Tcp End

CMakeLists.txt

@@ -13,6 +13,7 @@ slime_option(USE_MACA                  "USE in MACA Platform"             OFF)
 
 slime_option(BUILD_NVLINK              "Build NVLINK"                     OFF)
 slime_option(BUILD_ASCEND_DIRECT       "Build Ascend direct transport"    OFF)
+slime_option(BUILD_TCP                 "Build TCP transport"               ON)
 
 # Slime options for custom python wrapper
 slime_option(BUILD_PYTHON              "Build python wrapper"             OFF)

dlslime/csrc/CMakeLists.txt

@@ -27,6 +27,10 @@ if(BUILD_RDMA)
     target_link_libraries(dlslime INTERFACE _slime_rdma)
 endif()
 
+if(BUILD_TCP)
+    target_link_libraries(dlslime INTERFACE _slime_tcp)
+endif()
+
 # rpc/ has no independent C++ consumers (the session API is all
 # pybind11). It is compiled straight into _slime_c.so by the python
 # subdirectory below. Keep the source files here as a marker so future

dlslime/csrc/device/cuda/cuda_signal.h

@@ -8,7 +8,7 @@
 #include "dlslime/csrc/device/signal.h"
 #include "dlslime/csrc/engine/rdma/rdma_env.h"
 #include "dlslime/csrc/logging.h"
-#include "dlslime/csrc/pause.h"
+#include "dlslime/csrc/common/pause.h"
 #include "nvtx_helper.h"
 
 namespace dlslime {

dlslime/csrc/engine/CMakeLists.txt

@@ -34,3 +34,7 @@ endif()
 if (BUILD_RDMA)
     add_subdirectory(rdma)
 endif()
+
+if (BUILD_TCP)
+    add_subdirectory(tcp)
+endif()

dlslime/csrc/engine/tcp/CMakeLists.txt

@@ -0,0 +1,47 @@
+# asio is header-only.  Try find_package, fall back to manual detection.
+find_package(asio QUIET)
+if(NOT asio_FOUND)
+    if(EXISTS /usr/include/asio.hpp)
+        add_library(asio::asio INTERFACE IMPORTED)
+        target_include_directories(asio::asio INTERFACE /usr/include)
+    elseif(EXISTS /usr/include/boost/asio.hpp)
+        add_library(asio::asio INTERFACE IMPORTED)
+        target_include_directories(asio::asio INTERFACE /usr/include/boost)
+        target_compile_definitions(asio::asio INTERFACE ASIO_STANDALONE)
+    else()
+        message(FATAL_ERROR "asio not found. Install libasio-dev or boost.")
+    endif()
+endif()
+
+add_library(_slime_tcp SHARED
+    tcp_memory_pool.cpp
+    tcp_connection_pool.cpp
+    tcp_context.cpp
+    tcp_session.cpp
+    tcp_endpoint.cpp
+)
+
+target_compile_definitions(_slime_tcp PRIVATE ASIO_STANDALONE)
+
+if (USE_CUDA)
+    find_package(CUDAToolkit REQUIRED)
+    target_compile_definitions(_slime_tcp PRIVATE USE_CUDA)
+    target_include_directories(_slime_tcp PRIVATE ${CUDAToolkit_INCLUDE_DIRS})
+    target_link_libraries(_slime_tcp PUBLIC CUDA::cudart)
+endif()
+
+target_link_libraries(_slime_tcp PUBLIC
+    asio::asio
+    _slime_device
+    _slime_engine
+)
+
+set_target_properties(_slime_tcp PROPERTIES
+    BUILD_WITH_INSTALL_RPATH TRUE
+    INSTALL_RPATH "${ORIGIN}"
+)
+
+install(TARGETS _slime_tcp
+    EXPORT dlslimeTargets
+    LIBRARY DESTINATION ${DLSLIME_INSTALL_PATH}
+)

dlslime/csrc/engine/tcp/build_and_test.sh

@@ -0,0 +1,65 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+REPO_ROOT="$(cd "$SCRIPT_DIR/../../../.." && pwd)"
+BUILD_DIR="$REPO_ROOT/build_tcp"
+MODE="${1:-all}"
+
+# Optional: USE_CUDA=ON ./build_and_test.sh all
+USE_CUDA="${USE_CUDA:-OFF}"
+
+header() { echo; echo -e "\033[1;36m==>\033[m \033[1m$*\033[m"; }
+ok()     { echo -e "  \033[1;32mOK\033[m   $*"; }
+
+do_build() {
+    rm -rf ${BUILD_DIR}
+    local cuda_label=""
+    [[ "$USE_CUDA" == "ON" ]] && cuda_label=" + USE_CUDA=ON"
+
+    header "Configuring (BUILD_TCP=ON, BUILD_RDMA=OFF${cuda_label})"
+    cmake -S "$REPO_ROOT" -B "$BUILD_DIR" -G Ninja \
+        -DCMAKE_BUILD_TYPE=Release \
+        -DDLSLIME_INSTALL_PATH=dlslime \
+        -DBUILD_PYTHON=ON \
+        -DBUILD_RDMA=OFF \
+        -DBUILD_TCP=ON \
+        -DBUILD_NVLINK=OFF \
+        -DBUILD_ASCEND_DIRECT=OFF \
+        -DUSE_CUDA="$USE_CUDA" \
+        -DSKBUILD_PROJECT_NAME=dlslime 2>&1 | tail -3
+    ok "CMake configure"
+
+    header "Building _slime_c"
+    cmake --build "$BUILD_DIR" --target _slime_c -j"$(nproc)" 2>&1 | tail -8
+    ok "Build complete"
+
+    cp "$BUILD_DIR/lib/"*.so "$REPO_ROOT/dlslime/"
+    ok "Copied .so files to dlslime/"
+}
+
+do_test() {
+    header "Running TcpEndpoint tests"
+    export SLIME_LOG_LEVEL=1
+    export LD_LIBRARY_PATH="$REPO_ROOT/dlslime"
+    export PYTHONPATH="$REPO_ROOT"
+    python3 "$SCRIPT_DIR/test_tcp_endpoint.py" 2>&1 | while IFS= read -r line; do
+        if   [[ "$line" == *"PASSED"*  ]]; then echo -e "  \033[1;32m✓\033[m $line"
+        elif [[ "$line" == *"SKIP"*    ]]; then echo -e "  \033[1;33m⊘\033[m $line"
+        elif [[ "$line" == *"FAIL"*    ]]; then echo -e "  \033[1;91m✗\033[m $line"
+        else echo "  $line"
+        fi
+    done
+    echo " tests done "
+}
+
+case "$MODE" in
+    all)    do_build; do_test ;;
+    build)  do_build ;;
+    test)   do_test ;;
+    clean)  rm -rf "$BUILD_DIR" "$REPO_ROOT/dlslime/_slime_c"*.so "$REPO_ROOT/dlslime/lib_slime_"*.so
+            ok "Cleaned" ;;
+    *)      echo "Usage: $0 {all|build|test|clean}" >&2
+            echo "  USE_CUDA=ON $0 all   # build + test with CUDA" >&2
+            exit 1 ;;
+esac