Enable RDC (ROCm Data Center Tool) build in TheRock (#1981)

This change integrates ROCm Data Center Tools (RDC) support into TheRock and includes comprehensive updates across the build system, third-party dependency bundling, and documentation.

Major changes:
1. RDC Feature Integration:
    - Adds a new build flag (THEROCK_ENABLE_DC_TOOLS) to enable data center tools as a group, including a specific flag (THEROCK_ENABLE_RDC) for ROCm Data Center Tool (RDC). These appear in CMake and are documented in the README.
    - Introduces a new CMake-controlled subdirectory (dctools/) for RDC, which is Linux-only and requires rocprofiler-sdk v3.
    - Integrates RDC validation and binary checks into the build and test systems.
2. gRPC Static Library Bundling:
    - Adds a new third-party/grpc/ directory to fetch, build, and bundle gRPC as a static library for RDC's use, with pre- and post-build scripts for correct static linking and symbol handling, primarily on Linux.
    - Ensures relocatability by patching install paths in pc and CMake files.
3. libcap Bundling on Linux:
    - Bundles libcap as a new third-party dependency under third-party/sysdeps/linux/libcap for privileged operations on Linux.
    - Provides full build, patching, and symbol versioning logic so the library can be bundled, relocated, and used safely within TheRock.
    - Integrates libcap as a dependency of RDC and documents its use in the developer docs.
4. Artifact and Build Infrastructure Updates:
    - Updates the artifact manifest and sysdeps dependency tracking to include libcap and gRPC in the correct build groups.
    - Ensures consistent interface, rpath, and packaging for all new libraries.
5. Documentation Updates:
    - Extends the README and developer docs to document new options, usage scenarios, and library inclusion (RDC, libcap, gRPC).
    - Updates RFC0007 to reflect the latest artifact layout for RDC and the static linking rationale.


Signed-off-by: Wang, Yanyao <[email protected]>

Author: WBobby

CMakeLists.txt

@@ -184,6 +184,7 @@ option(THEROCK_ENABLE_COMM_LIBS "Enable building of comm libraries" "${THEROCK_E
 option(THEROCK_ENABLE_MATH_LIBS "Enable building of math libraries" "${THEROCK_ENABLE_ALL}")
 option(THEROCK_ENABLE_ML_LIBS "Enable building of ML libraries" "${THEROCK_ENABLE_ALL}")
 option(THEROCK_ENABLE_PROFILER "Enable building the profiler libraries" "${THEROCK_ENABLE_ALL}")
+option(THEROCK_ENABLE_DC_TOOLS "Enable building of data center tools" "${THEROCK_ENABLE_ALL}")
 option(THEROCK_ENABLE_HOST_MATH "Build all bundled host math libraries by default" OFF)
 option(THEROCK_RESET_FEATURES "One-shot flag which forces all feature flags to their default state for this configuration run" OFF)
 
@@ -261,6 +262,16 @@ else()
   set(_optional_profiler_requirements "")
 endif()
 
+# Systems Tools Features
+if(NOT WIN32)
+  # RDC is Linux-only and requires rocprofiler-sdk v3
+  therock_add_feature(RDC
+    GROUP DC_TOOLS
+    DESCRIPTION "Enables ROCm Data Center Tool (RDC)"
+    REQUIRES CORE_RUNTIME ROCPROFV3
+  )
+endif()
+
 # Comm-libs Features.
 if(NOT WIN32)
   # RCCL is Linux-only.
@@ -419,6 +430,7 @@ endif()
 
 set(THEROCK_BUNDLED_BZIP2)
 set(THEROCK_BUNDLED_ELFUTILS)
+set(THEROCK_BUNDLED_LIBCAP)
 set(THEROCK_BUNDLED_LIBDRM)
 set(THEROCK_BUNDLED_LIBLZMA)
 set(THEROCK_BUNDLED_NUMACTL)
@@ -439,6 +451,7 @@ if(THEROCK_BUNDLE_SYSDEPS)
     endif()
     set(THEROCK_BUNDLED_BZIP2 therock-bzip2)
     set(THEROCK_BUNDLED_ELFUTILS therock-elfutils)
+    set(THEROCK_BUNDLED_LIBCAP therock-libcap)
     set(THEROCK_BUNDLED_LIBDRM therock-libdrm)
     set(THEROCK_BUNDLED_LIBLZMA therock-liblzma)
     set(THEROCK_BUNDLED_NUMACTL therock-numactl)
@@ -480,6 +493,8 @@ add_subdirectory(core)
 # Note that rocprofiler-register is in base and is what higher level clients
 # depend on. The profiler itself is independent.
 add_subdirectory(profiler)
+# Data center tools (RDC, etc.)
+add_subdirectory(dctools)
 add_subdirectory(comm-libs)
 add_subdirectory(math-libs)
 add_subdirectory(ml-libs)

README.md

@@ -159,6 +159,7 @@ enable/disable selected subsets:
 | `-DTHEROCK_ENABLE_MATH_LIBS=OFF` | Disables all math libraries          |
 | `-DTHEROCK_ENABLE_ML_LIBS=OFF`   | Disables all ML libraries            |
 | `-DTHEROCK_ENABLE_PROFILER=OFF`  | Disables profilers                   |
+| `-DTHEROCK_ENABLE_DC_TOOLS=OFF`  | Disables data center tools           |
 
 Individual features can be controlled separately (typically in combination with
 `-DTHEROCK_ENABLE_ALL=OFF` or `-DTHEROCK_RESET_FEATURES=ON` to force a
@@ -182,6 +183,7 @@ minimal build):
 | `-DTHEROCK_ENABLE_MIOPEN_PLUGIN=ON` | Enables MIOpen_plugin                         |
 | `-DTHEROCK_ENABLE_HIPDNN=ON`        | Enables hipDNN                                |
 | `-DTHEROCK_ENABLE_ROCWMMA=ON`       | Enables rocWMMA                               |
+| `-DTHEROCK_ENABLE_RDC=ON`           | Enables ROCm Data Center Tool (Linux only)    |
 
 > [!TIP]
 > Enabling any features will implicitly enable their *minimum* dependencies. Some

dctools/CMakeLists.txt

@@ -0,0 +1,106 @@
+# Data Center Tools
+
+if(THEROCK_ENABLE_RDC)
+  if(WIN32)
+    message(FATAL_ERROR "RDC requires Linux (Windows not supported)")
+  endif()
+
+  # Get gRPC stage directory for RDC's GRPC_ROOT requirement
+  get_target_property(_grpc_stage therock-grpc THEROCK_STAGE_DIR)
+
+  therock_cmake_subproject_declare(rdc
+    EXTERNAL_SOURCE_DIR "${THEROCK_ROCM_SYSTEMS_SOURCE_DIR}/projects/rdc"
+    BINARY_DIR "${CMAKE_CURRENT_BINARY_DIR}/rdc"
+    BACKGROUND_BUILD
+
+    CMAKE_ARGS
+      -DBUILD_PROFILER=ON
+      -DBUILD_STANDALONE=ON
+      -DBUILD_RUNTIME=ON
+      -DBUILD_RVS=OFF
+      -DBUILD_TESTS=${THEROCK_BUILD_TESTING}
+      -DHIP_PLATFORM=amd
+      -DCMAKE_CXX_STANDARD=17
+      -DGRPC_ROOT=${_grpc_stage}
+
+    BUILD_DEPS
+      therock-grpc
+
+    RUNTIME_DEPS
+      ROCR-Runtime
+      amdsmi
+      rocprofiler-sdk
+      ${THEROCK_BUNDLED_LIBCAP}
+      ${THEROCK_BUNDLED_ZLIB}
+
+    INSTALL_DESTINATION
+      portable-rdc
+
+    INTERFACE_LINK_DIRS
+      lib
+
+    INTERFACE_INSTALL_RPATH_DIRS
+      lib
+  )
+  therock_cmake_subproject_glob_c_sources(rdc
+    SUBDIRS .
+  )
+  therock_cmake_subproject_provide_package(rdc rdc lib/cmake/rdc)
+  therock_cmake_subproject_activate(rdc)
+
+  # RDC validation tests (runtime tests excluded - require GPU hardware)
+
+  therock_test_validate_shared_lib(
+    PATH rdc/stage/portable-rdc/lib
+    LIB_NAMES
+      librdc_bootstrap.so
+      librdc_client.so
+  )
+
+  # Verify gRPC symbols are hidden to prevent ODR violations
+  add_test(
+    NAME rdc-validate-no-grpc-symbol-pollution
+    COMMAND bash -c "! nm -D ${CMAKE_CURRENT_BINARY_DIR}/rdc/stage/portable-rdc/lib/librdc_client.so* 2>/dev/null | grep -q ' T.*grpc::'"
+  )
+  set_tests_properties(rdc-validate-no-grpc-symbol-pollution PROPERTIES
+    LABELS "rdc;symbol-visibility"
+  )
+
+  # Verify binaries exist
+  add_test(
+    NAME rdc-verify-binaries-exist
+    COMMAND bash -c "test -x ${CMAKE_CURRENT_BINARY_DIR}/rdc/stage/portable-rdc/bin/rdcd && test -x ${CMAKE_CURRENT_BINARY_DIR}/rdc/stage/portable-rdc/bin/rdci"
+  )
+  set_tests_properties(rdc-verify-binaries-exist PROPERTIES
+    LABELS "rdc;build-verification"
+  )
+
+  # Verify library SONAME structure
+  add_test(
+    NAME rdc-verify-library-sonames
+    COMMAND bash -c "test -L ${CMAKE_CURRENT_BINARY_DIR}/rdc/stage/portable-rdc/lib/librdc.so && test -L ${CMAKE_CURRENT_BINARY_DIR}/rdc/stage/portable-rdc/lib/librdc_client.so && test -L ${CMAKE_CURRENT_BINARY_DIR}/rdc/stage/portable-rdc/lib/librdc_bootstrap.so"
+  )
+  set_tests_properties(rdc-verify-library-sonames PROPERTIES
+    LABELS "rdc;build-verification"
+  )
+
+  # Provide RDC artifact (portable-rdc/ structure per RFC0007)
+  therock_provide_artifact(rdc
+    TARGET_NEUTRAL
+    DESCRIPTOR artifact-rdc.toml
+    COMPONENTS
+      dbg
+      dev
+      doc
+      lib
+      run
+      test
+    SUBPROJECT_DEPS
+      amdsmi
+      rocprofiler-sdk
+      rdc
+      ${THEROCK_BUNDLED_LIBCAP}
+      ${THEROCK_BUNDLED_ZLIB}
+  )
+
+endif(THEROCK_ENABLE_RDC)

dctools/artifact-rdc.toml

@@ -0,0 +1,30 @@
+# RDC (ROCm Data Center Tool)
+# Portable distribution with statically-linked gRPC per RFC0007
+
+[components.dbg."dctools/rdc/stage/portable-rdc"]
+
+[components.dev."dctools/rdc/stage/portable-rdc"]
+
+[components.doc."dctools/rdc/stage/portable-rdc"]
+include = [
+  "share/doc/**",
+]
+
+[components.lib."dctools/rdc/stage/portable-rdc"]
+include = [
+  "lib/*.so*",
+  "lib/rdc/*.so*",
+  "lib/rdc/hsaco/**/*.hsaco",
+]
+
+[components.run."dctools/rdc/stage/portable-rdc"]
+include = [
+  "bin/**",
+  "libexec/**",
+  "share/rdc/conf/**",
+]
+
+[components.test."dctools/rdc/stage/portable-rdc"]
+include = [
+  "share/rdc/rdctst_tests/**",
+]

docs/development/dependencies.md

@@ -26,6 +26,7 @@ project wide:
   bundling is not enabled or supported for the target OS):
   - `THEROCK_BUNDLED_BZIP2`
   - `THEROCK_BUNDLED_ELFUTILS`
+  - `THEROCK_BUNDLED_LIBCAP`
   - `THEROCK_BUNDLED_LIBDRM`
   - `THEROCK_BUNDLED_LIBLZMA`
   - `THEROCK_BUNDLED_NUMACTL`
@@ -68,6 +69,14 @@ Supported sub-libraries: `libelf`, `libdw`.
 - Import library: `libdw::libdw`
 - Alternatives: `pkg_check_modules(DW libdw)`
 
+## libcap
+
+Provides Linux capabilities for privileged operations (used by RDC).
+
+- Canonical method: `find_package(Libcap)`
+- Import library: `Libcap::Libcap`
+- Alternatives: `pkg_check_modules(LIBCAP libcap)` or direct linking (used by RDC)
+
 ## libdrm
 
 Supported sub-libraries: `libdrm`, `libdrm_amdgpu`

docs/rfcs/RFC0007-rdc-therock-integration.md

@@ -67,8 +67,10 @@ portable-rdc/
 │       ├── librdc_rocr.so      # 1MB (optional module)
 │       ├── librdc_rocp.so      # 500KB (optional module)
 │       └── librdc_rvs.so       # 200KB (optional module)
-└── etc/
-    └── rdc_options.conf
+└── share/
+    └── rdc/
+        └── conf/
+            └── rdc_options.conf  # User-level configuration
 ```
 
 **Static Linking Rationale:**

third-party/CMakeLists.txt

@@ -37,3 +37,7 @@ if(THEROCK_BUNDLE_SYSDEPS)
     add_subdirectory(sysdeps/windows)
   endif()
 endif()
+
+# gRPC: Static library for RDC (built on-demand when depended upon)
+# See: docs/rfcs/RFC0007-rdc-therock-integration.md
+add_subdirectory(grpc)

third-party/grpc/CMakeLists.txt

@@ -0,0 +1,56 @@
+# gRPC static library for RDC
+therock_subproject_fetch(therock-grpc-sources
+  CMAKE_PROJECT
+  # Originally mirrored from: https://github.com/grpc/grpc.git (git tag v1.67.1)
+  URL "https://rocm-third-party-deps.s3.us-east-2.amazonaws.com/grpc-v1.67.1.tar.gz"
+  URL_HASH "SHA256=ed0051771a1ed0ec6583be4ddb0fe9ee4473bc2f8a59d214ff49290ed5edb4ef"
+)
+
+set(_binary_dir "${CMAKE_CURRENT_BINARY_DIR}/build")
+
+therock_cmake_subproject_declare(therock-grpc
+  EXTERNAL_SOURCE_DIR "${CMAKE_CURRENT_BINARY_DIR}/source"
+  BINARY_DIR "${_binary_dir}"
+  NO_MERGE_COMPILE_COMMANDS
+  BACKGROUND_BUILD
+  OUTPUT_ON_FAILURE
+
+  CMAKE_ARGS
+    # Static linking configuration
+    -DBUILD_SHARED_LIBS=OFF
+    -DgRPC_BUILD_SHARED_LIBS=OFF
+    -DgRPC_PREFER_STATIC_LIBS=ON
+    -Dprotobuf_BUILD_SHARED_LIBS=OFF
+
+    # Dependency providers
+    -DgRPC_ZLIB_PROVIDER=package
+    -DgRPC_ABSL_PROVIDER=module
+    -DgRPC_CARES_PROVIDER=module
+    -DgRPC_RE2_PROVIDER=module
+    -DgRPC_SSL_PROVIDER=module
+    -DgRPC_PROTOBUF_PROVIDER=module
+
+    # Disable tests and unnecessary language plugins
+    -DgRPC_BUILD_TESTS=OFF
+    -DgRPC_BUILD_CSHARP_EXT=OFF
+    -DgRPC_BUILD_GRPC_CSHARP_PLUGIN=OFF
+    -DgRPC_BUILD_GRPC_NODE_PLUGIN=OFF
+    -DgRPC_BUILD_GRPC_OBJECTIVE_C_PLUGIN=OFF
+    -DgRPC_BUILD_GRPC_PHP_PLUGIN=OFF
+    -DgRPC_BUILD_GRPC_PYTHON_PLUGIN=OFF
+    -DgRPC_BUILD_GRPC_RUBY_PLUGIN=OFF
+
+    # Installation configuration
+    -DgRPC_INSTALL=ON
+    -DCMAKE_INSTALL_LIBDIR=lib
+    "-DCMAKE_INSTALL_RPATH=\\$ORIGIN:\\$ORIGIN/../lib"
+    -DCMAKE_BUILD_RPATH_USE_ORIGIN=ON
+    -DCMAKE_SKIP_BUILD_RPATH=OFF
+
+  RUNTIME_DEPS
+    therock-zlib
+)
+therock_cmake_subproject_provide_package(therock-grpc gRPC lib/cmake/grpc)
+therock_cmake_subproject_provide_package(therock-grpc protobuf lib/cmake/protobuf)
+therock_cmake_subproject_activate(therock-grpc)
+add_dependencies(therock-third-party therock-grpc)

third-party/grpc/fetch_grpc_tarball.sh

@@ -0,0 +1,69 @@
+#!/bin/bash
+set -euo pipefail
+
+# gRPC source tarball generation script
+
+GRPC_VERSION="${GRPC_VERSION:-v1.67.1}"
+GRPC_TAG="${GRPC_VERSION}"
+OUTPUT_DIR="${PWD}/tarballs"
+TEMP_DIR=$(mktemp -d)
+
+echo "===================================="
+echo "gRPC Tarball Generation"
+echo "===================================="
+echo "Version: ${GRPC_VERSION}"
+echo "Tag: ${GRPC_TAG}"
+echo "Output: ${OUTPUT_DIR}"
+echo "Temp: ${TEMP_DIR}"
+echo ""
+
+# Create output directory
+mkdir -p "${OUTPUT_DIR}"
+
+# Clone gRPC repository
+echo "Step 1: Cloning gRPC repository..."
+cd "${TEMP_DIR}"
+git clone --depth 1 --branch "${GRPC_TAG}" --recurse-submodules \
+  https://github.com/grpc/grpc.git grpc-${GRPC_VERSION}
+
+# Enter the cloned directory
+cd "grpc-${GRPC_VERSION}"
+
+# Ensure all submodules are initialized and updated
+echo ""
+echo "Step 2: Syncing submodules..."
+git submodule sync --recursive
+git submodule update --init --recursive --depth 1
+
+# Remove all .git directories and files
+echo ""
+echo "Step 3: Removing .git directories..."
+find . -name ".git" -exec rm -rf {} + 2>/dev/null || true
+find . -name ".gitignore" -delete
+find . -name ".gitmodules" -delete
+find . -name ".gitattributes" -delete
+
+# Return to temp directory
+cd "${TEMP_DIR}"
+
+# Create tarball
+echo ""
+echo "Step 4: Creating tarball..."
+TARBALL_NAME="grpc-${GRPC_VERSION}.tar.gz"
+tar -czf "${OUTPUT_DIR}/${TARBALL_NAME}" "grpc-${GRPC_VERSION}"
+
+# Calculate SHA256 hash
+echo ""
+echo "Step 5: Calculating SHA256 hash..."
+cd "${OUTPUT_DIR}"
+SHA256=$(sha256sum "${TARBALL_NAME}" | awk '{print $1}')
+
+# Display results
+echo ""
+echo "===================================="
+echo "Tarball created successfully!"
+echo "===================================="
+echo "File: ${OUTPUT_DIR}/${TARBALL_NAME}"
+echo "Size: $(du -h "${TARBALL_NAME}" | cut -f1)"
+echo "SHA256: ${SHA256}"
+echo ""