Improve CMake options

.azure-pipelines/integration-test-rocm.yml

@@ -29,7 +29,7 @@ jobs:
       targetType: 'inline'
       script: |
         mkdir build && cd build
-        CXX=/opt/rocm/bin/hipcc cmake -DCMAKE_BUILD_TYPE=Release -DBYPASS_GPU_CHECK=ON -DUSE_ROCM=ON ..
+        CXX=/opt/rocm/bin/hipcc cmake -DCMAKE_BUILD_TYPE=Release -DMSCCLPP_BYPASS_GPU_CHECK=ON -DMSCCLPP_USE_ROCM=ON ..
         make -j
       workingDirectory: '$(System.DefaultWorkingDirectory)'
 

.azure-pipelines/integration-test.yml

@@ -31,7 +31,7 @@ jobs:
       targetType: 'inline'
       script: |
         mkdir build && cd build
-        cmake -DCMAKE_BUILD_TYPE=Release -DBYPASS_GPU_CHECK=ON -DUSE_CUDA=ON ..
+        cmake -DCMAKE_BUILD_TYPE=Release -DMSCCLPP_BYPASS_GPU_CHECK=ON -DMSCCLPP_USE_CUDA=ON ..
         make -j
       workingDirectory: '$(System.DefaultWorkingDirectory)'
 

.azure-pipelines/multi-nodes-test.yml

@@ -26,7 +26,7 @@ jobs:
       targetType: 'inline'
       script: |
         mkdir build && cd build
-        cmake -DCMAKE_BUILD_TYPE=Release -DBYPASS_GPU_CHECK=ON -DUSE_CUDA=ON ..
+        cmake -DCMAKE_BUILD_TYPE=Release -DMSCCLPP_BYPASS_GPU_CHECK=ON -DMSCCLPP_USE_CUDA=ON ..
         make -j
       workingDirectory: '$(System.DefaultWorkingDirectory)'
 

.github/workflows/codeql-analysis.yml

@@ -46,7 +46,7 @@ jobs:
     - name: Build
       run: |
         rm -rf build && mkdir build && cd build
-        cmake -DBYPASS_GPU_CHECK=ON -DUSE_CUDA=ON ..
+        cmake -DMSCCLPP_BYPASS_GPU_CHECK=ON -DMSCCLPP_USE_CUDA=ON ..
         make -j
 
     - name: Perform CodeQL Analysis
@@ -91,7 +91,7 @@ jobs:
     - name: Build
       run: |
         rm -rf build && mkdir build && cd build
-        CXX=/opt/rocm/bin/hipcc cmake -DBYPASS_GPU_CHECK=ON -DUSE_ROCM=ON ..
+        CXX=/opt/rocm/bin/hipcc cmake -DMSCCLPP_BYPASS_GPU_CHECK=ON -DMSCCLPP_USE_ROCM=ON ..
         make -j
 
     - name: Perform CodeQL Analysis

CMakeLists.txt

@@ -14,19 +14,19 @@ enable_language(CXX)
 list(APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 
 # Options
-option(ENABLE_TRACE "Enable tracing" OFF)
-option(BUILD_TESTS "Build tests" ON)
-option(BUILD_PYTHON_BINDINGS "Build Python bindings" ON)
-option(BUILD_APPS_NCCL "Build NCCL interfaces" ON)
-option(USE_CUDA "Use NVIDIA/CUDA." OFF)
-option(USE_ROCM "Use AMD/ROCm." OFF)
-option(BYPASS_GPU_CHECK "Bypass GPU check." OFF)
-
-if(BYPASS_GPU_CHECK)
-    if(USE_CUDA)
+option(MSCCLPP_ENABLE_TRACE "Enable tracing" OFF)
+option(MSCCLPP_BUILD_TESTS "Build tests" ON)
+option(MSCCLPP_BUILD_PYTHON_BINDINGS "Build Python bindings" ON)
+option(MSCCLPP_BUILD_APPS_NCCL "Build NCCL interfaces" ON)
+option(MSCCLPP_USE_CUDA "Use NVIDIA/CUDA." OFF)
+option(MSCCLPP_USE_ROCM "Use AMD/ROCm." OFF)
+option(MSCCLPP_BYPASS_GPU_CHECK "Bypass GPU check." OFF)
+
+if(MSCCLPP_BYPASS_GPU_CHECK)
+    if(MSCCLPP_USE_CUDA)
         message("Bypassing GPU check: using NVIDIA/CUDA.")
         find_package(CUDAToolkit REQUIRED)
-    elseif(USE_ROCM)
+    elseif(MSCCLPP_USE_ROCM)
         message("Bypassing GPU check: using AMD/ROCm.")
         # Temporal fix for rocm5.6
         set(CMAKE_PREFIX_PATH "/opt/rocm;${CMAKE_PREFIX_PATH}")
@@ -40,16 +40,16 @@ else()
     include(CheckAmdGpu)
     if(NVIDIA_FOUND AND AMD_FOUND)
         message("Detected NVIDIA/CUDA and AMD/ROCm: prioritizing NVIDIA/CUDA.")
-        set(USE_CUDA ON)
-        set(USE_ROCM OFF)
+        set(MSCCLPP_USE_CUDA ON)
+        set(MSCCLPP_USE_ROCM OFF)
     elseif(NVIDIA_FOUND)
         message("Detected NVIDIA/CUDA.")
-        set(USE_CUDA ON)
-        set(USE_ROCM OFF)
+        set(MSCCLPP_USE_CUDA ON)
+        set(MSCCLPP_USE_ROCM OFF)
     elseif(AMD_FOUND)
         message("Detected AMD/ROCm.")
-        set(USE_CUDA OFF)
-        set(USE_ROCM ON)
+        set(MSCCLPP_USE_CUDA OFF)
+        set(MSCCLPP_USE_ROCM ON)
     else()
         message(FATAL_ERROR "Neither NVIDIA/CUDA nor AMD/ROCm is found.")
     endif()
@@ -58,7 +58,7 @@ endif()
 # Declare project
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wextra")
-if(USE_CUDA)
+if(MSCCLPP_USE_CUDA)
     set(CMAKE_CUDA_STANDARD 17)
     set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler -Wall,-Wextra")
     project(mscclpp LANGUAGES CXX CUDA)
@@ -115,13 +115,13 @@ if(IBVERBS_FOUND)
     target_compile_definitions(mscclpp_obj PUBLIC USE_IBVERBS)
 endif()
 set_target_properties(mscclpp_obj PROPERTIES LINKER_LANGUAGE CXX POSITION_INDEPENDENT_CODE 1 VERSION ${MSCCLPP_VERSION} SOVERSION ${MSCCLPP_SOVERSION})
-if(USE_CUDA)
-    target_compile_definitions(mscclpp_obj PRIVATE USE_CUDA)
-elseif(USE_ROCM)
-    target_compile_definitions(mscclpp_obj PRIVATE USE_ROCM)
+if(MSCCLPP_USE_CUDA)
+    target_compile_definitions(mscclpp_obj PRIVATE MSCCLPP_USE_CUDA)
+elseif(MSCCLPP_USE_ROCM)
+    target_compile_definitions(mscclpp_obj PRIVATE MSCCLPP_USE_ROCM)
 endif()
-if(ENABLE_TRACE)
-    target_compile_definitions(mscclpp_obj PRIVATE ENABLE_TRACE)
+if(MSCCLPP_ENABLE_TRACE)
+    target_compile_definitions(mscclpp_obj PRIVATE MSCCLPP_ENABLE_TRACE)
 endif()
 if(NPKIT_FLAGS)
     target_compile_definitions(mscclpp_obj PRIVATE ${NPKIT_FLAGS})
@@ -150,17 +150,17 @@ install(TARGETS mscclpp_static
     ARCHIVE DESTINATION ${INSTALL_PREFIX}/lib)
 
 # Tests
-if(BUILD_TESTS)
+if(MSCCLPP_BUILD_TESTS)
     enable_testing() # Called here to allow ctest from the build directory
     add_subdirectory(test)
 endif()
 
 # Python bindings
-if(BUILD_PYTHON_BINDINGS)
+if(MSCCLPP_BUILD_PYTHON_BINDINGS)
     add_subdirectory(python)
 endif()
 
 # NCCL interfaces
-if(BUILD_APPS_NCCL)
+if(MSCCLPP_BUILD_APPS_NCCL)
     add_subdirectory(apps/nccl)
 endif()

apps/nccl/CMakeLists.txt

@@ -4,7 +4,7 @@
 file(GLOB_RECURSE SOURCES CONFIGURE_DEPENDS src/*)
 file(GLOB_RECURSE HEADERS CONFIGURE_DEPENDS include/nccl.h)
 
-if(USE_ROCM)
+if(MSCCLPP_USE_ROCM)
     set_source_files_properties(${SOURCES} PROPERTIES LANGUAGE CXX)
 endif()
 
@@ -14,10 +14,10 @@ target_sources(mscclpp_nccl_obj PUBLIC FILE_SET HEADERS FILES ${HEADERS})
 target_include_directories(mscclpp_nccl_obj PRIVATE include ${PROJECT_SOURCE_DIR}/src/include SYSTEM PRIVATE ${GPU_INCLUDE_DIRS})
 target_link_libraries(mscclpp_nccl_obj PRIVATE ${GPU_LIBRARIES} PUBLIC mscclpp_obj)
 set_target_properties(mscclpp_nccl_obj PROPERTIES LINKER_LANGUAGE CXX POSITION_INDEPENDENT_CODE 1 VERSION ${MSCCLPP_VERSION} SOVERSION ${MSCCLPP_SOVERSION})
-if(USE_CUDA)
-    target_compile_definitions(mscclpp_nccl_obj PRIVATE USE_CUDA)
-elseif(USE_ROCM)
-    target_compile_definitions(mscclpp_nccl_obj PRIVATE USE_ROCM)
+if(MSCCLPP_USE_CUDA)
+    target_compile_definitions(mscclpp_nccl_obj PRIVATE MSCCLPP_USE_CUDA)
+elseif(MSCCLPP_USE_ROCM)
+    target_compile_definitions(mscclpp_nccl_obj PRIVATE MSCCLPP_USE_ROCM)
 endif()
 
 add_library(mscclpp_nccl SHARED)
@@ -34,6 +34,6 @@ install(TARGETS mscclpp_nccl
 install(TARGETS mscclpp_nccl_static
     ARCHIVE DESTINATION ${INSTALL_PREFIX}/lib)
 
-if(BUILD_TESTS)
+if(MSCCLPP_BUILD_TESTS)
     add_subdirectory(test)
 endif()

docs/getting-started/quickstart.md

@@ -13,7 +13,17 @@
     * AMD MI250X GPUs + ROCm >= 5.7
     * AMD MI300X GPUs + ROCm >= 6.0
 * OS: tested over Ubuntu 18.04 and 20.04
-* Libraries: [libnuma](https://github.com/numactl/numactl), MPI (optional)
+* Libraries
+    * [libnuma](https://github.com/numactl/numactl)
+        ```bash
+        sudo apt-get install libnuma-dev
+        ```
+    * (Optional, for [building the Python module](#install-from-source-python-module)) Python >= 3.8 and Python Development Package
+        ```bash
+        sudo apt-get satisfy "python3 (>=3.8), python3-dev (>=3.8)"
+        ```
+        If you don't want to build Python module, you need to set `-DMSCCLPP_BUILD_PYTHON_BINDINGS=OFF` in your `cmake` command (see details in [Install from Source (Libraries and Headers)](#install-from-source-libraries-and-headers)).
+    * (Optional, for benchmarks) MPI
 * Others
     * For NVIDIA platforms, `nvidia_peermem` driver should be loaded on all nodes. Check it via:
         ```
@@ -60,11 +70,12 @@ $ CXX=/path/to/hipcc cmake -DCMAKE_BUILD_TYPE=Release ..
 $ make -j
 ```
 
+(install-from-source-libraries-and-headers)=
 ## Install from Source (Libraries and Headers)
 
 ```bash
 # Install the generated headers and binaries to /usr/local/mscclpp
-$ cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local/mscclpp -DBUILD_PYTHON_BINDINGS=OFF ..
+$ cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local/mscclpp -DMSCCLPP_BUILD_PYTHON_BINDINGS=OFF ..
 $ make -j mscclpp mscclpp_static
 $ sudo make install/fast
 ```

pyproject.toml

@@ -16,8 +16,8 @@ wheel.packages = ["python/mscclpp", "python/mscclpp_benchmark"]
 wheel.install-dir = "mscclpp"
 
 [tool.scikit-build.cmake.define]
-BUILD_PYTHON_BINDINGS = "ON"
-BUILD_TESTS = "OFF"
+MSCCLPP_BUILD_PYTHON_BINDINGS = "ON"
+MSCCLPP_BUILD_TESTS = "OFF"
 
 [tool.black]
 line-length = 120

src/bootstrap/socket.cc

@@ -78,7 +78,7 @@ static int envSocketFamily(void) {
 
 static int findInterfaces(const char* prefixList, char* names, union SocketAddress* addrs, int sock_family,
                           int maxIfNameSize, int maxIfs) {
-#ifdef ENABLE_TRACE
+#ifdef MSCCLPP_ENABLE_TRACE
   char line[SOCKET_NAME_MAXLEN + 1];
 #endif
   struct mscclpp::netIf userIfs[MAX_IFS];
@@ -184,7 +184,7 @@ static bool matchSubnet(struct ifaddrs local_if, union SocketAddress* remote) {
 
 int FindInterfaceMatchSubnet(char* ifNames, union SocketAddress* localAddrs, union SocketAddress* remoteAddr,
                              int ifNameMaxSize, int maxIfs) {
-#ifdef ENABLE_TRACE
+#ifdef MSCCLPP_ENABLE_TRACE
   char line[SOCKET_NAME_MAXLEN + 1];
 #endif
   char line_a[SOCKET_NAME_MAXLEN + 1];
@@ -436,7 +436,7 @@ void Socket::bind() {
 
 void Socket::bindAndListen() {
   bind();
-#ifdef ENABLE_TRACE
+#ifdef MSCCLPP_ENABLE_TRACE
   char line[SOCKET_NAME_MAXLEN + 1];
   TRACE(MSCCLPP_INIT | MSCCLPP_NET, "Listening on socket %s", SocketToString(&addr_, line));
 #endif
@@ -452,7 +452,7 @@ void Socket::bindAndListen() {
 
 void Socket::connect(int64_t timeout) {
   mscclpp::Timer timer;
-#ifdef ENABLE_TRACE
+#ifdef MSCCLPP_ENABLE_TRACE
   char line[SOCKET_NAME_MAXLEN + 1];
 #endif
   const int one = 1;

src/include/atomic.hpp

@@ -4,14 +4,14 @@
 #ifndef MSCCLPP_ATOMIC_HPP_
 #define MSCCLPP_ATOMIC_HPP_
 
-#if defined(USE_CUDA)
+#if defined(MSCCLPP_USE_CUDA)
 #define MSCCLPP_DEVICE_CUDA
 #include <mscclpp/atomic_device.hpp>
 #undef MSCCLPP_DEVICE_CUDA
-#else  // !defined(USE_CUDA)
+#else  // !defined(MSCCLPP_USE_CUDA)
 #define MSCCLPP_DEVICE_HIP
 #include <mscclpp/atomic_device.hpp>
 #undef MSCCLPP_DEVICE_HIP
-#endif  // !defined(USE_CUDA)
+#endif  // !defined(MSCCLPP_USE_CUDA)
 
 #endif  // MSCCLPP_ATOMIC_HPP_

src/include/debug.h

@@ -107,7 +107,7 @@ extern char mscclppLastError[];
 #define INFO(FLAGS, ...) mscclppDebugLog(MSCCLPP_LOG_INFO, (FLAGS), __func__, __LINE__, __VA_ARGS__)
 #define TRACE_CALL(...) mscclppDebugLog(MSCCLPP_LOG_TRACE, MSCCLPP_CALL, __func__, __LINE__, __VA_ARGS__)
 
-#ifdef ENABLE_TRACE
+#ifdef MSCCLPP_ENABLE_TRACE
 #define TRACE(FLAGS, ...) mscclppDebugLog(MSCCLPP_LOG_TRACE, (FLAGS), __func__, __LINE__, __VA_ARGS__)
 extern std::chrono::steady_clock::time_point mscclppEpoch;
 #else

test/CMakeLists.txt

@@ -11,7 +11,7 @@ set(TEST_LIBS_GTEST GTest::gtest_main GTest::gmock_main)
 set(TEST_INC_COMMON PRIVATE ${PROJECT_SOURCE_DIR}/include SYSTEM PRIVATE ${GPU_INCLUDE_DIRS})
 set(TEST_INC_INTERNAL PRIVATE ${PROJECT_SOURCE_DIR}/src/include)
 
-if(USE_ROCM)
+if(MSCCLPP_USE_ROCM)
     file(GLOB_RECURSE CU_SOURCES CONFIGURE_DEPENDS *.cu)
     set_source_files_properties(${CU_SOURCES} PROPERTIES LANGUAGE CXX)
 endif()

test/mscclpp-test/CMakeLists.txt

@@ -5,7 +5,7 @@ FetchContent_Declare(json URL https://github.com/nlohmann/json/releases/download
 FetchContent_MakeAvailable(json)
 
 function(add_mscclpp_test_executable name sources)
-    if(USE_ROCM)
+    if(MSCCLPP_USE_ROCM)
         set_source_files_properties(${sources} PROPERTIES LANGUAGE CXX)
     endif()
     add_executable(${name} ${sources} common.cc)