[CI] GPU quick tests (#3433)

Added quick tests to GPU using our server as runner. However, instead of
building and running in the same job, uses a similar pipeline as CPU by
pretending that `build_gpu` and `test_quick_gpu` are run across
different runners, so that this can be adjusted easily once we have
proper runners.

Author: qwang98

.github/actions/init-testing-instance-gpu/action.yml

@@ -0,0 +1,37 @@
+name: "Init testing instance (GPU)"
+description: "Initialises a testing instance with all required tools and fetches the precomputed tests archive named `tests_archive_gpu`"
+
+runs:
+  using: "composite"
+  steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: recursive
+      - name: Download build artifacts (GPU)
+        uses: actions/download-artifact@v4
+        with:
+          name: tests_archive_gpu
+      - name: Install Rust toolchain nightly-2025-05-14 (with clippy and rustfmt)
+        shell: bash
+        run: rustup toolchain install nightly-2025-05-14 --component clippy,rustfmt,rust-src
+      - name: Install riscv target
+        shell: bash
+        run: rustup target add riscv32imac-unknown-none-elf --toolchain nightly-2025-10-01
+      # TODO: runner on our GPU server has no sudo access, so we manually installed these; uncomment these once we have proper runners
+      # - name: Install test dependencies
+      #  shell: bash
+      #  run: sudo apt-get update && sudo apt-get install -y binutils-riscv64-unknown-elf lld
+      - name: Install Rust deps
+        shell: bash
+        run: rustup install nightly-2025-10-01 --component rust-src
+      - name: Install Rust deps
+        shell: bash
+        run: rustup install nightly-2025-02-14 --component rust-src
+      - name: Install OpenVM guest toolchain (nightly-2025-08-02)
+        shell: bash
+        run: |
+          rustup toolchain install nightly-2025-08-02
+          rustup component add rust-src --toolchain nightly-2025-08-02
+      - uses: taiki-e/install-action@nextest
+
+

.github/workflows/pr-tests.yml

@@ -199,4 +199,38 @@ jobs:
       - name: Lint no default features
         run: cargo clippy --all --all-targets --all-features --profile pr-tests --verbose -- -D warnings
       - name: Build (GPU)
-        run: cargo build --all-targets --all-features --all --profile pr-tests --verbose
+        run: cargo build --all-targets --all-features --all --profile pr-tests --verbose
+      - uses: taiki-e/install-action@nextest
+      - name: Create tests archive (GPU, features=cuda)
+        run: cargo nextest archive --archive-file tests_gpu.tar.zst --cargo-profile pr-tests --workspace --package powdr-openvm --features cuda
+      - name: Upload build artifacts (GPU)
+        uses: actions/upload-artifact@v4
+        with:
+          name: tests_archive_gpu
+          path: |
+            tests_gpu.tar.zst
+
+  test_quick_gpu:
+    needs: build_gpu
+    runs-on: [self-hosted, GPU]
+    # TODO: we only have one runner on our GPU server, so can't partition yet; uncomment these once we have proper runners
+    # strategy:
+    #   matrix:
+    #     test:
+    #       - "1"
+    #       - "2"
+    #       - "3"
+    #       - "4"
+    #       - "5"
+    #       - "6"
+    #       - "7"
+
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: recursive
+      - name: Init testing instance (GPU)
+        uses: ./.github/actions/init-testing-instance-gpu
+      - name: Run quick GPU tests from powdr-openvm only
+        run: cargo nextest run --archive-file tests_gpu.tar.zst --workspace-remap . --verbose --no-tests=warn
+        # run: cargo nextest run --archive-file tests_gpu.tar.zst --workspace-remap . --verbose --partition count:"${{ matrix.test }}"/7 --no-tests=warn