Pr a ci build

.github/workflows/lint.yml

@@ -7,41 +7,38 @@
 
 jobs:
   linters:
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-24.04
 
     steps:
     - name: Check out Git repository
       uses: actions/checkout@v4
 
     - name: Install ClangFormat
       run: sudo apt-get install -y clang-format
 
-    - name: Run git-clang-format
-      run: git clang-format --style=file --diff
-
     - name: Set up Python
-      uses: actions/setup-python@v4
+      uses: actions/setup-python@v5
       with:
-        python-version: 3.8
+        python-version: '3.12'
 
     - name: Install Python dependencies
-      run: python3.8 -m pip install black
+      run: pip install black
 
-    - name: Run black
-      run: python3.8 -m black --check --config pyproject.toml .
+    - name: Run lint
+      run: bash tools/lint.sh dry
 
   spelling:
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-24.04
 
     steps:
     - name: Check out Git repository
       uses: actions/checkout@v4
 
     - name: Download misspell
       run: |
         curl -L https://github.com/client9/misspell/releases/download/v0.3.4/misspell_0.3.4_linux_64bit.tar.gz -o /tmp/misspell_0.3.4_linux_64bit.tar.gz
         tar -xzf /tmp/misspell_0.3.4_linux_64bit.tar.gz -C .
 
     - name: Check spelling
       run: |
         ./misspell -error .github ark examples python scripts

.github/workflows/ut-cuda.yml → .github/workflows/ut.yml

@@ -1,4 +1,4 @@
-name: "Unit Tests (CUDA)"
+name: "Unit Tests"
 
 on:
   push:
@@ -8,49 +8,68 @@ on:
     branches:
       - main
     types: [opened, synchronize, reopened, ready_for_review]
+  schedule:
+    - cron: '42 20 * * 4'
 
 jobs:
   UnitTest:
-    runs-on: [ self-hosted, A100 ]
+    if: github.event_name != 'schedule'
     defaults:
       run:
         shell: bash
-    timeout-minutes: 30
+    timeout-minutes: 60
+    permissions:
+      actions: read
+      contents: read
     strategy:
+      fail-fast: false
       matrix:
-        cuda: [ cuda11.8, cuda12.2 ]
+        include:
+          - platform: cuda
+            runner: [self-hosted, CUDA]
+            container: nvcr.io/nvidia/pytorch:26.03-py3
+            container_options: --privileged --ipc=host --gpus=all --ulimit memlock=-1:-1
+          - platform: rocm
+            runner: [self-hosted, ROCM]
+            container: rocm/pytorch:rocm6.2.3_ubuntu22.04_py3.10_pytorch_release_2.3.0
+            container_options: --privileged --ipc=host --security-opt seccomp=unconfined --group-add video --ulimit memlock=-1:-1
+    runs-on: ${{ matrix.runner }}
     concurrency:
-      group: ${{ github.workflow }}-${{ github.ref }}-${{ matrix.cuda }}
+      group: ${{ github.workflow }}-${{ matrix.platform }}-${{ github.ref }}
       cancel-in-progress: true
     container:
-      image: "ghcr.io/microsoft/ark/ark:base-dev-${{ matrix.cuda }}"
-      options: --privileged --ipc=host --gpus=all --ulimit memlock=-1:-1
+      image: ${{ matrix.container }}
+      options: ${{ matrix.container_options }}
 
     steps:
       - name: Checkout
         uses: actions/checkout@v4
 
-      - name: LockGPUClock
-        run: |
-          sudo nvidia-smi -pm 1
-          for i in $(seq 0 $(( $(nvidia-smi -L | wc -l) - 1 ))); do
-            sudo nvidia-smi -ac $(nvidia-smi --query-gpu=clocks.max.memory,clocks.max.sm --format=csv,noheader,nounits -i $i | sed 's/\ //') -i $i
-          done
-
       - name: Dubious ownership exception
         run: |
           git config --global --add safe.directory /__w/ark/ark
 
       - name: Build
         run: |
+          apt-get update && apt-get install -y lcov
           mkdir build && cd build
-          cmake -DCMAKE_BUILD_TYPE=Debug ..
+          CMAKE_ARGS="-DCMAKE_BUILD_TYPE=Debug"
+          if [ "${{ matrix.platform }}" = "rocm" ]; then
+            CMAKE_ARGS="$CMAKE_ARGS -DCMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc"
+          fi
+          cmake $CMAKE_ARGS ..
           make -j ut ark_py
 
       - name: Run C++ UT
+        if: github.event_name != 'schedule'
         run: |
           cd build
           ARK_ROOT=$PWD ctest --stop-on-failure --verbose --schedule-random
+
+      - name: C++ Coverage
+        if: github.event_name != 'schedule'
+        run: |
+          cd build
           lcov --capture --directory . --output-file cpp_coverage.info
           lcov --remove cpp_coverage.info \
               '/usr/*' \
@@ -65,19 +84,22 @@ jobs:
           lcov --list cpp_coverage.info
 
       - name: Install Python Dependencies
+        if: github.event_name != 'schedule'
         run: |
           python3 -m pip install -r requirements.txt
 
       - name: Run Python UT
+        if: github.event_name != 'schedule'
         run: |
           cd build
           PYTHONPATH=$PWD/python ARK_ROOT=$PWD python3 -m pytest \
               --cov=python/ark \
               --cov-report lcov:py_coverage.info \
               --verbose \
-              ../python/unittest/test.py
+              ../python/unittest/
 
       - name: Report Coverage
+        if: github.event_name != 'schedule'
         env:
           CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
         run: |
@@ -86,9 +108,12 @@ jobs:
           bash <(curl -s https://codecov.io/bash) -f coverage.info || echo "Codecov did not collect coverage reports"
 
       - name: Install Python
+        if: github.event_name != 'schedule'
         run: |
           python3 -m pip install .
 
       - name: Run Tutorials
+        if: github.event_name != 'schedule'
         run: |
           python3 ./examples/tutorial/quickstart_tutorial.py
+

CMakeLists.txt

@@ -65,13 +65,27 @@ if(ARK_USE_CUDA)
     endif()
 
     # Set CUDA architectures
-    if(CUDAToolkit_VERSION_MAJOR GREATER_EQUAL 11)
+    if(CUDAToolkit_VERSION_MAJOR GREATER_EQUAL 13)
+        # CUDA 13+ dropped sm_60 and sm_70
+        set(CMAKE_CUDA_ARCHITECTURES 80 90)
+    elseif(CUDAToolkit_VERSION_MAJOR GREATER_EQUAL 12)
+        set(CMAKE_CUDA_ARCHITECTURES 60 70 80 90)
+    elseif(CUDAToolkit_VERSION_MAJOR GREATER_EQUAL 11)
         set(CMAKE_CUDA_ARCHITECTURES 60 70 80)
     endif()
 
-    # Hopper architecture
-    if(CUDAToolkit_VERSION_MAJOR GREATER_EQUAL 12)
-        set(CMAKE_CUDA_ARCHITECTURES ${CMAKE_CUDA_ARCHITECTURES} 90)
+    # CUDA 13+ moved CCCL headers into a cccl/ subdirectory.
+    # Add it to the include path so third-party code (e.g. MSCCL++)
+    # that includes <cuda/atomic> can still find the headers.
+    if(CUDAToolkit_VERSION_MAJOR GREATER_EQUAL 13)
+        list(GET CUDAToolkit_INCLUDE_DIRS 0 _CUDA_INCLUDE_FIRST)
+        set(CCCL_INCLUDE_DIR "${_CUDA_INCLUDE_FIRST}/cccl")
+        if(EXISTS "${CCCL_INCLUDE_DIR}")
+            include_directories(SYSTEM "${CCCL_INCLUDE_DIR}")
+            message(STATUS "CUDA 13+: added CCCL include dir ${CCCL_INCLUDE_DIR}")
+        else()
+            message(WARNING "CUDA 13+: CCCL include dir not found at ${CCCL_INCLUDE_DIR}. Build may fail.")
+        endif()
     endif()
 else()  # ARK_USE_ROCM
     set(CMAKE_HIP_STANDARD 17)

README.md

@@ -4,15 +4,13 @@ A GPU-driven system framework for scalable AI applications.
 
 [![Latest Release](https://img.shields.io/github/release/microsoft/ark.svg)](https://github.com/microsoft/ark/releases/latest)
 [![License](https://img.shields.io/github/license/microsoft/ark.svg)](LICENSE)
-[![CodeQL](https://github.com/microsoft/ark/actions/workflows/codeql.yml/badge.svg)](https://github.com/microsoft/ark/actions/workflows/codeql.yml)
+[![Unit Tests](https://github.com/microsoft/ark/actions/workflows/ut.yml/badge.svg)](https://github.com/microsoft/ark/actions/workflows/ut.yml)
 [![codecov](https://codecov.io/gh/microsoft/ark/graph/badge.svg?token=XmMOK85GOB)](https://codecov.io/gh/microsoft/ark)
 
 | Pipelines         | Build Status      |
 |-------------------|-------------------|
 | Unit Tests (CUDA) | [![Build Status](https://dev.azure.com/binyli/HPC/_apis/build/status%2Fark-test?branchName=main)](https://dev.azure.com/binyli/HPC/_build/latest?definitionId=6&branchName=main) |
-| Unit Tests (ROCm) | [![Unit Tests (ROCm)](https://github.com/microsoft/ark/actions/workflows/ut-rocm.yml/badge.svg?branch=main)](https://github.com/microsoft/ark/actions/workflows/ut-rocm.yml) |
-
-*NOTE (Nov 2023): ROCm unit tests will be replaced into an Azure pipeline in the future.*
+| Unit Tests | [![Unit Tests](https://github.com/microsoft/ark/actions/workflows/ut.yml/badge.svg?branch=main)](https://github.com/microsoft/ark/actions/workflows/ut.yml) |
 
 See [Quick Start](docs/quickstart.md) to quickly get started.
 

ark/api/context.cpp

@@ -29,8 +29,6 @@ void Context::set(const std::string& key, const std::string& value,
     this->impl_->set(key, value_json, type);
 }
 
-std::string Context::dump() const {
-    return this->impl_->dump().dump();
-}
+std::string Context::dump() const { return this->impl_->dump().dump(); }
 
 }  // namespace ark