Merge branch 'master' into feature/topk

Author: ndgrigorian

.flake8

@@ -31,6 +31,6 @@ per-file-ignores =
     dpctl/utils/_compute_follows_data.pyx: E999, E225, E227
     dpctl/utils/_onetrace_context.py: E501, W505
     dpctl/tensor/_array_api.py: E501, W505
-    examples/cython/sycl_buffer/syclbuffer/_buffer_example.pyx: E999, E225, E402
-    examples/cython/usm_memory/blackscholes/blackscholes.pyx: E999, E225, E226, E402
+    examples/cython/sycl_buffer/syclbuffer/_syclbuffer.pyx: E999, E225, E402
+    examples/cython/usm_memory/blackscholes/_blackscholes_usm.pyx: E999, E225, E226, E402
     examples/cython/use_dpctl_sycl/use_dpctl_sycl/_cython_api.pyx: E999, E225, E226, E402

.github/workflows/conda-package.yml

@@ -24,7 +24,7 @@ jobs:
 
     strategy:
       matrix:
-        python: ['3.9', '3.10', '3.11', '3.12']
+        python: ['3.9', '3.10', '3.11', '3.12', '3.13']
     steps:
       - uses: actions/[email protected]
         with:
@@ -64,12 +64,12 @@ jobs:
             $CHANNELS \
             conda-recipe
       - name: Upload artifact
-        uses: actions/upload-artifact@v4.4.3
+        uses: actions/upload-artifact@v4.5.0
         with:
           name: ${{ env.PACKAGE_NAME }} ${{ runner.os }} Python ${{ matrix.python }}
           path: /usr/share/miniconda/conda-bld/linux-64/${{ env.PACKAGE_NAME }}-*.tar.bz2
       - name: Upload wheels artifact
-        uses: actions/upload-artifact@v4.4.3
+        uses: actions/upload-artifact@v4.5.0
         with:
           name: ${{ env.PACKAGE_NAME }} ${{ runner.os }} Wheels Python ${{ matrix.python }}
           path: ${{ env.WHEELS_OUTPUT_FOLDER }}${{ env.PACKAGE_NAME }}-*.whl
@@ -80,7 +80,7 @@ jobs:
 
     strategy:
       matrix:
-        python: ['3.9', '3.10', '3.11', '3.12']
+        python: ['3.9', '3.10', '3.11', '3.12', '3.13']
     steps:
       - uses: actions/[email protected]
         with:
@@ -127,13 +127,13 @@ jobs:
           conda build --no-test --python ${{ matrix.python }} --numpy 2 -c conda-forge --override-channels conda-recipe
 
       - name: Upload artifact
-        uses: actions/upload-artifact@v4.4.3
+        uses: actions/upload-artifact@v4.5.0
         with:
           name: ${{ env.PACKAGE_NAME }} ${{ runner.os }} Python ${{ matrix.python }}
           path: ${{ env.CONDA_BLD }}${{ env.PACKAGE_NAME }}-*.tar.bz2
 
       - name: Upload wheels artifact
-        uses: actions/upload-artifact@v4.4.3
+        uses: actions/upload-artifact@v4.5.0
         with:
           name: ${{ env.PACKAGE_NAME }} ${{ runner.os }} Wheels Python ${{ matrix.python }}
           path: ${{ env.WHEELS_OUTPUT_FOLDER }}${{ env.PACKAGE_NAME }}-*.whl
@@ -145,7 +145,7 @@ jobs:
 
     strategy:
       matrix:
-        python: ['3.9', '3.10', '3.11', '3.12']
+        python: ['3.9', '3.10', '3.11', '3.12', '3.13']
         experimental: [false]
         runner: [ubuntu-22.04]
     continue-on-error: ${{ matrix.experimental }}
@@ -239,7 +239,7 @@ jobs:
         shell: cmd /C CALL {0}
     strategy:
       matrix:
-        python: ['3.9', '3.10', '3.11', '3.12']
+        python: ['3.9', '3.10', '3.11', '3.12', '3.13']
         experimental: [false]
         runner: [windows-2019]
     continue-on-error: ${{ matrix.experimental }}
@@ -408,7 +408,7 @@ jobs:
     timeout-minutes: 20
     strategy:
       matrix:
-        python: ['3.9', '3.10', '3.11', '3.12']
+        python: ['3.9', '3.10', '3.11', '3.12', '3.13']
     steps:
       - name: Download conda artifact
         uses: actions/download-artifact@v4
@@ -445,7 +445,7 @@ jobs:
     timeout-minutes: 20
     strategy:
       matrix:
-        python: ['3.9', '3.10', '3.11', '3.12']
+        python: ['3.9', '3.10', '3.11', '3.12', '3.13']
     steps:
       - name: Download artifact
         uses: actions/download-artifact@v4
@@ -549,7 +549,7 @@ jobs:
       - name: Install example requirements
         shell: bash -ex -l {0}
         env:
-          DPCPP_CMPLR: "dpcpp_linux-64>=2024.2"
+          DPCPP_CMPLR: "dpcpp_linux-64>=2025.0"
         run: |
           CHANNELS="${{ env.CHANNELS }}"
           . $CONDA/etc/profile.d/conda.sh
@@ -569,6 +569,7 @@ jobs:
               ${{ env.DPCPP_CMPLR }} "${DPCTL_DEPENDS}" \
               "sysroot_linux-64>=2.28"
           echo "Compiler installed"
+          conda list -n ${{ env.BUILD_ENV_NAME }}
       - name: Install dpctl
         shell: bash -l {0}
         run: |
@@ -586,14 +587,8 @@ jobs:
           for d in $(find . -maxdepth 1 -type d -not -path ".")
           do
             pushd $d
-            export MKLROOT=${CONDA_PREFIX}
-            export TBBROOT=${CONDA_PREFIX}
             conda activate --stack build_env
-            CC=icx CXX=icpx python setup.py build_ext --inplace -G Ninja -- \
-                -DTBB_LIBRARY_DIR=${TBBROOT}/lib \
-                -DMKL_LIBRARY_DIR=${MKLROOT}/lib \
-                -DMKL_INCLUDE_DIR=${MKLROOT}/include \
-                -DTBB_INCLUDE_DIR=${TBBROOT}/include || exit 1
+            CC=icx CXX=icpx python setup.py build_ext --inplace -G Ninja || exit 1
             conda deactivate
             if [ -e tests ]
             then
@@ -614,7 +609,7 @@ jobs:
           do
             pushd $d
             conda activate --stack ${{ env.BUILD_ENV_NAME }}
-            python setup.py build_ext --inplace || exit 1
+            CC=icx CXX=icpx python setup.py develop -G Ninja  || exit 1
             conda deactivate
             python -m pytest tests || exit 1
             popd

.github/workflows/generate-docs.yml

@@ -108,7 +108,7 @@ jobs:
           git push tokened_docs gh-pages
       - name: Save built docs as an artifact
         if: ${{ github.event.pull_request && github.event.pull_request.head.repo.fork && github.event.action != 'closed'}}
-        uses: actions/upload-artifact@v4.4.3
+        uses: actions/upload-artifact@v4.5.0
         with:
           name: ${{ env.PACKAGE_NAME }} rendered documentation
           path: ~/docs

.github/workflows/openssf-scorecard.yml

@@ -61,14 +61,14 @@ jobs:
       # Upload the results as artifacts (optional). Commenting out will disable uploads of run results in SARIF
       # format to the repository Actions tab.
       - name: "Upload artifact"
-        uses: actions/upload-artifact@184d73b71b93c222403b2e7f1ffebe4508014249 # v4.4.0
+        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b # v4.4.0
         with:
           name: SARIF file
           path: results.sarif
           retention-days: 14
 
       # Upload the results to GitHub's code scanning dashboard.
       - name: "Upload to code-scanning"
-        uses: github/codeql-action/upload-sarif@aa578102511db1f4524ed59b8cc2bae4f6e88195 # v3.27.6
+        uses: github/codeql-action/upload-sarif@48ab28a6f5dbc2a99bf1e0131198dd8f1df78169 # v3.28.0
         with:
           sarif_file: results.sarif

.github/workflows/run-tests-from-dppy-bits.yaml

@@ -24,7 +24,7 @@ jobs:
 
     strategy:
       matrix:
-        python: ['3.9', '3.10', '3.11', '3.12']
+        python: ['3.9', '3.10', '3.11', '3.12', '3.13']
         experimental: [false]
         runner: [ubuntu-22.04, ubuntu-24.04]
     continue-on-error: ${{ matrix.experimental }}
@@ -75,7 +75,7 @@ jobs:
         shell: cmd /C CALL {0}
     strategy:
       matrix:
-        python: ['3.9', '3.10', '3.11', '3.12']
+        python: ['3.9', '3.10', '3.11', '3.12', '3.13']
         experimental: [false]
         runner: [windows-2019]
 

CHANGELOG.md

@@ -39,6 +39,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 * Add missing include of SYCL header to "math_utils.hpp" [gh-1899](https://github.com/IntelPython/dpctl/pull/1899)
 * Add support of CV-qualifiers in `is_complex<T>` helper [gh-1900](https://github.com/IntelPython/dpctl/pull/1900)
 * Tuning work for elementwise functions with modest performance gains (under 10%) [gh-1889](https://github.com/IntelPython/dpctl/pull/1889)
+* Support for Python 3.13 for `dpctl` [gh-1941](https://github.com/IntelPython/dpctl/pull/1941)
 
 ## [0.18.3] - Dec. 07, 2024
 

dpctl/_host_task_util.hpp

@@ -58,8 +58,14 @@ DPCTLSyclEventRef async_dec_ref(DPCTLSyclQueueRef QRef,
                 cgh.depends_on(*(unwrap<sycl::event>(depERefs[ev_id])));
             }
             cgh.host_task([obj_array_size, obj_vec]() {
+                const bool initialized = Py_IsInitialized();
+#if PY_VERSION_HEX < 0x30d0000
+                const bool finalizing = _Py_IsFinalizing();
+#else
+                const bool finalizing = Py_IsFinalizing();
+#endif
                 // if the main thread has not finilized the interpreter yet
-                if (Py_IsInitialized() && !_Py_IsFinalizing()) {
+                if (initialized && !finalizing) {
                     PyGILState_STATE gstate;
                     gstate = PyGILState_Ensure();
                     for (size_t i = 0; i < obj_array_size; ++i) {

dpctl/apis/include/dpctl4pybind11.hpp

@@ -198,7 +198,13 @@ class dpctl_capi
     {
         void operator()(py::object *p) const
         {
-            bool guard = (Py_IsInitialized() && !_Py_IsFinalizing());
+            const bool initialized = Py_IsInitialized();
+#if PY_VERSION_HEX < 0x30d0000
+            const bool finilizing = _Py_IsFinalizing();
+#else
+            const bool finilizing = Py_IsFinalizing();
+#endif
+            const bool guard = initialized && !finilizing;
 
             if (guard) {
                 delete p;

dpctl/tensor/_indexing_functions.py

@@ -491,7 +491,11 @@ def take_along_axis(x, indices, /, *, axis=-1, mode="wrap"):
             "from input arguments. "
         )
     mode_i = _get_indexing_mode(mode)
-    indexes_dt = ti.default_device_index_type(exec_q.sycl_device)
+    indexes_dt = (
+        dpt.uint64
+        if indices.dtype == dpt.uint64
+        else ti.default_device_index_type(exec_q.sycl_device)
+    )
     _ind = tuple(
         (
             indices
@@ -567,7 +571,11 @@ def put_along_axis(x, indices, vals, /, *, axis=-1, mode="wrap"):
         )
     out_usm_type = dpctl.utils.get_coerced_usm_type(usm_types_)
     mode_i = _get_indexing_mode(mode)
-    indexes_dt = ti.default_device_index_type(exec_q.sycl_device)
+    indexes_dt = (
+        dpt.uint64
+        if indices.dtype == dpt.uint64
+        else ti.default_device_index_type(exec_q.sycl_device)
+    )
     _ind = tuple(
         (
             indices