Category: News
-
New Adaptyst release delayed
Featured image based on work from: Bernerlover, Wikimedia Commons The modular redesign of Adaptyst mentioned several times in previous news items and roadmap has been mostly completed and it involved many significant code changes. The finished work includes the CPU “perf” module, the new output format of performance analysis results, and DaCe/SDFG integration. However, due…
-
Preliminary NVIDIA GPU support will be available by 23 June
Featured image based on work from: Bernerlover, Wikimedia Commons The first iteration of the modular design and DaCe integration will feature a new module inspecting NVIDIA GPU activity of codes. Therefore, by 23 June, you will be able to start using Adaptyst for analysing performance of your workflows running on NVIDIA GPUs. Under the hood,…
-
Current status of the DaCe integration and modular design work
Featured image based on work from: Bernerlover, Wikimedia Commons The news this week is short: the work on the DaCe integration and modular design is well under way and we are still on track for releasing their first iteration by 23 June. Stay tuned!
-
AMD CPU roofline profiling support currently not available
Featured image based on work from: Bernerlover, Wikimedia Commons When cache-aware roofline profiling was introduced in Adaptyst, we could only confirm its support for Intel CPUs as we had solely Intel-CPU-based machines for testing the feature. Therefore, the AMD CPU support was marked as experimental. Luckily, we have recently secured access to an AMD-CPU-based machine.…
-
First iteration of modular design and DaCe integration coming in June + CVMFS update
Featured image based on work from: Bernerlover, Wikimedia Commons We have recently started working on the first iteration of the modular design of Adaptyst and DaCe integration, as touched upon in our roadmap and the news from 14/03/2025. If there are no unexpected serious technical obstacles, it will become available for you to test by…
-
New versioning scheme and off-CPU timeline sampling settings
Featured image based on work from: Bernerlover, Wikimedia Commons To make versioning more consistent, the new scheme has been introduced along with git tags and GitHub releases. The current list of releases can be found here: Adaptyst and Adaptyst Analyser. If you have installed Adaptyst Analyser before, you may have to remove it completely before…
-
Working on optimising timeline rendering in Adaptyst Analyser + Easter break in news
Featured image based on work from: Bernerlover, Wikimedia Commons In some more complex cases, e.g. a profiled program with lots of interleaving on-CPU and off-CPU activities, timeline rendering in Adaptyst Analyser can be very resource-heavy, negatively impacting the user experience. This problem can be currently counteracted by turning on off-CPU timeline sampling, but the option…
-
Making the build system fully-CMake-based and preparing a CVMFS release
Featured image based on work from: Bernerlover, Wikimedia Commons In order to streamline the installation-from-source process of Adaptyst, we have moved all build-related steps to the already-existing CMake setup and discontinued the build.sh and install.sh scripts along with the “perf” git submodule. We have also added an option to change patched “perf” installation settings and…
-
Updates this week: stack filtering at profile-time and new flame graph features
Featured image based on work from: Bernerlover, Wikimedia Commons Here are the latest development updates for this week:
-
Full cache-aware roofline support is available in Adaptyst now
Featured image based on work from: Bernerlover, Wikimedia Commons As part of the SYCLOPS project, we collaborate with Prof. Aleksandar Ilic’s group at INESC-ID on integrating their work on cache-aware roofline profiling (including the CARM Tool) into Adaptyst. Partial support of roofline has already existed in our tool for some time, namely displaying performance bounds…