Removes default visibility in debug mode and updates unit tests for alt_rsmi impl (#2091)

* Update unit tests for alt_rsmi impl

- Create distinct test executable for alt_rsmi testing
- Updated alt_rsmi tests to use public methods
- Compiles alt_rsmi.cc with ARSMI_TEST_BUILD
- Enables external linkage of internal variables
- Only for AltRsmiTests.cpp that manipulates internals
- Clean separation for test behavior

* Address review comments

* restore hidden symbol visibility

[ROCm/rccl commit: 74690ea705]

projects/rccl/CMakeLists.txt

@@ -1095,6 +1095,7 @@ target_compile_options(rccl PRIVATE -Werror=deprecated-copy-with-user-provided-c
 target_compile_options(rccl PRIVATE -Wno-format-nonliteral)
 target_compile_options(rccl PRIVATE -Wno-unused-function)
 target_compile_options(rccl PRIVATE -fgpu-rdc)
+target_compile_options(rccl PRIVATE -fvisibility=hidden)
 
 if(QUIET_WARNINGS)
   target_compile_options(rccl PRIVATE -Wno-invalid-offsetof)
@@ -1118,8 +1119,7 @@ if(ENABLE_CODE_COVERAGE)
   message(STATUS "Code coverage is enabled with build type '${CMAKE_BUILD_TYPE}'.")
 
   target_compile_options(rccl PRIVATE
-    -fvisibility=default -Xarch_host -fprofile-instr-generate
-    -Xarch_host -fcoverage-mapping)
+    -Xarch_host -fprofile-instr-generate -Xarch_host -fcoverage-mapping)
 
   set(COVERAGE_SHARED_LINKER_FLAGS
     -fprofile-generate
@@ -1133,14 +1133,6 @@ if(ENABLE_CODE_COVERAGE)
 
   target_link_options(rccl PRIVATE ${COVERAGE_SHARED_LINKER_FLAGS})
   target_link_options(rccl PRIVATE ${COVERAGE_EXE_LINKER_FLAGS})
-elseif(BUILD_TESTS) # Enable default/hidden visibility based on build type and ROCM_VERSION
-  if (ROCM_VERSION VERSION_GREATER_EQUAL "60400" AND CMAKE_BUILD_TYPE MATCHES "Debug")
-    target_compile_options(rccl PRIVATE -fvisibility=default)
-  else()
-    target_compile_options(rccl PRIVATE -fvisibility=hidden)
-  endif()
-else()  # Enable hidden visibility for library without tests/code coverage enabled
-  target_compile_options(rccl PRIVATE -fvisibility=hidden)
 endif()
 
 if (HAVE_KERNARG_PRELOAD)

projects/rccl/src/misc/alt_rsmi.cc

@@ -48,18 +48,29 @@ struct ARSMI_systemNode {
     std::string s_card;
 };
 
-static const char *kKFDNodesPathRoot = "/sys/class/kfd/kfd/topology/nodes";
+// Internal implementation details
+// When compiled into the test binary (ARSMI_TEST_BUILD defined), these have external linkage
+// so AltRsmiTestUtils.cpp can access them via extern declarations.
+// When compiled into the production library (ARSMI_TEST_BUILD not defined), these are static.
+#ifdef ARSMI_TEST_BUILD
+  // Test build: external linkage for cross-TU access
+  #define ARSMI_LOCAL thread_local
+#else
+  // Production build: static/internal linkage
+  #define ARSMI_LOCAL static thread_local
+#endif
+
+ARSMI_LOCAL const char *kKFDNodesPathRoot = "/sys/class/kfd/kfd/topology/nodes";
 static const uint32_t kAmdGpuId = 0x1002;
 
 // Vector containing data about each node, ordered by bdf ID
-static thread_local std::vector<ARSMI_systemNode> ARSMI_orderedNodes;
+ARSMI_LOCAL std::vector<ARSMI_systemNode> ARSMI_orderedNodes;
 
 // 2-D matrix with link information between each pair of nodes.
-static thread_local std::vector<std::vector<ARSMI_linkInfo>> ARSMI_orderedLinks;
+ARSMI_LOCAL std::vector<std::vector<ARSMI_linkInfo>> ARSMI_orderedLinks;
 
 // Number of devices recognized
-static thread_local int ARSMI_num_devices=-1;
-
+ARSMI_LOCAL int ARSMI_num_devices=-1;
 
 // Public API functions
 int ARSMI_init(void)

projects/rccl/test/CMakeLists.txt

@@ -236,6 +236,25 @@ if(BUILD_TESTS)
       list(APPEND RCCL_TEST_EXECUTABLES rccl-UnitTestsMPI)
 
     endif()
+
+    # rccl-UnitTestsAltRsmi: Uses TEST BUILD alt_rsmi.cc (ARSMI_TEST_BUILD)
+    # This separate executable compiles alt_rsmi.cc with ARSMI_TEST_BUILD,
+    # enabling external linkage of internal variables so that
+    # tests can access and manipulate them for testing.
+    list(APPEND RCCL_TEST_EXECUTABLES rccl-UnitTestsAltRsmi)
+
+    set(TEST_ALTRSMI_SOURCE_FILES
+      AltRsmiTests.cpp
+      ../src/misc/alt_rsmi.cc
+      common/main_fixtures.cpp
+      common/EnvVars.cpp
+      common/ProcessIsolatedTestRunner.cpp
+    )
+
+    add_executable(rccl-UnitTestsAltRsmi ${TEST_ALTRSMI_SOURCE_FILES})
+
+    # Define ARSMI_TEST_BUILD specifically for rccl-UnitTestsAltRsmi
+    target_compile_definitions(rccl-UnitTestsAltRsmi PRIVATE ARSMI_TEST_BUILD)
   endif()
 
   foreach(test_executable IN LISTS RCCL_TEST_EXECUTABLES)