You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* Move openmp patches to versioned directories
- Add patches for 15.0.7, 16.0.6
* Ignore .cache clangd directories
* Make device 0 default for the examples hello-target.py
* Use LLVM_version env in setup.py to install openmp
* Use LLVM_VERSION env var in setup-miniconda3.sh
* Update init to import decorators and remove LLVM version check
* Replace cffi with ctypes to call LLVM openmp runtime symbols
- Avoid ffi dlopen namespace issues with initialization
* Replace deprecated compile_isolated interface with njit
* Add pointee type information in tags
- Recent versions of LLVM use opaque pointers. The numba frontend must
communicate the pointee types to the LLVM plugin pass for lowering
* Require LLVM_DIR env var and ensure LLVM is found only in it
- Avoids using other installation directories of LLVM in the path
* Make OpenMPCUDACodegen a singleton
- Ensure CUDA context is initalized once
* Update LLVM API and use pointee types from numba in the plugin pass
* Add patch for linking LLVM statically in openmp 16.0.6
* Update tests
- Fix single tests to nest in parallel (as it must)
- Skip target data tests that are problematic if no target region
executes
- Clean up
* Add compatibility matrix in README
* Fix typo setup.py
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* Fix comment in CGIntrinsicsOpenMP.cpp
Co-authored-by: Todd A. Anderson <drtodd13@comcast.net>
* Rename hasher in setup.py to avoid conflict with parameter
* Remove unused output
* Move pointee type information at the right branch of arg_to_str
- Append when x is a str
- Cleanup dead code in arg_to_str
* Add compatibility matrix to README
* Use LLVM_VERSION env var in project.toml and setup-miniconda3.sh
* Build wheels for numba 0.61 and harden github action
- Use LLVM 15 for building to be compatible with numba 0.61
- Pin all actions to the commit ref
- Add least permissions at the workflow level
* Add print message in pass CMakeLists
* Fix includes and order
* Use one conda env and export absolute paths
* Use clang-tools and clang instead of clangdev in setup-miniconda3.sh
* Update conda build files for numba 0.61 and use LLVM_VERSION
* Make more robust the check for NVIDIA GPUs in conda build test
- Use command to check if nvidia-smi exists
* Work around assert in tests with python 3.10 and numba lowering
* Update file tests for LLVM 15 in conda meta.yaml
* Update gitlab CI
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Todd A. Anderson <drtodd13@comcast.net>
0 commit comments