[SDK] Update UUID (rocprofiler_uuid_t) (#390)

* changing uuid abi * fix * review comments * fix CI fail * review comments * fix * adding static asserts * making constructor constexpr * fix CI fail * upadate UUID length to 16 bytes * fixing value64 Co-authored-by: Madsen, Jonathan <Jonathan.Madsen@amd.com> * Update CHANGELOG.md Co-authored-by: Madsen, Jonathan <Jonathan.Madsen@amd.com> --------- Co-authored-by: Madsen, Jonathan <Jonathan.Madsen@amd.com> [ROCm/rocprofiler-sdk commit: fcfb5bd918]
2025-07-18 11:24:48 -05:00
@@ -198,6 +198,7 @@ Full documentation for ROCprofiler-SDK is available at [rocm.docs.amd.com/projec

 - SDK no longer creates a background thread when every tool returns a nullptr from `rocprofiler_configure`.
 - Updated disassembly.hpp's vaddr-to-file-offset mapping to use the dedicated comgr API.
+- rocprofiler_uuid_t ABI is changed to hold 128 bit value.
 - rocprofv3 shorthand argument for `--collection-period` is now `-P` (upper-case) as `-p` (lower-case) is reserved for later use
 - default output format for rocprofv3 is now `rocpd` (SQLite3 database)
 - rocprofv3 avail tool renamed from rocprofv3_avail to rocprofv3-avail tool
@@ -1132,7 +1132,7 @@ template <typename ArchiveT>
 void
 save(ArchiveT& ar, rocprofiler_uuid_t data)
 {
-    ROCP_SDK_SAVE_DATA_FIELD(value);
+    ROCP_SDK_SAVE_DATA_FIELD(bytes);
 }

 template <typename ArchiveT>
@@ -572,10 +572,9 @@ typedef union rocprofiler_address_t
 * @brief Stores UUID for devices.
 *
 */
-typedef union rocprofiler_uuid_t
+typedef struct rocprofiler_uuid_t
 {
-    uint64_t value;  ///< numerical value
-    void*    bytes;  ///< uuid in hexadecimal
+    uint8_t bytes[16];  // numerical value
 } rocprofiler_uuid_t;

 //--------------------------------------------------------------------------------------//
@@ -503,7 +503,8 @@ update_agent_runtime_visibility(rocprofiler_agent_t& agent_info)
                {
                    auto _uuid =
                        std::strtoull(itr.substr(uuid_prefix.length()).c_str(), nullptr, 16);
-                    if(_uuid == agent_info.uuid.value) return parse_result{true, _idx_v};
+                    auto uuid_view = uuid_view_t{agent_info.uuid};
+                    if(_uuid == uuid_view.value64[0]) return parse_result{true, _idx_v};
                }
                else
                {
@@ -720,7 +721,7 @@ read_topology()
        agent_info.name         = "";
        agent_info.product_name = "";
        agent_info.vendor_name  = "";
-        agent_info.uuid         = {.value = 0};
+        memset(&agent_info.uuid.bytes, 0, sizeof(agent_info.uuid.bytes));
        if(agent_info.type == ROCPROFILER_AGENT_TYPE_GPU)
        {
            constexpr auto workgrp_max = 1024;
@@ -729,7 +730,10 @@ read_topology()
            constexpr auto grid_max_y  = std::numeric_limits<uint16_t>::max();
            constexpr auto grid_max_z  = std::numeric_limits<uint16_t>::max();

-            read_property(properties, "unique_id", agent_info.uuid.value);
+            auto     _uuid    = uuid_view_t{};
+            uint64_t uuid_val = 0;
+            read_property(properties, "unique_id", uuid_val);
+            _uuid.value64[0] = uuid_val;
            read_property(
                properties, "max_engine_clk_fcompute", agent_info.max_engine_clk_fcompute);
            read_property(properties, "local_mem_size", agent_info.local_mem_size);
@@ -744,6 +748,7 @@ read_topology()
            agent_info.grid_max_dim       = {grid_max_x, grid_max_y, grid_max_z};
            agent_info.cu_count           = agent_info.simd_count / agent_info.simd_per_cu;

+            agent_info.uuid = static_cast<rocprofiler_uuid_t>(_uuid);
            if(int drm_fd = 0; (drm_fd = drmOpenRender(agent_info.drm_render_minor)) >= 0)
            {
                uint32_t major_version = 0;
@@ -37,6 +37,45 @@ namespace rocprofiler
 {
 namespace agent
 {
+struct uuid_view_t
+{
+    union
+    {
+        uint8_t  bytes[16];   ///< raw bytes
+        uint64_t value64[2];  /// view as 64 bit chunks
+    };
+
+    constexpr uuid_view_t()
+    : bytes()
+    {
+        for(uint8_t& byte : bytes)
+        {
+            byte = 0;
+        }
+    }
+
+    explicit uuid_view_t(rocprofiler_uuid_t _uuid)
+    {
+        static_assert(sizeof(bytes) == sizeof(_uuid.bytes));
+        for(size_t i = 0; i < sizeof(bytes); i++)
+        {
+            bytes[i] = _uuid.bytes[i];
+        }
+    }
+
+    explicit constexpr operator rocprofiler_uuid_t() const
+    {
+        auto _uuid = rocprofiler_uuid_t{};
+        static_assert(std::is_same_v<std::remove_extent_t<decltype(bytes)>,
+                                     std::remove_extent_t<decltype(_uuid.bytes)>> == true);
+        for(size_t i = 0; i < sizeof(bytes); i++)
+        {
+            _uuid.bytes[i] = bytes[i];
+        }
+        return _uuid;
+    }
+};
+
 std::vector<const rocprofiler_agent_t*>
 get_agents();

@@ -142,7 +142,7 @@ constexpr rocprofiler_agent_t default_agent =
                        .logical_node_id            = 0,
                        .logical_node_type_id       = 0,
                        .runtime_visibility         = {0, 0, 0, 0, 0},
-                        .uuid                       = {.value = 0}};
+                        .uuid = static_cast<rocprofiler_uuid_t>(agent::uuid_view_t{})};
 }  // namespace

 void
@@ -111,7 +111,7 @@ TEST(rocprofiler_lib, agent_abi)
    EXPECT_EQ(offsetof(rocprofiler_agent_t, uuid), 296) << msg;
    // Add test for offset of new field above this. Do NOT change any existing values!

-    constexpr auto expected_rocp_agent_size = 304;
+    constexpr auto expected_rocp_agent_size = 312;
    // If a new field is added, increase this value by the size of the new field(s)
    EXPECT_EQ(sizeof(rocprofiler_agent_t), expected_rocp_agent_size)
        << "ABI break. If you added a new field, make sure that this is the only new check that "
@@ -450,7 +450,9 @@ TEST(rocprofiler_lib, agent_visibility_multigpu)
        for(const auto* itr : _agents)
        {
            ordinals.emplace(itr->id, itr->logical_node_type_id);
-            uuids.emplace(itr->id, fmt::format("GPU-{:X}", itr->uuid.value));
+            auto uuid_view  = rocprofiler::agent::uuid_view_t{itr->uuid};
+            auto uuid_value = uuid_view.value64[0];
+            uuids.emplace(itr->id, fmt::format("GPU-{:X}", uuid_value));
            ROCP_WARNING << ordinals.at(itr->id) << " :: " << uuids.at(itr->id);

            all_ordinals = fmt::format("{},{}", all_ordinals, ordinals.at(itr->id));
@@ -637,7 +639,10 @@ TEST(rocprofiler_lib, agent_visibility_inverted_multigpu)

        for(const auto* itr : _agents)
        {
-            auto _uuid = fmt::format("GPU-{:X}", itr->uuid.value);
+            auto uuid_view  = rocprofiler::agent::uuid_view_t{itr->uuid};
+            auto uuid_value = uuid_view.value64[0];
+            auto _uuid      = fmt::format("GPU-{:X}", uuid_value);
+
            if(ordinals.empty())
            {
                devices_reversed_with_ordinal_0_uuid = fmt::format("1,{}", _uuid);
@@ -761,7 +766,9 @@ TEST(rocprofiler_lib, agent_visibility_inverted_multigpu)

    for(const auto* itr : get_gpu_agents())
    {
-        const auto curr_uuid = fmt::format("GPU-{:X}", itr->uuid.value);
+        auto       uuid_view  = rocprofiler::agent::uuid_view_t{itr->uuid};
+        auto       uuid_value = uuid_view.value64[0];
+        const auto curr_uuid  = fmt::format("GPU-{:X}", uuid_value);
        if(ordinals.at(itr->id) == 0)
        {
            ASSERT_EQ(itr->runtime_visibility.hsa, 1) << "agent-" << itr->node_id;