variorum
variorum copied to clipboard
LLNL
Vendor-neutral library for exposing power and performance features across diverse architectures
The existing approach requires the developer to pre-allocate memory for the batch operations, which opens up the possibility of multiple hard-to-find bugs. Memory allocation should be done automatically and only...
Current updates from @masterleinad are here: https://github.com/kokkos/kokkos-tools/pull/265. Once the energy API in Variorum is updated we can update the associated Kokkos PR as needed too.
# Description Extend the new energy API from v0.8 to include GPU energy on NVIDIA and AMD GPUs. **6/26: This is WIP and won't compile/work just yet.** To Do 6/27:...
Currently, we're sampling every 250ms -- as a result, when we encounter a code that doesn't do much work and takes less than 250ms, energy deltas get reported as zero....
Update the API to support nested calls in general, especially in Caliper-like tools. This might be useful for the Kokkos-update as well. Merge #559 and #563 first, and then add...
Tagging @kshoga1 so she can follow this issue as needed.
This issue will track the Variorum service update in Caliper.
We want to be able to support use cases where `var_monitor` just runs in the background on each node (with sleep) and users can specify a `var_monitor start` and `var_monitor...
Update the CMake build in variorum to set the `C_FLAGS` and linker options to include address sanitizer checks.
# Description Without this fix NVML can't be found via standard CMake variables. With this, I can do ``` cmake -DVARIORUM_WITH_NVIDIA_GPU=ON -DNVML_ROOT=/usr/local/cuda-11.7/targets/x86_64-linux/ ../src ``` do find `NVML`. Requiring finding the...
