Support numba 0.66-0.67 with LLVM 22 (#78)

* Port to LLVM 22 and numba 0.66-0.67

numba 0.66 and 0.67 use llvmlite 0.48 (LLVM 22), so build the pass library
and the OpenMP runtimes against LLVM 22.1.8.

- Port the pass library to the LLVM 22 OpenMPIRBuilder API
  (__kmpc_parallel_60, TargetKernelArgs, ReductionInfo, PassPlugin.h), and
  set the OpenMPIRBuilder config, which createReductions now reads.
- Build Linux wheels with the conda-forge LLVM 22 toolchain, compiling
  against the manylinux gcc-toolset libstdc++.
- Build the nvptx and amdgcn device runtime bitcode from openmp/device,
  ignoring host CFLAGS and CXXFLAGS.
- Keep three runtime patches for 22.1.8 (static LLVM, skip liboffload,
  __tgt_get_device_info) and drop the patch sets for older LLVM versions.
- Require numba >=0.66,<0.68 and test numba 0.66.0 and 0.67.0.
- Drop the conda is_freethreading variant, which no numba pin uses now.

The libomptarget CUDA plugin in LLVM 22 includes llvm/llvm-project#159354
(fixes #71), and llvmlite 0.48 knows sm_110 (fixes #68).

The OpenMPIRBuilder config setting, the device runtime host-flag reset, and
the __tgt_get_device_info patch port are from #75.

* Read AVX512_SKX from the NumPy SIMD features in conda tests

numba 0.66 removed the "NumPy AVX512_SKX detected" sysinfo entry, so the
conda test script failed with a KeyError before running the tests. Check the
"NumPy Supported SIMD features" list instead. The fix is from #75.

* Document PyOMP 0.6 compatibility

---------

Co-authored-by: Aleksander Wennersteen <aleksander.wennersteen@pasqal.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Author: 3 people

.github/workflows/build-upload-conda.yml

@@ -64,7 +64,7 @@ jobs:
 
           # Select the python variant; 3.14t needs the free-threaded ABI.
           if [[ "${{ matrix.python-version }}" == "3.14t" ]]; then
-            PY_ARGS=(--python "3.14.* *_cp314t" -m buildscripts/conda-recipes/pyomp/freethreading.yaml)
+            PY_ARGS=(--python "3.14.* *_cp314t")
           else
             PY_ARGS=(--python ${{ matrix.python-version }})
           fi

.github/workflows/build-upload-wheels.yml

@@ -52,7 +52,7 @@ jobs:
         # Set LLVM_VERSION for the host to forward to the cibuildwheel
         # environment.
         env:
-          LLVM_VERSION: "20.1.8"
+          LLVM_VERSION: "22.1.8"
         run: python -m cibuildwheel --output-dir wheelhouse
 
       - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -69,7 +69,7 @@ jobs:
 
       - name: Build sdist
         env:
-          LLVM_VERSION: "20.1.8"
+          LLVM_VERSION: "22.1.8"
         run: pipx run build --sdist
 
       - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
@@ -85,23 +85,7 @@ jobs:
       matrix:
         os: [ubuntu-latest, macos-latest, ubuntu-24.04-arm]
         python-version: ['3.10', '3.11', '3.12', '3.13', '3.14', '3.14t']
-        numba-version: ['0.62.0', '0.62.1', '0.63.0', '0.63.1', '0.64.0', '0.65.0', '0.65.1']
-        exclude:
-          - python-version: '3.14'
-            numba-version: '0.62.0'
-          - python-version: '3.14'
-            numba-version: '0.62.1'
-          # Numba supports free-threaded Python starting with 0.65.
-          - python-version: '3.14t'
-            numba-version: '0.62.0'
-          - python-version: '3.14t'
-            numba-version: '0.62.1'
-          - python-version: '3.14t'
-            numba-version: '0.63.0'
-          - python-version: '3.14t'
-            numba-version: '0.63.1'
-          - python-version: '3.14t'
-            numba-version: '0.64.0'
+        numba-version: ['0.66.0', '0.67.0']
 
     steps:
       - name: Download built wheels

README.md

@@ -45,11 +45,12 @@ conda install -c python-for-hpc -c conda-forge pyomp
 
 | PyOMP | Numba           |
 | ----- | --------------- |
+| 0.6.x | 0.66.x - 0.67.x |
 | 0.5.x | 0.62.x - 0.65.x |
 | 0.4.x | 0.61.x          |
 | 0.3.x | 0.57.x - 0.60.x |
 
-Python 3.14t (free-threaded) requires Numba 0.65.x.
+Python 3.14t (free-threaded) requires PyOMP 0.5.2 or later and Numba 0.65.x or later.
 
 Besides a standard installation, we also provide the following options to
 quickly try out PyOMP online or through a container.

buildscripts/cibuildwheel/setup-miniforge3.sh

@@ -58,6 +58,7 @@ bash _downloads/miniforge3.sh -b -f -p "_stage/miniforge3"
 echo "Miniforge installed"
 source "_stage/miniforge3/bin/activate" base
 
-# Create conda environment with tools and libraries for the LLVM_VERSION.
+# Create conda environment with tools and libraries for the LLVM_VERSION. lld
+# links the AMDGPU device runtime.
 echo "Installing llvmdev ${LLVM_VERSION}..."
-conda create -n llvmdev-${LLVM_VERSION} --override-channels -c conda-forge -q -y clang=${LLVM_VERSION} clangxx=${LLVM_VERSION} clang-tools=${LLVM_VERSION} llvmdev=${LLVM_VERSION} zstd
+conda create -n llvmdev-${LLVM_VERSION} --override-channels -c conda-forge -q -y clang=${LLVM_VERSION} clangxx=${LLVM_VERSION} clang-tools=${LLVM_VERSION} llvmdev=${LLVM_VERSION} lld=${LLVM_VERSION} zstd

buildscripts/conda-recipes/pyomp/conda_build_config.yaml

@@ -1,8 +1,3 @@
-# --- Free-threading ---
-# Default for the is_freethreading selector in meta.yaml; override for 3.14t.
-is_freethreading:
-  - false
-
 # --- Compilers ---
 c_compiler_version:
   - 9                       # [linux]

buildscripts/conda-recipes/pyomp/freethreading.yaml

@@ -1,5 +1,5 @@
 {% set version = environ.get('GIT_DESCRIBE_TAG', '0.0.0').lstrip('v') %}
-{% set LLVM_VERSION = environ.get('LLVM_VERSION', '20.1.8') %}
+{% set LLVM_VERSION = environ.get('LLVM_VERSION', '22.1.8') %}
 {% set LLVM_VERSION_MAJOR = LLVM_VERSION.split('.')[0] %}
 {% set LLVM_VERSION_MINOR = LLVM_VERSION.split('.')[1] %}
 
@@ -45,26 +45,24 @@ requirements:
     - sysroot_linux-aarch64  # [aarch64]
     - setuptools
     - setuptools_scm
-    - numba >=0.62, <0.66
-    - numba >=0.65, <0.66  # [is_freethreading]
+    - numba >=0.66, <0.68
     - clang {{ LLVM_VERSION }}
     - clangxx {{ LLVM_VERSION }}
     - clang-tools {{ LLVM_VERSION }}
     - llvmdev {{ LLVM_VERSION }}
+    # lld links the AMDGPU device runtime.
+    - lld {{ LLVM_VERSION }} # [linux]
     - zlib
     # require llvm-openmp for the openmp cpu runtime.
-    - llvm-openmp 21.1.8.* # [osx]
-    - llvm-openmp {{ LLVM_VERSION }} # [not osx]
+    - llvm-openmp {{ LLVM_VERSION }}
     - elfutils # [linux]
     - libffi # [linux]
   run:
     - python
     - setuptools
-    - numba >=0.62, <0.66
-    - numba >=0.65, <0.66  # [is_freethreading]
+    - numba >=0.66, <0.68
     # require llvm-openmp for the openmp cpu runtime.
-    - llvm-openmp 21.1.8.* # [osx]
-    - llvm-openmp {{ LLVM_VERSION }} # [not osx]
+    - llvm-openmp {{ LLVM_VERSION }}
     - lark
     - cffi
 

buildscripts/conda-recipes/pyomp/meta.yaml

@@ -12,7 +12,7 @@ export PYTHONFAULTHANDLER=1
 # of low accuracy SVML libm replacements in ufunc loops.
 _NPY_CMD='from numba.misc import numba_sysinfo;\
           sysinfo=numba_sysinfo.get_sysinfo();\
-          print(sysinfo["NumPy AVX512_SKX detected"] and
+          print("AVX512_SKX" in sysinfo["NumPy Supported SIMD features"] and
                 sysinfo["NumPy Version"]>="1.22")'
 NUMPY_DETECTS_AVX512_SKX_NP_GT_122=$(python -c "$_NPY_CMD")
 echo "NumPy >= 1.22 with AVX512_SKX detected: $NUMPY_DETECTS_AVX512_SKX_NP_GT_122"

buildscripts/conda-recipes/pyomp/run_test.sh

@@ -8,7 +8,7 @@
 
 
 PYTHON_VERSIONS = ("3.10", "3.11", "3.12", "3.13", "3.14")
-NUMBA_VERSION = "0.65.1"
+NUMBA_VERSION = "0.67.0"
 MINIFORGE_VERSION = "26.3.2-3"
 MINIFORGE_SHA256 = "848194851a98903134187fbb4ab50efe87b003e0c0f808f97644b7524a62bf2c"
 WHEEL_DIRECTORY = (

buildscripts/modal/test_gpu.py

@@ -24,14 +24,16 @@ supported.
 +--------+---------------------+
 | PyOMP  | Numba               |
 +========+=====================+
+| 0.6.x  | 0.66.x - 0.67.x     |
++--------+---------------------+
 | 0.5.x  | 0.62.x - 0.65.x     |
 +--------+---------------------+
 | 0.4.x  | 0.61.x              |
 +--------+---------------------+
 | 0.3.x  | 0.57.x - 0.60.x     |
 +--------+---------------------+
 
-Python 3.14t (free-threaded) requires Numba 0.65.x.
+Python 3.14t (free-threaded) requires PyOMP 0.5.2 or later and Numba 0.65.x or later.
 
 Additional options
 ------------------