[rocprofiler-systems] AMD SMI collector using policy-based design

projects/rocprofiler-systems/source/lib/common/CMakeLists.txt

@@ -26,6 +26,7 @@ target_sources(
         ${CMAKE_CURRENT_SOURCE_DIR}/invoke.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/join.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/setup.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/span.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/traits.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/md5sum.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/static_object.hpp

projects/rocprofiler-systems/source/lib/core/agent_manager.cpp

@@ -47,8 +47,8 @@ agent_manager::insert_agent(agent& _agent)
     _agent.device_type_index = _agent_counts[_agent.type]++;
     _agents.emplace_back(std::make_shared<agent>(_agent));
 
-    LOG_TRACE("Inserting agent with device handle: {}, and agent id: {}, device type: {}",
-              _agent.device_id, _agent.device_type_index, to_string(_agent.type));
+    LOG_INFO("Inserting agent with device handle: {}, and agent id: {}, device type: {}",
+             _agent.device_id, _agent.device_type_index, to_string(_agent.type));
 }
 
 const agent&

projects/rocprofiler-systems/source/lib/core/config.cpp

@@ -2555,6 +2555,12 @@ get_caching_perfetto()
     return _v;
 }
 
+bool
+get_use_cache_output()
+{
+    return get_use_rocpd() || get_caching_perfetto();
+}
+
 int
 get_kill_delay()
 {

projects/rocprofiler-systems/source/lib/core/config.hpp

@@ -385,6 +385,9 @@ get_use_rocpd() ROCPROFSYS_HOT;
 bool&
 get_caching_perfetto() ROCPROFSYS_HOT;
 
+bool
+get_use_cache_output() ROCPROFSYS_HOT;
+
 bool
 get_merge_perfetto_files();
 

projects/rocprofiler-systems/source/lib/core/trace_cache/cache_manager.hpp

@@ -26,9 +26,9 @@
 #include "core/trace_cache/metadata_registry.hpp"
 #include "core/trace_cache/sample_type.hpp"
 #include "core/trace_cache/storage_parser.hpp"
-
+#include "library/pmc/collectors/gpu/sample.hpp"
+#include "library/pmc/collectors/nic/sample.hpp"
 #include "library/runtime.hpp"
-
 #include <memory>
 #include <unistd.h>
 
@@ -40,8 +40,9 @@ namespace trace_cache
 using storage_parser_t =
     storage_parser<type_identifier_t, kernel_dispatch_sample, memory_copy_sample,
                    memory_allocate_sample, region_sample, in_time_sample,
-                   pmc_event_with_sample, amd_smi_sample, cpu_freq_sample,
-                   backtrace_region_sample, scratch_memory_sample, ainic_sample>;
+                   pmc_event_with_sample, pmc::collectors::gpu::sample,
+                   pmc::collectors::nic::sample, cpu_freq_sample, backtrace_region_sample,
+                   scratch_memory_sample>;
 
 using buffer_storage_t = buffer_storage<flush_worker_factory_t, type_identifier_t>;
 

projects/rocprofiler-systems/source/lib/core/trace_cache/cache_type_traits.hpp

@@ -22,6 +22,7 @@
 
 #pragma once
 #include "common/span.hpp"
+#include <array>
 #include <cstdint>
 #include <optional>
 #include <string_view>
@@ -104,6 +105,17 @@ struct is_vector<std::vector<T>> : std::true_type
 template <typename T>
 inline constexpr bool is_vector_v = is_vector<T>::value;
 
+template <typename T>
+struct is_array : std::false_type
+{};
+
+template <typename T, size_t N>
+struct is_array<std::array<T, N>> : std::true_type
+{};
+
+template <typename T>
+inline constexpr bool is_array_v = is_array<T>::value;
+
 template <typename T>
 static constexpr bool is_string_view_v =
     std::is_same_v<std::decay_t<T>, std::string_view>;
@@ -122,7 +134,7 @@ inline constexpr bool is_optional_v = is_optional<T>::value;
 template <typename T>
 inline constexpr bool is_supported_type_v =
     is_span_v<T> || std::is_integral_v<T> || std::is_floating_point_v<T> ||
-    is_string_view_v<T> || is_vector_v<T> || is_optional_v<T>;
+    is_string_view_v<T> || is_vector_v<T> || is_optional_v<T> || is_array_v<T>;
 
 template <typename T>
 struct is_enum_class

projects/rocprofiler-systems/source/lib/core/trace_cache/metadata_registry.hpp

@@ -37,7 +37,7 @@
 #include <rocprofiler-sdk/callback_tracing.h>
 #include <rocprofiler-sdk/cxx/name_info.hpp>
 #include <set>
-#include <sstream>
+#include <spdlog/fmt/ranges.h>
 #include <stdint.h>
 #include <string.h>
 #include <string>
@@ -62,18 +62,6 @@ struct process
     uint32_t    end;
 };
 
-template <typename Category>
-inline std::string
-annotate_category(std::optional<int> first_section  = std::nullopt,
-                  std::optional<int> second_section = std::nullopt)
-{
-    std::stringstream ss;
-    ss << std::string(tim::trait::name<Category>::value);
-    if(first_section) ss << "_" << std::to_string(*first_section);
-    if(second_section) ss << "_" << std::to_string(*second_section);
-    return ss.str();
-}
-
 struct pmc
 {
     agent_type  type;
@@ -131,16 +119,12 @@ struct thread
 
 template <typename Category>
 inline std::string
-annotate_with_device_id(uint32_t           device_id,
-                        std::optional<int> first_section  = std::nullopt,
-                        std::optional<int> second_section = std::nullopt)
+format_track_name(std::optional<int> first_section  = std::nullopt,
+                  std::optional<int> second_section = std::nullopt)
 {
-    std::stringstream ss;
-    ss << std::string(tim::trait::name<Category>::value) + " [" +
-              std::to_string(device_id) + "]";
-    if(first_section) ss << "_" << std::to_string(*first_section);
-    if(second_section) ss << "_" << std::to_string(*second_section);
-    return ss.str();
+    return fmt::format("{}{}{}", tim::trait::name<Category>::value,
+                       first_section ? fmt::format("_{}", *first_section) : "",
+                       second_section ? fmt::format("_{}", *second_section) : "");
 }
 
 template <typename Category>