Merge branch 'main' into update-readme-conda

.ycm_extra_conf.py

@@ -5,13 +5,13 @@
 
 CONDA_PREFIX = os.environ["CONDA_PREFIX"]
 NVJITLINK_DIR = Path(__file__).parent
-VER = sys.version_info
-PYTHON_INCLUDE_NAME = "python3.10"  # Not ideal
+PYTHON_VERSION = sys.version_info
+PYTHON_INCLUDE_NAME = "python%d.%d" % PYTHON_VERSION
 
 CONDA_INCLUDE_DIR = Path(CONDA_PREFIX, "include")
 PYTHON_INCLUDE_DIR = Path(CONDA_INCLUDE_DIR, PYTHON_INCLUDE_NAME)
 NVJITLINK_INCLUDE_DIR = Path(NVJITLINK_DIR, "include")
-CUDA_INCLUDE_DIR = "/usr/local/cuda-12.0/include"
+CUDA_INCLUDE_DIR = "/usr/local/cuda/include"
 
 flags = [
     "--cuda-gpu-arch=sm_50",

CMakeLists.txt

@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.26.4 FATAL_ERROR)
 
 project(
   pynvjitlink
-  VERSION 0.1.12
+  VERSION ${SKBUILD_PROJECT_VERSION}
   LANGUAGES CXX CUDA
 )
 

README.md

@@ -1,7 +1,7 @@
 # <div align="left"><img src="https://rapids.ai/assets/images/rapids_logo.png" width="90px"/>&nbsp;pynvjitlink</div>
 
 The [RAPIDS](https://rapids.ai) pynvjitlink library provides a Python binding for the
-[nvJitLink library](https://docs.nvidia.com/cuda/nvJitLink/index.html).
+[nvJitLink library](https://docs.nvidia.com/cuda/nvjitlink/index.html).
 
 ## Installation with pip
 

ci/install_latest_cuda_toolkit.sh

@@ -17,4 +17,4 @@ else
     exit 1
 fi
 
-yum install -y cuda-toolkit-12-3
+yum install -y cuda-toolkit-12-4

ci/release/update-version.sh

@@ -17,5 +17,3 @@ function sed_runner() {
 
 # Centralized version file update
 echo "${NEXT_FULL_TAG}" > pynvjitlink/VERSION
-sed_runner 's/'"^  VERSION [0-9\.]*"'/'"  VERSION ${NEXT_FULL_TAG}"'/g' CMakeLists.txt
-sed_runner 's/^version = "[0-9\.]*"/version = "'${NEXT_FULL_TAG}'"/g' pyproject.toml

ci/test_conda.sh

@@ -5,15 +5,17 @@ set -euo pipefail
 
 . /opt/conda/etc/profile.d/conda.sh
 
-rapids-logger "Generate testing dependencies"
+rapids-logger "Install testing dependencies"
 # TODO: Replace with rapids-dependency-file-generator
 rapids-mamba-retry create -n test \
     c-compiler \
     cxx-compiler \
     cuda-nvcc \
+    cuda-nvrtc \
     cuda-version=${RAPIDS_CUDA_VERSION%.*} \
     "numba>=0.58" \
     make \
+    psutil \
     pytest \
     python=${RAPIDS_PY_VERSION}
 
@@ -40,15 +42,20 @@ popd
 rapids-logger "Check GPU usage"
 nvidia-smi
 
+rapids-logger "Show Numba system info"
+python -m numba --sysinfo
+
 EXITCODE=0
 trap "EXITCODE=1" ERR
 set +e
 
-rapids-logger "pytest pynvjitlink"
-pytest \
+rapids-logger "Run Tests"
+pushd pynvjitlink/tests
+python -m pytest \
   --cache-clear \
   --junitxml="${RAPIDS_TESTS_DIR}/junit-pynvjitlink.xml" \
   -v
+popd
 
 rapids-logger "Test script exiting with value: $EXITCODE"
 exit ${EXITCODE}

ci/test_patch.sh

@@ -5,13 +5,14 @@ set -euo pipefail
 
 . /opt/conda/etc/profile.d/conda.sh
 
-rapids-logger "Generate testing dependencies"
+rapids-logger "Install testing dependencies"
 # TODO: Replace with rapids-dependency-file-generator
 rapids-mamba-retry create -n test \
-    cuda-nvcc-impl \
+    cuda-nvcc \
     cuda-nvrtc \
     cuda-version=${RAPIDS_CUDA_VERSION%.*} \
     "numba>=0.58" \
+    psutil \
     python=${RAPIDS_PY_VERSION}
 
 # Temporarily allow unbound variables for conda activation.
@@ -30,6 +31,9 @@ rapids-mamba-retry install \
 rapids-logger "Check GPU usage"
 nvidia-smi
 
+rapids-logger "Show Numba system info"
+python -m numba --sysinfo
+
 EXITCODE=0
 trap "EXITCODE=1" ERR
 set +e

ci/test_wheel.sh

@@ -3,22 +3,38 @@
 
 set -euo pipefail
 
+rapids-logger "Install testing dependencies"
+# TODO: Replace with rapids-dependency-file-generator
+python -m pip install \
+    "numba>=0.58" \
+    psutil \
+    pytest
+
 rapids-logger "Download Wheel"
 RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen ${RAPIDS_CUDA_VERSION})"
 RAPIDS_PY_WHEEL_NAME="pynvjitlink_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-s3 ./dist/
 
-# This is the version of the suffix with a preceding hyphen. It's used
-# everywhere except in the final wheel name.
-PACKAGE_CUDA_SUFFIX="-${RAPIDS_PY_CUDA_SUFFIX}"
+RAPIDS_TESTS_DIR=${RAPIDS_TESTS_DIR:-"${PWD}/test-results"}/
+mkdir -p "${RAPIDS_TESTS_DIR}"
 
 rapids-logger "Install wheel"
-pip install --find-links ./dist pynvjitlink${PACKAGE_CUDA_SUFFIX}
+python -m pip install $(echo ./dist/pynvjitlink_${RAPIDS_PY_CUDA_SUFFIX}*.whl)
 
 rapids-logger "Build Tests"
 pushd test_binary_generation
 make
 popd
 
+rapids-logger "Check GPU usage"
+nvidia-smi
+
+rapids-logger "Show Numba system info"
+python -m numba --sysinfo
+
 rapids-logger "Run Tests"
-pip install pytest
-pytest pynvjitlink/tests
+pushd pynvjitlink/tests
+python -m pytest \
+  --cache-clear \
+  --junitxml="${RAPIDS_TESTS_DIR}/junit-pynvjitlink.xml" \
+  -v
+popd

conda/recipes/pynvjitlink/conda_build_config.yaml

@@ -8,4 +8,4 @@ cuda_compiler:
   - cuda-nvcc
 
 cuda_compiler_version:
-  - 12.2
+  - 12.4

conda/recipes/pynvjitlink/meta.yaml

@@ -1,12 +1,16 @@
 # Copyright (c) 2018-2024, NVIDIA CORPORATION.
 
 {% set data = load_file_data("pyproject.toml") %}
+{% set version = load_file_regex(
+    load_file="pynvjitlink/VERSION",
+    regex_pattern="(?P<value>.*)"
+)[0] %}
 {% set project_data = data.get("project") %}
 {% set project_urls = project_data["urls"] %}
 
 package:
   name: pynvjitlink
-  version: {{ project_data["version"] }}
+  version: {{ version }}
 
 source:
   path: ../../..

pynvjitlink/VERSION

@@ -1 +1 @@
-0.1.12
+0.1.13

pynvjitlink/patch.py

@@ -104,7 +104,7 @@ def add_file(self, path, kind):
         elif kind == FILE_EXTENSION_MAP["fatbin"]:
             fn = self._linker.add_fatbin
         elif kind == FILE_EXTENSION_MAP["a"]:
-            raise LinkerError("Don't know how to link archives")
+            fn = self._linker.add_library
         elif kind == FILE_EXTENSION_MAP["ptx"]:
             return self.add_ptx(data, name)
         elif kind == FILE_EXTENSION_MAP["o"]:

pynvjitlink/tests/test_pynvjitlink_api.py

@@ -94,11 +94,6 @@ def test_add_cubin_with_fatbin_error(device_functions_fatbin):
         nvjitlinker.add_cubin(device_functions_fatbin, name)
 
 
-@pytest.mark.xfail(
-    reason="Fails in conda due to using CUDA 12.2u2. This should be fixed "
-    "when requiring CUDA 12.3 nvjitlink library. See: "
-    "https://github.com/rapidsai/pynvjitlink/pull/33#issuecomment-1885182905"
-)
 def test_add_fatbin_with_cubin_error(device_functions_cubin):
     nvjitlinker = NvJitLinker("-arch=sm_75")
     name = "test_device_functions.cubin"

pyproject.toml

@@ -11,7 +11,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "pynvjitlink"
-version = "0.1.12"
+dynamic = ["version"]
 description = "nvJitLink Python binding"
 readme = { file = "README.md", content-type = "text/markdown" }
 authors = [
@@ -29,5 +29,7 @@ License = "https://github.com/rapidsai/pynvjitlink/blob/main/LICENSE"
 [tool.setuptools]
 license-files = ["LICENSE"]
 
-[tool.setuptools.dynamic]
-version = {file = "pynvjitlink/VERSION"}
+[tool.scikit-build.metadata.version]
+provider = "scikit_build_core.metadata.regex"
+input = "pynvjitlink/VERSION"
+regex = "(?P<value>.*)"