Deprecate the DeviceNDArray class and public APIs that return instances

.github/workflows/simulator-test.yaml

@@ -102,7 +102,7 @@ jobs:
           set -euo pipefail
           CUDA_VER_PART="$(cut -d. -f1-2 <<< "${{ matrix.CUDA_VER }}" | tr . -)"
           PY_VER_PART="py$(tr -d . <<< "${{ matrix.PY_VER }}")"
-          PIXI_ENV="cu-${CUDA_VER_PART}-${PY_VER_PART}"
+          PIXI_ENV="sim-cu-${CUDA_VER_PART}-${PY_VER_PART}"
           echo "PIXI_ENV=${PIXI_ENV}" >> "${GITHUB_ENV}"
       - name: Setup pixi
         uses: prefix-dev/setup-pixi@fef5c9568ca6c4ff7707bf840ab0692ba3f08293  # v0.9.0

ci/coverage_report.sh

@@ -5,6 +5,7 @@
 set -euo pipefail
 
 CUDA_VER_MAJOR_MINOR=${CUDA_VER%.*}
+CUDA_VER_MAJOR=${CUDA_VER%.*.*}
 
 rapids-logger "Install wheel with test dependencies and coverage tools"
 package=$(realpath "${NUMBA_CUDA_ARTIFACTS_DIR}"/*.whl)
@@ -13,6 +14,7 @@ python -m pip install \
     "${package}" \
     "cuda-python==${CUDA_VER_MAJOR_MINOR%.*}.*" \
     "cuda-core" \
+    "cupy-cuda${CUDA_VER_MAJOR}x" \
     pytest-cov \
     coverage \
     --group test

ci/tools/run-tests

@@ -9,8 +9,20 @@
 set -euo pipefail
 
 echo "Installing numba-cuda wheel with test dependencies"
+
+# Check Python version to determine if we should install cupy
+PYTHON_VERSION=$(python -c "import sys; print(f'{sys.version_info.major}.{sys.version_info.minor}')")
+INSTALL_CUPY=true
+if [[ "${PYTHON_VERSION}" == "3.14" ]]; then
+  echo "Python 3.14 detected - skipping cupy installation (not yet available)"
+  INSTALL_CUPY=false
+fi
+
 if [[ "${LOCAL_CTK}" == 1 ]]; then
   pip install "${NUMBA_CUDA_ARTIFACTS_DIR}"/*.whl "cuda-bindings==${TEST_CUDA_MAJOR}.*" --group test
+  if [[ "${INSTALL_CUPY}" == "true" ]]; then
+    pip install "cupy-cuda${TEST_CUDA_MAJOR}x"
+  fi
 else
   pip install $(ls "${NUMBA_CUDA_ARTIFACTS_DIR}"/*.whl)["cu${TEST_CUDA_MAJOR}"] "cuda-toolkit==${TEST_CUDA_MAJOR}.${TEST_CUDA_MINOR}.*" --group "test-cu${TEST_CUDA_MAJOR}"
 fi

docs/source/user/ipc.rst

@@ -18,6 +18,10 @@ Sharing between processes is implemented using the Legacy CUDA IPC API
 Export device array to another process
 --------------------------------------
 
+.. note::
+  DeviceNDArray is deprecated. Prefer CuPy for array operations including
+  inter-process communication.
+
 A device array can be shared with another process in the same machine using
 the CUDA IPC API.  To do so, use the ``.get_ipc_handle()`` method on the device
 array to get a ``IpcArrayHandle`` object, which can be transferred to another
@@ -34,6 +38,10 @@ process.
 Import IPC memory from another process
 --------------------------------------
 
+.. note::
+  DeviceNDArray is deprecated. Prefer CuPy for array operations including
+  inter-process communication.
+
 The following function is used to open IPC handle from another process
 as a device array.
 

docs/source/user/memory.rst

@@ -11,6 +11,10 @@ Memory management
 Data transfer
 =============
 
+.. note::
+  DeviceNDArray is deprecated. Prefer CuPy for array operations including
+  memory transfers.
+
 Even though Numba can automatically transfer NumPy arrays to the device,
 it can only do so conservatively by always transferring device memory back to
 the host when a kernel finishes. To avoid the unnecessary transfer for
@@ -38,6 +42,9 @@ buffer using the following APIs:
 Device arrays
 -------------
 
+.. note::
+  DeviceNDArray is deprecated. Prefer CuPy for array operations.
+
 Device array references have the following methods.  These methods are to be
 called in host code, not within CUDA-jitted functions.
 
@@ -52,6 +59,11 @@ called in host code, not within CUDA-jitted functions.
 Pinned memory
 =============
 
+.. note::
+  DeviceNDArray is deprecated. Prefer CuPy for array operations including
+  allocating pinned memory
+
+
 .. autofunction:: numba.cuda.pinned
    :noindex:
 .. autofunction:: numba.cuda.pinned_array
@@ -63,6 +75,10 @@ Pinned memory
 Mapped memory
 =============
 
+.. note::
+  DeviceNDArray is deprecated. Prefer CuPy for array operations including
+  allocating mapped memory
+
 .. autofunction:: numba.cuda.mapped
    :noindex:
 .. autofunction:: numba.cuda.mapped_array
@@ -76,6 +92,10 @@ Mapped memory
 Managed memory
 ==============
 
+.. note::
+  DeviceNDArray is deprecated. Prefer CuPy for array operations including
+  allocating managed memory
+
 .. autofunction:: numba.cuda.managed_array
    :noindex: