Support for a single memory space writeable by both GPU and CPU #242
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…ides appropriate synchronizations when needed instead of performing any memory copies.
robinson96
changed the title
|
This looks fine, but can you remove the tpl export change? We are going to bump to latest BLT and use the new TPL macros. |
Fixed. |
|
It sounds like the CZ is back up now, so is there an easy way to retrigger the pipelines other than adding a commit? |
liu15
approved these changes
Feb 14, 2024
adayton1
added a commit
that referenced
this pull request
Mar 5, 2024
* Depend on develop version of TPLs * Tweak version constraints * Update Umpire package * Fix syntax error in Umpire package * Update packages for CHAI and Umpire * Depend on newer camp * Update versions for RAJA and Umpire * Fixes for libcpp builds * Add SLURM_OVERLAP variable * Add greater control for when CHAI is a submodule * Fix hip build with resource manager disabled * Depend on camp@main * Missed one reference to camp@master * Bug fixes for HIP and managed_ptr * Clean up managed_ptr.hpp * Fix host build warning * Customize block size for HIP build * Add a managed_ptr example * Fix HIP kernel launches * Add reproducer for HIP managed_ptr errors * Clean up managed_ptr reproducer * Add a simpler reproducer * Add simplest reproducer * Fix macro * Fix mutex lock in evict() that overlaps with lock in free() * initial changes to add CHAI_ cmake prefix * fixing cmake option command syntax * trying to fix azure pipeline image and raja plugin flag * make edits to azure pipeline file * Switching from quartz to ruby in gitlab CI * Updating umpire camp dependency * Release v2022.03.0 (#208) * Update Spack packages for CI (#210) * Add CHAI + RAJA launch integration test. (#219) * Release v2022.10.0 (#217) * Make operator overloads bind in more cases (#222) * MigrateInner on CHAI Copyable for PINNED / UM (#223) * Use shared CI (#213) * Deprecate API for synchronization after device kernels (#228) * setExecutionSpace thread safety(#230) Co-authored-by: Neela Kausik <[email protected]> * Release v2023.06.0 (#232) * [Woptim] Update shared ci to v2023.08.0 (#236) * Remove explicit device synchronization (#235) * Update radiuss-shared-ci to new release (with radiuss-spack-configs) (#238) * Use free instead of realloc to zero size (#239) * [Woptim] shared ci 2023.12.0 (#241) * Update Shared CI and Spack Configs to add poodle * Activate poodle subpipeline * Reduce allocated time * Fine tune allocation duration * Fine tune allocation duration in CI * Comment alloc command choices * Update RSC to main * Fix missing poodle alloc info, update to radiuss Shared CI 2023.12.0, sync build_and_test script with RAJA * Update radiuss-spack-configs * CI updates (#244) * Updates radiuss-spack-configs to commit a37c3e1 * Changes to match the CI of other projects * Support for a single memory space writeable by both GPU and CPU (#242) * Support for a single memory space writeable by both GPU and CPU. Provides appropriate synchronizations when needed instead of performing any memory copies. --------- Co-authored-by: Alan Dayton <[email protected]> * Clean up CHAI dependency handling * Revert "Clean up CHAI dependency handling" This reverts commit 5afb66a. * Clean up CMake in CHAI (#247) * Update copyright to 2024 (#248) * Update to BLT v0.6.1 (#243) * Update to BLT v0.6.1 * Update to RAJA v2024.02.0 * Update to Umpire 2024.02.0 * Use new way of exporting tpls * Disable failing container builds since container images are being revamped * Allow failure for the job using libcpp - doesn't play nicely with fmt at the moment * Use branch of radiuss-spack-configs to fix spack builds * Add install tests * Add a file with release notes * Ignoring runner issues on tioga since the jobs passed previously * [Woptim] update spack + update to rocm 5.7.1 (#245) * Update Spack version * Update RADIUSS Spack Configs with February 2024 release changes --------- Co-authored-by: Alan Dayton <[email protected]> * Update version numbers --------- Co-authored-by: David Beckingsale <[email protected]> Co-authored-by: Benjamin T. Liu <[email protected]> Co-authored-by: Kristi <[email protected]> Co-authored-by: Kristi <[email protected]> Co-authored-by: Arturo Vargas <[email protected]> Co-authored-by: robinson96 <[email protected]> Co-authored-by: Adrien Bernede <[email protected]> Co-authored-by: Neela Kausik <[email protected]> Co-authored-by: Ben Liu <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
SUMMARY
Provides appropriate synchronizations when needed instead of performing any memory copies.
DETAILS
The configuration variable CHAI_THIN_GPU_ALLOCATE, when enabled in combination with CHAI_DISABLE_RM set to True, will configure CHAI to only perform necessary synchronization in lieu of necessary memory motion.
Changes ManagedArray::m_size to ManagedArray::m_elems - now tracks #of bytes instead of number of entries in the array. This makes the ManagedArray more robust to type conversions.