Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
openfpm_pdata
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Build
Pipelines
Jobs
Pipeline schedules
Artifacts
Deploy
Releases
Model registry
Operate
Environments
Monitor
Incidents
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
Sbalzarini Lab
Software
Parallel Computing
OpenFPM
openfpm_pdata
Commits
0c4f0f5b
Commit
0c4f0f5b
authored
1 year ago
by
Pietro Incardona
Browse files
Options
Downloads
Patches
Plain Diff
Adding missing files
parent
a9ba989e
No related branches found
Branches containing commit
No related tags found
No related merge requests found
Changes
2
Hide whitespace changes
Inline
Side-by-side
Showing
2 changed files
example/Performance/Syncthreads_kokkos_benchmark/Makefile
+75
-0
75 additions, 0 deletions
example/Performance/Syncthreads_kokkos_benchmark/Makefile
example/Performance/Syncthreads_kokkos_benchmark/team.cpp
+77
-0
77 additions, 0 deletions
example/Performance/Syncthreads_kokkos_benchmark/team.cpp
with
152 additions
and
0 deletions
example/Performance/Syncthreads_kokkos_benchmark/Makefile
0 → 100644
+
75
−
0
View file @
0c4f0f5b
include
example.mk
### This is a trick to avoid "Command not found if you no not have NVCC compiler". In practice the normal C++ compiler is used
### internally the example disable with the preprocessor its code if not compiled with nvcc
CUDA_CC
=
CUDA_CC_LINK
=
ifdef
HIP
CUDA_CC
=
hipcc
CUDA_OPTIONS
=
-D__NVCC__
-D__HIP__
-DCUDART_VERSION
=
11000
-D__CUDACC__
-D__CUDACC_VER_MAJOR__
=
11
-D__CUDACC_VER_MINOR__
=
0
-D__CUDACC_VER_BUILD__
=
0
LIBS_SELECT
=
$(
LIBS
)
CC
=
hipcc
CUDA_CC_LINK
=
hipcc
else
ifdef
CUDA_ON_CPU
CUDA_CC
=
mpic++
-x
c++
$(
INCLUDE_PATH
)
INCLUDE_PATH_NVCC
=
CUDA_CC_LINK
=
mpic++
CUDA_OPTIONS
=
-D__NVCC__
-DCUDART_VERSION
=
11000
LIBS_SELECT
=
$(
LIBS
)
else
ifeq
(, $(shell which nvcc))
CUDA_CC
=
mpic++
-x
c++
$(
INCLUDE_PATH
)
INCLUDE_PATH_NVCC
=
CUDA_CC_LINK
=
mpic++
CUDA_OPTIONS
=
LIBS_SELECT
=
$(
LIBS
)
else
CUDA_CC
=
nvcc
-ccbin
=
mpic++
CUDA_CC_LINK
=
nvcc
-ccbin
=
mpic++
CUDA_OPTIONS
=
-use_fast_math
-arch
=
sm_61
-lineinfo
LIBS_SELECT
=
$(
LIBS_NVCC
)
endif
endif
endif
ifeq
($(PROFILE),ON)
CUDA_CC
=
scorep
--nocompiler
--cuda
--mpp
=
mpi nvcc
-ccbin
=
mpic++
CUDA_CC_LINK
=
scorep
--nocompiler
--cuda
--mpp
=
mpi nvcc
-ccbin
=
mpic++
else
CUDA_CC
:=
$(
CUDA_CC
)
CUDA_CC_LINK
:=
$(
CUDA_CC_LINK
)
endif
LDIR
=
OPT
=
OBJ
=
main.o
cuda
:
sync_test
:
OPT += -DTEST_RUN
sync_dlb_test
:
sync_dlb
%.o
:
%.cu
$(
CUDA_CC
)
-O3
$(
OPT
)
$(
CUDA_OPTIONS
)
-g
-c
--std
=
c++14
-o
$@
$<
$(
INCLUDE_PATH_NVCC
)
%.o
:
%.cpp
$(
CC
)
-O0
$(
OPT
)
-g
-c
--std
=
c++14
-o
$@
$<
$(
INCLUDE_PATH
)
cuda
:
$(OBJ)
$(
CUDA_CC_LINK
)
-o
$@
$^
$(
CFLAGS
)
$(
LIBS_PATH
)
$(
LIBS_SELECT
)
sync_dlb2
:
$(OBJ)
$(
CUDA_CC_LINK
)
-o
$@
$^
$(
CFLAGS
)
$(
LIBS_PATH
)
$(
LIBS_SELECT
)
all
:
cuda
run
:
sync_dlb_test
mpirun
--oversubscribe
-np
2 ./sync_dlb
.PHONY
:
clean all run
clean
:
rm
-f
*
.o
*
~ core sync_dlb
This diff is collapsed.
Click to expand it.
example/Performance/Syncthreads_kokkos_benchmark/team.cpp
0 → 100644
+
77
−
0
View file @
0c4f0f5b
#include
<Kokkos_Core.hpp>
#include
<cstdio>
int
main
(
int
argc
,
char
*
argv
[])
{
Kokkos
::
initialize
(
argc
,
argv
);
printf
(
"LayoutRight
\n
"
);
{
//////////////////////// TEAMS ////////////////////////
using
Kokkos
::
TeamPolicy
;
using
Kokkos
::
parallel_for
;
typedef
TeamPolicy
<
Kokkos
::
OpenMP
>::
member_type
member_type
;
// Create an instance of the policy
int
team_sz
=
1
;
int
sz
=
512
;
TeamPolicy
<
Kokkos
::
OpenMP
>
policy
(
sz
*
sz
,
team_sz
);
// Launch a kernel
Kokkos
::
fence
();
Kokkos
::
Timer
timer
;
parallel_for
(
policy
,
KOKKOS_LAMBDA
(
member_type
team_member
)
{
// Calculate a global thread id
int
k
=
team_member
.
league_rank
()
*
team_member
.
team_size
()
+
team_member
.
team_rank
();
// Calculate the sum of the global thread ids of this team
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
team_member
.
team_barrier
();
// Atomically add the value to a global value
});
Kokkos
::
fence
();
double
time
=
timer
.
seconds
();
std
::
cout
<<
"TIME: "
<<
time
/
(
sz
*
sz
*
team_sz
*
24
)
*
1e9
<<
" ns"
<<
std
::
endl
;
///////////////////////////////////////////////////////
}
printf
(
"LayoutLeft
\n
"
);
Kokkos
::
finalize
();
}
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
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!
Save comment
Cancel
Please
register
or
sign in
to comment