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
be4b61ae
Commit
be4b61ae
authored
Jun 07, 2016
by
Andreas Marek
Browse files
Double precision AVX512 block6 kernel
parent
366d0f43
Changes
2
Pipelines
1
Expand all
Hide whitespace changes
Inline
Side-by-side
src/elpa2_kernels/elpa2_kernels_real_avx512_6hv_double_precision.c
View file @
be4b61ae
This diff is collapsed.
Click to expand it.
src/mod_compute_hh_trafo_real.F90
View file @
be4b61ae
...
...
@@ -695,6 +695,64 @@ module compute_hh_trafo_real
#endif /* WITH_NO_SPECIFIC_REAL_KERNEL */
#endif /* WITH_REAL_AVX_BLOCK6_KERNEL || WITH_REAL_AVX2_BLOCK6_KERNEL */
#if defined(WITH_REAL_AVX512_BLOCK6_KERNEL)
#if defined(WITH_NO_SPECIFIC_REAL_KERNEL)
if
((
THIS_REAL_ELPA_KERNEL
.eq.
REAL_ELPA_KERNEL_AVX512_BLOCK6
))
then
#endif /* WITH_NO_SPECIFIC_REAL_KERNEL */
! X86 INTRINSIC CODE, USING 6 HOUSEHOLDER VECTORS
do
j
=
ncols
,
6
,
-6
w
(:,
1
)
=
bcast_buffer
(
1
:
nbw
,
j
+
off
)
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
)
w
(:,
5
)
=
bcast_buffer
(
1
:
nbw
,
j
+
off
-4
)
w
(:,
6
)
=
bcast_buffer
(
1
:
nbw
,
j
+
off
-5
)
#ifdef WITH_OPENMP
call
hexa_hh_trafo_real_avx512_6hv_double
(
c_loc
(
a
(
1
,
j
+
off
+
a_off
-5
,
istripe
,
my_thread
)),
w
,
&
nbw
,
nl
,
stripe_width
,
nbw
)
#else
call
hexa_hh_trafo_real_avx512_6hv_double
(
c_loc
(
a
(
1
,
j
+
off
+
a_off
-5
,
istripe
)),
w
,
&
nbw
,
nl
,
stripe_width
,
nbw
)
#endif
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
)
#ifdef WITH_OPENMP
call
quad_hh_trafo_real_avx512_4hv_double
(
c_loc
(
a
(
1
,
jj
+
off
+
a_off
-3
,
istripe
,
my_thread
)),
w
,
&
nbw
,
nl
,
stripe_width
,
nbw
)
#else
call
quad_hh_trafo_real_avx512_4hv_double
(
c_loc
(
a
(
1
,
jj
+
off
+
a_off
-3
,
istripe
)),
w
,
&
nbw
,
nl
,
stripe_width
,
nbw
)
#endif
enddo
do
jjj
=
jj
,
2
,
-2
w
(:,
1
)
=
bcast_buffer
(
1
:
nbw
,
jjj
+
off
)
w
(:,
2
)
=
bcast_buffer
(
1
:
nbw
,
jjj
+
off
-1
)
#ifdef WITH_OPENMP
call
double_hh_trafo_real_avx512_2hv_double
(
c_loc
(
a
(
1
,
jjj
+
off
+
a_off
-1
,
istripe
,
my_thread
)),
&
w
,
nbw
,
nl
,
stripe_width
,
nbw
)
#else
call
double_hh_trafo_real_avx512_2hv_double
(
c_loc
(
a
(
1
,
jjj
+
off
+
a_off
-1
,
istripe
)),
&
w
,
nbw
,
nl
,
stripe_width
,
nbw
)
#endif
enddo
#ifdef WITH_OPENMP
if
(
jjj
==
1
)
call
single_hh_trafo_real_cpu_openmp_double
(
a
(
1
:
stripe_width
,
1
+
off
+
a_off
:
1
+
off
+
a_off
+
nbw
-1
,
&
istripe
,
my_thread
),
&
bcast_buffer
(
1
:
nbw
,
off
+1
),
nbw
,
nl
,
stripe_width
)
#else
if
(
jjj
==
1
)
call
single_hh_trafo_real_cpu_double
(
a
(
1
:
stripe_width
,
1
+
off
+
a_off
:
1
+
off
+
a_off
+
nbw
-1
,
istripe
),
&
bcast_buffer
(
1
:
nbw
,
off
+1
),
nbw
,
nl
,
stripe_width
)
#endif
#if defined(WITH_NO_SPECIFIC_REAL_KERNEL)
endif
#endif /* WITH_NO_SPECIFIC_REAL_KERNEL */
#endif /* WITH_REAL_AVX512_BLOCK6_KERNEL */
endif
! GPU_KERNEL
#ifdef WITH_OPENMP
...
...
@@ -1304,6 +1362,67 @@ module compute_hh_trafo_real
#endif /* WITH_NO_SPECIFIC_REAL_KERNEL */
#endif /* WITH_REAL_AVX_BLOCK6_KERNEL || WITH_REAL_AVX2_BLOCK6_KERNEL */
#if defined(WITH_REAL_AVX512_BLOCK6_KERNEL)
#if defined(WITH_NO_SPECIFIC_REAL_KERNEL)
if
((
THIS_REAL_ELPA_KERNEL
.eq.
REAL_ELPA_KERNEL_AVX512_BLOCK6
))
then
#endif /* WITH_NO_SPECIFIC_REAL_KERNEL */
! X86 INTRINSIC CODE, USING 6 HOUSEHOLDER VECTORS
do
j
=
ncols
,
6
,
-6
w
(:,
1
)
=
bcast_buffer
(
1
:
nbw
,
j
+
off
)
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
)
w
(:,
5
)
=
bcast_buffer
(
1
:
nbw
,
j
+
off
-4
)
w
(:,
6
)
=
bcast_buffer
(
1
:
nbw
,
j
+
off
-5
)
#ifdef WITH_OPENMP
call
hexa_hh_trafo_real_avx512_6hv_single
(
c_loc
(
a
(
1
,
j
+
off
+
a_off
-5
,
istripe
,
my_thread
)),
w
,
&
nbw
,
nl
,
stripe_width
,
nbw
)
#else
call
hexa_hh_trafo_real_avx512_6hv_single
(
c_loc
(
a
(
1
,
j
+
off
+
a_off
-5
,
istripe
)),
w
,
&
nbw
,
nl
,
stripe_width
,
nbw
)
#endif
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
)
#ifdef WITH_OPENMP
call
quad_hh_trafo_real_avx512_4hv_single
(
c_loc
(
a
(
1
,
jj
+
off
+
a_off
-3
,
istripe
,
my_thread
)),
w
,
&
nbw
,
nl
,
stripe_width
,
nbw
)
#else
call
quad_hh_trafo_real_avx512_4hv_single
(
c_loc
(
a
(
1
,
jj
+
off
+
a_off
-3
,
istripe
)),
w
,
&
nbw
,
nl
,
stripe_width
,
nbw
)
#endif
enddo
do
jjj
=
jj
,
2
,
-2
w
(:,
1
)
=
bcast_buffer
(
1
:
nbw
,
jjj
+
off
)
w
(:,
2
)
=
bcast_buffer
(
1
:
nbw
,
jjj
+
off
-1
)
#ifdef WITH_OPENMP
call
double_hh_trafo_real_avx512_2hv_single
(
c_loc
(
a
(
1
,
jjj
+
off
+
a_off
-1
,
istripe
,
my_thread
)),
&
w
,
nbw
,
nl
,
stripe_width
,
nbw
)
#else
call
double_hh_trafo_real_avx512_2hv_single
(
c_loc
(
a
(
1
,
jjj
+
off
+
a_off
-1
,
istripe
)),
&
w
,
nbw
,
nl
,
stripe_width
,
nbw
)
#endif
enddo
#ifdef WITH_OPENMP
if
(
jjj
==
1
)
call
single_hh_trafo_real_cpu_openmp_single
(
a
(
1
:
stripe_width
,
1
+
off
+
a_off
:
1
+
off
+
a_off
+
nbw
-1
,
&
istripe
,
my_thread
),
&
bcast_buffer
(
1
:
nbw
,
off
+1
),
nbw
,
nl
,
stripe_width
)
#else
if
(
jjj
==
1
)
call
single_hh_trafo_real_cpu_single
(
a
(
1
:
stripe_width
,
1
+
off
+
a_off
:
1
+
off
+
a_off
+
nbw
-1
,
istripe
),
&
bcast_buffer
(
1
:
nbw
,
off
+1
),
nbw
,
nl
,
stripe_width
)
#endif
#if defined(WITH_NO_SPECIFIC_REAL_KERNEL)
endif
#endif /* WITH_NO_SPECIFIC_REAL_KERNEL */
#endif /* WITH_REAL_AVX512_BLOCK6_KERNEL */
endif
! GPU_KERNEL
#ifdef WITH_OPENMP
...
...
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