo_
2hv_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
)
void
double_hh_trafo_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
)
{
int
i
;
int
nb
=
*
pnb
;
...
...
@@ -115,7 +115,7 @@ void double_hh_trafo_2hv_(double* q, double* hh, int* pnb, int* pnq, int* pldq,
}
#if 0
void double_hh_trafo_
2hv_
fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
void double_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
{
int i;
int nb = *pnb;
...
...
ELPA_2011.12.Intrinsics/src/elpa2_kernels/elpa2_tum_kernels_real_sse-avx_4hv.c
View file @
fbb0e6eb
...
...
@@ -33,12 +33,12 @@ __forceinline void hh_trafo_kernel_4_SSE_4hv(double* q, double* hh, int nb, int
__forceinline
void
hh_trafo_kernel_6_SSE_4hv
(
double
*
q
,
double
*
hh
,
int
nb
,
int
ldq
,
int
ldh
,
double
s_1_2
,
double
s_1_3
,
double
s_2_3
,
double
s_1_4
,
double
s_2_4
,
double
s_3_4
);
#endif
void
double
_hh_trafo_
4hv_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
);
void
quad
_hh_trafo_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
);
#if 0
void
double
_hh_trafo_
4hv_
fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh);
void
quad
_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh);
#endif
void
double
_hh_trafo_
4hv_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
)
void
quad
_hh_trafo_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
)
{
int
i
;
int
nb
=
*
pnb
;
...
...
@@ -134,7 +134,7 @@ void double_hh_trafo_4hv_(double* q, double* hh, int* pnb, int* pnq, int* pldq,
}
#if 0
void
double
_hh_trafo_
4hv_
fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
void
quad
_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
{
int i;
int nb = *pnb;
...
...
ELPA_2011.12.Intrinsics/src/elpa2_kernels/elpa2_tum_kernels_real_sse-avx_6hv.c
View file @
fbb0e6eb
...
...
@@ -31,12 +31,12 @@ void hh_trafo_kernel_2_SSE_6hv(double* q, double* hh, int nb, int ldq, int ldh,
void
hh_trafo_kernel_4_SSE_6hv
(
double
*
q
,
double
*
hh
,
int
nb
,
int
ldq
,
int
ldh
,
double
*
scalarprods
);
#endif
void
double
_hh_trafo_
6hv_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
);
void
hexa
_hh_trafo_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
);
#if 0
void
double
_hh_trafo_
6hv_
fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh);
void
hexa
_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh);
#endif
void
double
_hh_trafo_
6hv_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
)
void
hexa
_hh_trafo_
(
double
*
q
,
double
*
hh
,
int
*
pnb
,
int
*
pnq
,
int
*
pldq
,
int
*
pldh
)
{
int
i
;
int
nb
=
*
pnb
;
...
...
@@ -206,7 +206,7 @@ void double_hh_trafo_6hv_(double* q, double* hh, int* pnb, int* pnq, int* pldq,
}
#if 0
void
double
_hh_trafo_
6hv_
fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
void
hexa
_hh_trafo_fast_(double* q, double* hh, int* pnb, int* pnq, int* pldq, int* pldh)
{
int i;
int nb = *pnb;
...
...
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