celerity-runtime
celerity-runtime copied to clipboard
celerity
High-level C++ for Accelerator Clusters
After being stuck on [a bug](https://github.com/celerity/celerity-runtime/actions/runs/2924269790/jobs/4663659961) for some time, recent versions of DPC++ are compatible with Celerity again. This PR bumps the minimum required version to a commit from the...
So far we've used clang-tidy only to enforce naming conventions. This PR enables a whole set of additional checks as suggested in #127. Notably this includes [`misc-const-correctness`](https://clang.llvm.org/extra/clang-tidy/checks/misc/const-correctness.html), a check that...
In addition to execution commands, scheduling a task will generate push / await-push / reduction cascades on participating nodes. So far, all of these implicit commands are `MPI_Sent` to workers...
This is the second PR in the Instruction Graph series, and the first one to touch the execution side of the new runtime. It adds the necessary infrastructure to perform...
This pull request adds a divergence checking mechanism for tasks. It does so by periodically gathering hashes of all tasks from task_recording and comparing them. When a divergence is detected...
RFC, based on #182. A common parallelization pattern, e.g. in dense matrix-vector products, is to map 1D thread-ids to the rows of a 2D matrix while iterating over all columns...
The IDAG runtime needs updates to both the minimum hipSYCL and the minimum DPC++ version for proper reduction support and various bugfixes. This PR bumps both versions and adjusts documentation...
This PR performs some small refactorings necessary for me to introduce the new executor- and backend types in another standalone PR _without_ replacing the entire runtime. It is best reviewed...
This PR upstreams and tests two threading facilities required for IDAG execution: The in-order `thread_queue` and the `double_buffered_queue` for fast communication between threads. - `async_event` gains the ability to return...