wip: tunercheck action

Signed-off-by: Raghu Raja <[email protected]>

Author: rajachan

.github/workflows/tunercheck.yaml

@@ -0,0 +1,56 @@
+name: Build Checks
+on: [push, pull_request]
+permissions:
+  contents: read
+  pull-requests: read
+
+jobs:
+  tuner-decisions-check:
+    runs-on: ubuntu-22.04
+    steps:
+      - name: Install Dependencies
+        run: |
+          sudo apt-key del 7fa2af80
+          wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
+          sudo dpkg -i cuda-keyring_1.1-1_all.deb
+          sudo apt-get update
+          sudo apt-get install cuda-toolkit libhwloc-dev
+          pip install uv
+
+
+      - name: Fetch and Install EFA Installer
+        run: |
+          curl -O https://efa-installer.amazonaws.com/aws-efa-installer-latest.tar.gz
+          tar -xf aws-efa-installer-*.tar.gz
+          pushd aws-efa-installer/
+              sudo ./efa_installer.sh -y --skip-kmod
+          popd
+
+      - uses: actions/checkout@v4
+      - name: Build Plugin
+        run: |
+          set -x
+
+          # actions/checkout@v4 would drop the plugin source in $PWD,
+          # so go ahead and build it.
+          ./autogen.sh
+          ./configure --with-mpi=/opt/amazon/openmpi \
+                      --with-libfabric=/opt/amazon/efa \
+                      --with-cuda=/usr/local/cuda/ \
+                      --enable-platform-aws \
+                      --prefix=$PWD/install
+          make -j 2
+          make install
+
+      - name: Tuner Check
+        run: |
+          pushd contrib/python/
+
+          OFI_NCCL_FORCE_PRODUCT_NAME=p5en.48xlarge uv run show-tuner-decisions ../..//install/lib/libnccl-ofi-tuner.so \
+             --min-ranks-per-node 1 --max-ranks-per-node 8 \
+             --min-nnodes 2 --max-nnodes 2048
+
+          OFI_NCCL_FORCE_PRODUCT_NAME=p5.48xlarge uv run show-tuner-decisions ../..//install/lib/libnccl-ofi-tuner.so \
+             --min-ranks-per-node 1 --max-ranks-per-node 8 \
+             --min-nnodes 2 --max-nnodes 2048
+          popd