npu: add demo workload

Signed-off-by: Tuomas Katila <[email protected]>

Author: tkatila

.github/workflows/lib-build.yaml

@@ -36,6 +36,7 @@ jobs:
           - sgx-sdk-demo
           - sgx-aesmd-demo
           - dsa-dpdk-dmadevtest
+          - intel-npu-demo
         builder: [buildah, docker]
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4

README.md

@@ -242,13 +242,13 @@ The summary of resources available via plugins in this repository is given in th
    * [intelgpu-job.yaml](demo/intelgpu-job.yaml)
  * `iaa.intel.com`  : `wq-user-[shared or dedicated]`
    * [iaa-accel-config-demo-pod.yaml](demo/iaa-accel-config-demo-pod.yaml)
+ * `npu.intel.com`  : `accel`
+   * [intel-npu-workload.yaml](demo/intel-npu-workload.yaml)
  * `qat.intel.com`  : `generic` or `cy`/`dc`/`asym-dc`/`sym-dc`
    * [compress-perf-dpdk-pod-requesting-qat-dc.yaml](deployments/qat_dpdk_app/compress-perf/compress-perf-dpdk-pod-requesting-qat-dc.yaml)
    * [crypto-perf-dpdk-pod-requesting-qat-cy.yaml](deployments/qat_dpdk_app/crypto-perf/crypto-perf-dpdk-pod-requesting-qat-cy.yaml)
  * `sgx.intel.com`  : `epc`
    * [intelsgx-job.yaml](deployments/sgx_enclave_apps/base/intelsgx-job.yaml)
- * `npu.intel.com`  : `accel`
-   * TODO
 
 ## Developers
 

cmd/npu_plugin/README.md

@@ -103,5 +103,43 @@ master
 
 ## Testing and Demos
 
-TODO
+The NPU plugin functionality can be verified by deploying a [npu-plugin-demo](../../demo/intel-npu-demo/) image which runs tests with the Intel NPU.
 
+1. Make the image available to the cluster:
+
+    Build image:
+
+    ```bash
+    $ make intel-npu-demo
+    ```
+
+    Tag and push the `intel-npu-demo` image to a repository available in the cluster. Then modify the [intel-npu-workload.yaml's](../../demo/intel-npu-workload.yaml) image location accordingly:
+
+    ```bash
+    $ docker tag intel/intel-npu-demo:devel <repository>/intel/intel-npu-demo:latest
+    $ docker push <repository>/intel/intel-npu-demo:latest
+    $ $EDITOR ${INTEL_DEVICE_PLUGINS_SRC}/demo/intel-npu-workload.yaml
+    ```
+
+    If you are running the demo on a single node cluster, and do not have your own registry, you can add image to node image cache instead. For example, to import docker image to containerd cache:
+
+    ```bash
+    $ docker save intel/intel-npu-demo:devel | ctr -n k8s.io images import -
+    ```
+    Running `ctr` may require the use of `sudo`.
+
+1. Create a job:
+
+    ```bash
+    $ kubectl apply -f ${INTEL_DEVICE_PLUGINS_SRC}/demo/intel-npu-workload.yaml
+    job.batch/npu-workload created
+    ```
+
+1. Review the job's logs:
+
+    ```bash
+    $ kubectl get pods | fgrep npu-workload
+    # substitute the 'xxxxx' below for the pod name listed above
+    $ kubectl logs npu-workload-xxxxx
+    <log output>
+    ```

demo/intel-npu-demo/Dockerfile

@@ -0,0 +1,24 @@
+FROM ubuntu:24.04 AS build
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+RUN apt update && apt install -y build-essential git git-lfs cmake python3 && rm -rf /var/lib/apt/lists/\*
+
+ENV NPU_RELEASE=v1.19.0
+ENV NPU_PROJECT=https://github.com/intel/linux-npu-driver/
+
+RUN mkdir /install
+
+RUN git clone -b ${NPU_RELEASE} --recurse-submodules --depth 1 ${NPU_PROJECT} && \
+  cd linux-npu-driver && \
+  cmake -DCMAKE_INSTALL_PREFIX=/install/usr -DCMAKE_INSTALL_LIBDIR=lib -B build -S . && \
+  cmake --build build --parallel $(nproc) && \
+  cmake --install build
+
+FROM ubuntu:24.04 AS final
+
+COPY --from=build /install /
+
+COPY npu-tests.sh /
+
+ENTRYPOINT [ "/npu-tests.sh" ]

demo/intel-npu-demo/npu-tests.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+echo "Run NPU tests"
+
+npu-kmd-test *:-Device.GroupOwnership || exit 1
+
+# use npu-umd-test -l to list all topics
+# note: some of the tests require the NPU compiler & openvino runtime
+# https://github.com/intel/linux-npu-driver/blob/main/docs/overview.md#building-a-driver-together-with-the-compiler
+topics="Device Driver Event EventSync MemoryAllocation MemoryExecution MemoryAllocationThreaded CommandCopyPerf MultiMemoryExecution MultiContext CommandMemoryFill"
+
+for topic in $topics; do
+    npu-umd-test -v ${topic}.* || exit 1
+done
+
+echo "Tests done"
+
+exit 0

demo/intel-npu-workload.yaml

@@ -0,0 +1,18 @@
+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: npu-workload
+spec:
+  template:
+    metadata:
+      labels:
+        app: example-job
+    spec:
+      containers:
+      - image: intel/intel-npu-demo:devel
+        name: workload
+        resources:
+          limits:
+            npu.intel.com/accel: 1
+      restartPolicy: Never
+  backoffLimit: 4