Releases: ledatelescope/bifrost
v0.11.0
Changelog:
- Dropped support for Python2
- Merged the ibverb-support branch that brings in a new, more flexible packet capture interface
- Improved documentation about packet capture
- Improved documentation about CUDA support
- Improved documentation build process
- Various fixes for CUDA 12.x releases
- Added support for gcc 13
- Changed the file locking so that the user is warned if a block is waiting to acquire the lock
- Various deprecation cleanups
v0.10.0
Changelog:
- Switched over to an autotools-based build system
- Added a .m4 file to help other autotools-based software find Bifrost
- Added a pkg-config file for Bifrost
- Made the Python API compatible with PEP479
- Added support for the cuda_managed space on Pascal and later GPUs
- Added support for ci32 in bf.map
- Added support for converting a bifrost.ndarray to/from a cupy.ndarray
- Switched to ctypesgen 1.0.2
- Added an example Python notebook that can run on Google Colab
- Added support for 'python -m bifrost.version'
- Removed bifrost_telemetry.py in favor of 'python -m bifrost.telemetry'
- Fixed a problem with like_bmon.py crashing when there are a large number of pipelines
- Added a CHANGELOG file
- Added a lightweight telemetry client for the Python API
v0.9.0
Changelog:
- Updated for Python3
- Updated the tools to use argparse over getopt
- Improved test suite coverage
- Fixed proclog removal on multi-user systems
- Added the
bifrost.romein
gridder module (@KentJames)
v0.8.0-benchmark
This release is an update to v0.7.0-benchmark, using Ben's new ctypesgen interface (faster Bifrost speeds).
The code and benchmark logs in this release (logfile: test/benchmarks/performance_vs_serial/aws_bench2/aws_bench2_$X.log.txt
where X in range(1, 64))
were used in the paper that was submitted to arxiv and JAI yesterday (which should be available as a public preprint by 8:00pm EDT; I will add it to this note).
The performance plots in the paper were produced using: https://github.com/MilesCranmer/bifrost_paper/blob/master/new_performance_figure.ipynb, specifically In[69]
and In[70]
:
Every measurement pixel is an average of 63 trials, and the errors show standard deviation.
The changes will be delineated in v0.8.0 (the Bifrost version also referenced in the paper).
v0.8.0
This is the version referenced in the paper submitted to JAI and arxiv (paper v1.0.2).
Changelog:
- @benbarsdell changed the entire ctypes interface from PyCLibrary into ctypesgen in one day. Bravo.
- Improves benchmark speeds significantly, as used for performance plots (v0.8.0-benchmark)
- Tutorials on bfMap and views added to documentation (@telegraphic)
- pretty print used for print_header block (@telegraphic)
- Several bug fixes by @telegraphic
- Several benchmark tests created
v0.7.0-benchmark
This release holds all the code that was used to create the benchmarks in the paper.
Note that the Makefile in /python
must be manually changed to install Bifrost with python setup.py install -O2
, and the benchmark script (in /test/benchmarks/performance_vs_serial
) should be ran with python -OO linear_fft_pipeline.py
. Or you can use the run_benchmarks.sh
script to automate it and generate benchmark logs for analysis.
Most of the code contained here is already on master, and will be included in upcoming releases. The full code changes will be described in the non-benchmark release.
v0.7.0
This the first release since the first git commit, establishing the start of a versioned history.
Future releases will have change logs.