parsec
parsec copied to clipboard
ICLDisco
PaRSEC is a generic framework for architecture aware scheduling and management of micro-tasks on distributed, GPU accelerated, many-core heterogeneous architectures. PaRSEC assigns computation threads...
Lazy/dynamic allocation of infos can be convenient for the users, but also detrimental to predictive behaviors. Examples of drawbacks are when testing memory constrained setups (e.g., when the test uses...
Add an mca control that disable cpu-gpu load balancing (disabled by default). This can be overriden in multiple ways: * setting mca param will enable cpu load balancing globally *...
- Make sure the parsec context is started only after the taskpool is created as the taskpool will register active message handlers for the first time via the user-triggered option...
## Description PaRSEC numbers its devices differently from the rest of world (starting with 0 for host, possibly 1 for recursive, 2... for devices). CUDA and HIP start numbering devices...
## Describe the bug Some flags that are valid for CC are not valid for the linker on OSX (this may also impact other platforms, although not seen in the...
## Description On some systems (e.g., Frontier) GPU memory is directly attached to the NIC, and motioning data to the host before transfer (and receiving in host memory then moving...
> the CI instances are tagged with their devices. As an example serotonin is tagger with `gpu_amd` while guyot is tagged with `gpu_nvidia`. The CI should use these tags to...
The `apply` releases user-allocated memory, which may cause a memory leak. Users need to manage that instead of PaRSEC. https://github.com/ICLDisco/parsec/blob/00470145f3ce392853871c76d3200b7cb8aa6f05/parsec/data_dist/matrix/apply_wrapper.c#L115-L116
This is a refactoring of the current CI jobs such that the jobs no longer rely on building the dependencies using spack. The script instead load the dependencies from a...
