Use consistent types in Kokkos_OpenMPTarget_Parallel.hpp #3703
Conversation
masterleinad
force-pushed
the
fix_threadrange_ompt_types
branch
from
d12494c
to
b85ef20
Compare
|
Do you think you can come up with a compile-only test that would catch the issue? |
masterleinad
force-pushed
the
fix_threadrange_ompt_types
branch
2 times, most recently
from
df94387
to
94c7c63
Compare
|
|
masterleinad
force-pushed
the
fix_threadrange_ompt_types
branch
from
d80a6d8
to
1ebb840
Compare
| league_size) {} | ||
| : m_flags( | ||
| Kokkos::view_alloc(Kokkos::WithoutInitializing, "flags"), | ||
| #ifdef KOKKOS_ENABLE_OPENMPTARGET |
There was a problem hiding this comment.
Let's ask @rgayatri23 if this restriction is considered to be permanent or not.
There was a problem hiding this comment.
Hopefully the restriction of needing at-least 32 threads per team is not permanent. We should probably add a FIXME_OPENMPTARGET everywhere we do this so that in the future if we can avoid this restriction, we can easily track these changes.
| @@ -65,6 +65,8 @@ TEST(TEST_CATEGORY, team_for) { | |||
| 1000); | |||
| } | |||
|
|
|||
| // FIXME_OPENMPTARGET wrong results | |||
| #ifndef KOKKOS_ENABLE_OPENMPTARGET | |||
| TEST(TEST_CATEGORY, team_reduce) { | |||
| TestTeamPolicy<TEST_EXECSPACE, | |||
| Kokkos::Schedule<Kokkos::Static> >::test_reduce(0); | |||
There was a problem hiding this comment.
Is there a reason why OpenMPTarget has a different problem size?
There was a problem hiding this comment.
What do you mean? For me, this test just gives wrong results so I disabled it for now.
There was a problem hiding this comment.
Oh Ok. I saw test_reduce(0) and test_reduce(1000), so assumed that the OPENMPTARGET backend is doing something different.
|
LGTM |
|
Please cleanup history (the enable test, disable and reenable at least) and I'll approve |
masterleinad
force-pushed
the
fix_threadrange_ompt_types
branch
from
18c1deb
to
299aed7
Compare
Done. |
Fixes #3702. I checked that all other backends follow this convention.