chore: Update blt, RAJA, CHAI, Caliper, adiak, conduit #3218
Conversation
rrsettgast
added
ci: run CUDA builds
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## develop #3218 +/- ##
===========================================
- Coverage 55.73% 55.72% -0.01%
===========================================
Files 1041 1041
Lines 88562 88559 -3
===========================================
- Hits 49360 49353 -7
- Misses 39202 39206 +4 ☔ View full report in Codecov by Sentry. |
| @@ -53,7 +53,9 @@ if(EXISTS ${GEOSX_TPL_DIR}/vtk) | |||
| endif() | |||
|
|
|||
| if(EXISTS ${GEOSX_TPL_DIR}/fmt) | |||
| set(FMT_DIR ${GEOSX_TPL_DIR}/fmt CACHE PATH "" FORCE) | |||
| # set(FMT_DIR ${GEOSX_TPL_DIR}/fmt CACHE PATH "" FORCE) | |||
| set(FMT_DIR ${GEOSX_TPL_DIR}/chai CACHE PATH "" FORCE) | |||
There was a problem hiding this comment.
Is FMT_DIR pointing to chai ?
There was a problem hiding this comment.
yes. there is a conflict. A lot of the TPL's have fmt integrated, and umpire which is included in CHAI changed the way they include it so now there can be a conflict....it is annoying to say the least.
| message( " ----> HDF5 version ${HDF5_VERSION}") | ||
|
|
||
| get_target_property(HDF5_INCLUDE_DIRS HDF5::HDF5 INTERFACE_INCLUDE_DIRECTORIES) | ||
| set_target_properties(HDF5::HDF5 PROPERTIES INTERFACE_SYSTEM_INCLUDE_DIRECTORIES "${HDF5_INCLUDE_DIRS}") |
There was a problem hiding this comment.
Is it necessary to put the includes into system interface ?
There was a problem hiding this comment.
We do this so that they are included with -isystem instead of -I to suppress any compiler warnings.
rrsettgast
changed the title
…rc/coreComponents/dataRepository/unitTests
| // events.emplace_back( forAll< parallelDeviceAsyncPolicy<> >( stream, var.size(), [=] GEOS_DEVICE ( localIndex ii ) | ||
| forAll< parallelDevicePolicy<> >( var.size(), [=] GEOS_DEVICE ( localIndex ii ) | ||
| { | ||
| reinterpret_cast< std::remove_const_t< T > * >( buffer )[ ii ] = var.data()[ ii ]; | ||
| } ) ); | ||
| } ); | ||
| //); |
There was a problem hiding this comment.
@wrtobin Why is this not something like:
std::remove_const_t <T> * const lhs = reinterpret_cast< std::remove_const_t< T > * >( buffer );
forAll< parallelDevicePolicy<> >( var.size(), [=] GEOS_DEVICE ( localIndex ii )
{
lhs[ ii ] = var[ ii ];
} );
There was a problem hiding this comment.
Thats a fine refactor, except we do need var.data()[ ii ] since we're packing arbitrary multidimensional slices and just want the underlying array buffer, which inside the loop is on device and outside the loop is on host.
Since the lhs is a pinned buffer moving the cast outside the loop should be fine and might be more clear, but the performance should be the same since the cast is a compile-time operation on memory location/alignment more than anything else.
rrsettgast
force-pushed
the
update/update_to_blt_0.6.2
branch
from
cf21dbe
to
89c3ba2
Compare
rrsettgast
changed the title
* update blt to 0.6.2 * update RAJA and CHAI to v2024.02.2 * update Caliper to 2.11.0 *update adiak to 0.4.0 * update conduit to 0.9.2 * change to new HDF import. use umpires fmt and add extra dependency to executable * changes for change in raja interface * move src/coreComponents/unitTests/dataRepository/testPacking.cpp to src/coreComponents/dataRepository/unitTests * switched to syncronous packing/upacking to avoid test failures on CI image running on streak
* update blt to 0.6.2 * update RAJA and CHAI to v2024.02.2 * update Caliper to 2.11.0 *update adiak to 0.4.0 * update conduit to 0.9.2 * change to new HDF import. use umpires fmt and add extra dependency to executable * changes for change in raja interface * move src/coreComponents/unitTests/dataRepository/testPacking.cpp to src/coreComponents/dataRepository/unitTests * switched to syncronous packing/upacking to avoid test failures on CI image running on streak
GEOS-DEV/LvArray#323
GEOS-DEV/thirdPartyLibs#275