rocDecode API Tracing Support (#49)

* rocDecode API Tracing support

* Test bin file added to rocdecode. Need to add validate python methods

* Added option to not make rocDecode tests

* Added rocdecode and rocprofv3 tests

* Added csv test

* Address PR comments. Changed tests to use built-in rocstreambit decoder to remove ffmpeg dependancy. Changed cmake option to disbale tests rather than not build them. Tests work locally, but will fail until rocDecode is built with tracing enabled on CI

* Add option to avoid building rocdecode tests

* Added option to avoid building rocdecode bin file

* Merge conflict error

* CMake files changed in response to review comments. Attempting to implement callbacks.

* Turned off test building for rocdecode

* Minor fixes for review comments

* Review comments

* Updated formatting

* Document changes and format.hpp reversion. Need to remove iterate args support for now for later update.

* Remove iterate args support

* Remove iterate-args

* enforce abi versioning in macro if

* Fix doc error

* removed spaces to fix indentation error

---------

Co-authored-by: Madsen, Jonathan <Jonathan.Madsen@amd.com>

source/lib/output/buffered_output.hpp

@@ -163,5 +163,7 @@ using counter_records_buffered_output_t =
 using pc_sampling_host_trap_buffered_output_t =
     buffered_output<rocprofiler::tool::rocprofiler_tool_pc_sampling_host_trap_record_t,
                     domain_type::PC_SAMPLING_HOST_TRAP>;
+using rocdecode_buffered_output_t =
+    buffered_output<rocprofiler_buffer_tracing_rocdecode_api_record_t, domain_type::ROCDECODE>;
 }  // namespace tool
 }  // namespace rocprofiler

source/lib/output/domain_type.cpp

@@ -61,6 +61,7 @@ DEFINE_BUFFER_TYPE_NAME(PC_SAMPLING_HOST_TRAP,
                         "PC_SAMPLING_HOST_TRAP",
                         "pc_sampling_host_trap",
                         "pc_sampling_host_trap_stats")
+DEFINE_BUFFER_TYPE_NAME(ROCDECODE, "ROCDECODE_API", "rocdecode_api_trace", "rocdecode_api_stats")
 
 #undef DEFINE_BUFFER_TYPE_NAME
 

source/lib/output/domain_type.hpp

@@ -37,6 +37,7 @@ enum class domain_type
     MEMORY_ALLOCATION,
     COUNTER_VALUES,
     PC_SAMPLING_HOST_TRAP,
+    ROCDECODE,
     LAST,
 };
 

source/lib/output/generateCSV.cpp

@@ -722,6 +722,48 @@ generate_csv(const output_config&                                           cfg,
     }
 }
 
+void
+generate_csv(const output_config&                                                cfg,
+             const metadata&                                                     tool_metadata,
+             const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& data,
+             const stats_entry_t&                                                stats)
+{
+    if(data.empty()) return;
+
+    if(cfg.stats && stats)
+        write_stats(get_stats_output_file(cfg, domain_type::ROCDECODE), stats.entries);
+
+    auto ofs = tool::csv_output_file{cfg,
+                                     domain_type::ROCDECODE,
+                                     tool::csv::api_csv_encoder{},
+                                     {"Domain",
+                                      "Function",
+                                      "Process_Id",
+                                      "Thread_Id",
+                                      "Correlation_Id",
+                                      "Start_Timestamp",
+                                      "End_Timestamp"}};
+    for(auto ditr : data)
+    {
+        for(auto record : data.get(ditr))
+        {
+            auto row_ss   = std::stringstream{};
+            auto api_name = tool_metadata.get_operation_name(record.kind, record.operation);
+            rocprofiler::tool::csv::api_csv_encoder::write_row(
+                row_ss,
+                tool_metadata.get_kind_name(record.kind),
+                api_name,
+                tool_metadata.process_id,
+                record.thread_id,
+                record.correlation_id.internal,
+                record.start_timestamp,
+                record.end_timestamp);
+
+            ofs << row_ss.str();
+        }
+    }
+}
+
 void
 generate_csv(const output_config&                                              cfg,
              const metadata&                                                   tool_metadata,

source/lib/output/generateCSV.hpp

@@ -87,6 +87,12 @@ generate_csv(const output_config&                                           cfg,
              const generator<rocprofiler_buffer_tracing_rccl_api_record_t>& data,
              const stats_entry_t&                                           stats);
 
+void
+generate_csv(const output_config&                                                cfg,
+             const metadata&                                                     tool_metadata,
+             const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& data,
+             const stats_entry_t&                                                stats);
+
 void
 generate_csv(const output_config&                                                    cfg,
              const metadata&                                                         tool_metadata,

source/lib/output/generateJSON.cpp

@@ -187,7 +187,8 @@ write_json(json_output& json_ar,
            generator<rocprofiler_buffer_tracing_scratch_memory_record_t>    scratch_memory_gen,
            generator<rocprofiler_buffer_tracing_rccl_api_record_t>          rccl_api_gen,
            generator<rocprofiler_buffer_tracing_memory_allocation_record_t> memory_allocation_gen,
-           generator<rocprofiler_tool_pc_sampling_host_trap_record_t>       pc_sampling_gen)
+           generator<rocprofiler_tool_pc_sampling_host_trap_record_t>       pc_sampling_gen,
+           generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>     rocdecode_api_gen)
 
 {
     // summary
@@ -229,6 +230,7 @@ write_json(json_output& json_ar,
         json_ar(cereal::make_nvp("memory_allocation", memory_allocation_gen));
         json_ar(cereal::make_nvp("scratch_memory", scratch_memory_gen));
         json_ar(cereal::make_nvp("pc_sample_host_trap", pc_sampling_gen));
+        json_ar(cereal::make_nvp("rocdecode_api", rocdecode_api_gen));
         json_ar.finishNode();
     }
 }

source/lib/output/generateJSON.hpp

@@ -94,6 +94,8 @@ write_json(json_output&                                                     json
            generator<rocprofiler_buffer_tracing_scratch_memory_record_t>    scratch_memory_gen,
            generator<rocprofiler_buffer_tracing_rccl_api_record_t>          rccl_api_gen,
            generator<rocprofiler_buffer_tracing_memory_allocation_record_t> memory_allocation_gen,
-           generator<rocprofiler_tool_pc_sampling_host_trap_record_t>       pc_sampling_gen);
+           generator<rocprofiler_tool_pc_sampling_host_trap_record_t>       pc_sampling_gen,
+           generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>     rocdecode_api_gen);
+
 }  // namespace tool
 }  // namespace rocprofiler

source/lib/output/generateOTF2.cpp

@@ -367,7 +367,8 @@ write_otf2(
     std::deque<rocprofiler_buffer_tracing_marker_api_record_t>*      marker_api_data,
     std::deque<rocprofiler_buffer_tracing_scratch_memory_record_t>* /*scratch_memory_data*/,
     std::deque<rocprofiler_buffer_tracing_rccl_api_record_t>*          rccl_api_data,
-    std::deque<rocprofiler_buffer_tracing_memory_allocation_record_t>* memory_allocation_data)
+    std::deque<rocprofiler_buffer_tracing_memory_allocation_record_t>* memory_allocation_data,
+    std::deque<rocprofiler_buffer_tracing_rocdecode_api_record_t>*     rocdecode_api_data)
 {
     namespace sdk = ::rocprofiler::sdk;
 
@@ -418,6 +419,8 @@ write_otf2(
             tids.emplace(itr.thread_id);
         for(auto itr : *rccl_api_data)
             tids.emplace(itr.thread_id);
+        for(auto itr : *rocdecode_api_data)
+            tids.emplace(itr.thread_id);
 
         for(auto itr : *memory_copy_data)
         {
@@ -614,6 +617,7 @@ write_otf2(
         add_event_data(hip_api_data, sdk::category::hip_api{});
         add_event_data(marker_api_data, sdk::category::marker_api{});
         add_event_data(rccl_api_data, sdk::category::rccl_api{});
+        add_event_data(rocdecode_api_data, sdk::category::rocdecode_api{});
     }
 
     for(auto itr : *memory_copy_data)

source/lib/output/generateOTF2.hpp

@@ -46,6 +46,7 @@ write_otf2(
     std::deque<rocprofiler_buffer_tracing_marker_api_record_t>*        marker_api_data,
     std::deque<rocprofiler_buffer_tracing_scratch_memory_record_t>*    scratch_memory_data,
     std::deque<rocprofiler_buffer_tracing_rccl_api_record_t>*          rccl_api_data,
-    std::deque<rocprofiler_buffer_tracing_memory_allocation_record_t>* memory_allocation_data);
+    std::deque<rocprofiler_buffer_tracing_memory_allocation_record_t>* memory_allocation_data,
+    std::deque<rocprofiler_buffer_tracing_rocdecode_api_record_t>*     rocdecode_api_data);
 }  // namespace tool
 }  // namespace rocprofiler

source/lib/output/generatePerfetto.cpp

@@ -72,7 +72,8 @@ write_perfetto(
     const generator<rocprofiler_buffer_tracing_marker_api_record_t>&      marker_api_gen,
     const generator<rocprofiler_buffer_tracing_scratch_memory_record_t>& /*scratch_memory_gen*/,
     const generator<rocprofiler_buffer_tracing_rccl_api_record_t>&          rccl_api_gen,
-    const generator<rocprofiler_buffer_tracing_memory_allocation_record_t>& memory_allocation_gen)
+    const generator<rocprofiler_buffer_tracing_memory_allocation_record_t>& memory_allocation_gen,
+    const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>&     rocdecode_api_gen)
 {
     namespace sdk = ::rocprofiler::sdk;
 
@@ -168,6 +169,9 @@ write_perfetto(
         for(auto ditr : rccl_api_gen)
             for(auto itr : rccl_api_gen.get(ditr))
                 tids.emplace(itr.thread_id);
+        for(auto ditr : rocdecode_api_gen)
+            for(auto itr : rocdecode_api_gen.get(ditr))
+                tids.emplace(itr.thread_id);
 
         for(auto ditr : memory_copy_gen)
             for(auto itr : memory_copy_gen.get(ditr))
@@ -399,6 +403,37 @@ write_perfetto(
                 tracing_session->FlushBlocking();
             }
 
+        for(auto ditr : rocdecode_api_gen)
+            for(auto itr : rocdecode_api_gen.get(ditr))
+            {
+                auto  name  = buffer_names.at(itr.kind, itr.operation);
+                auto& track = thread_tracks.at(itr.thread_id);
+
+                TRACE_EVENT_BEGIN(sdk::perfetto_category<sdk::category::rocdecode_api>::name,
+                                  ::perfetto::StaticString(name.data()),
+                                  track,
+                                  itr.start_timestamp,
+                                  ::perfetto::Flow::ProcessScoped(itr.correlation_id.internal),
+                                  "begin_ns",
+                                  itr.start_timestamp,
+                                  "end_ns",
+                                  itr.end_timestamp,
+                                  "delta_ns",
+                                  (itr.end_timestamp - itr.start_timestamp),
+                                  "tid",
+                                  itr.thread_id,
+                                  "kind",
+                                  itr.kind,
+                                  "operation",
+                                  itr.operation,
+                                  "corr_id",
+                                  itr.correlation_id.internal);
+                TRACE_EVENT_END(sdk::perfetto_category<sdk::category::rocdecode_api>::name,
+                                track,
+                                itr.end_timestamp);
+                tracing_session->FlushBlocking();
+            }
+
         for(auto ditr : memory_copy_gen)
             for(auto itr : memory_copy_gen.get(ditr))
             {

source/lib/output/generatePerfetto.hpp

@@ -46,6 +46,7 @@ write_perfetto(
     const generator<rocprofiler_buffer_tracing_marker_api_record_t>&        marker_api_gen,
     const generator<rocprofiler_buffer_tracing_scratch_memory_record_t>&    scratch_memory_gen,
     const generator<rocprofiler_buffer_tracing_rccl_api_record_t>&          rccl_api_gen,
-    const generator<rocprofiler_buffer_tracing_memory_allocation_record_t>& memory_allocation_gen);
+    const generator<rocprofiler_buffer_tracing_memory_allocation_record_t>& memory_allocation_gen,
+    const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>&     rocdecode_api_gen);
 }  // namespace tool
 }  // namespace rocprofiler

source/lib/output/generateStats.cpp

@@ -228,6 +228,24 @@ generate_stats(const output_config& /*cfg*/,
     return get_stats(rccl_stats);
 }
 
+stats_entry_t
+generate_stats(const output_config& /*cfg*/,
+               const metadata&                                                     tool_metadata,
+               const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& data)
+{
+    auto rocdecode_stats = stats_map_t{};
+    for(auto ditr : data)
+    {
+        for(auto record : data.get(ditr))
+        {
+            auto api_name = tool_metadata.get_operation_name(record.kind, record.operation);
+            rocdecode_stats[api_name] += (record.end_timestamp - record.start_timestamp);
+        }
+    }
+
+    return get_stats(rocdecode_stats);
+}
+
 namespace
 {
 void

source/lib/output/generateStats.hpp

@@ -75,6 +75,11 @@ generate_stats(const output_config& cfg,
                const metadata&      tool_metadata,
                const generator<rocprofiler_buffer_tracing_memory_allocation_record_t>& data);
 
+stats_entry_t
+generate_stats(const output_config&                                                cfg,
+               const metadata&                                                     tool_metadata,
+               const generator<rocprofiler_buffer_tracing_rocdecode_api_record_t>& data);
+
 stats_entry_t
 generate_stats(const output_config&                                              cfg,
                const metadata&                                                   tool_metadata,