From 5467c831881ca9476af9819a31a398a0ac9351a1 Mon Sep 17 00:00:00 2001 From: "Trowbridge, Ian" Date: Fri, 11 Apr 2025 16:56:36 -0500 Subject: [PATCH] rocDecode Buffer Tracing Support (#315) * Added buffer tracing support for rocdecode and updated tests to work with buffer tracing * Updated perfetto to output args individually rather than as a string list * Updated docstrings and operation type, changed OTF2 code to remove warning due to change in operation type * Updated tests for review comments * Test args exist and return value * Updated to use string entry * Change function name * Updated PR to reflect review comments * Updated for PR review comments * Change function name [ROCm/rocprofiler-sdk commit: 077723337a0cb28497215e1e46a7172715f49ac3] --- .../source/docs/how-to/using-rocprofv3.rst | 2 + .../include/rocprofiler-sdk/buffer_tracing.h | 30 ++- .../rocprofiler-sdk/cxx/serialization.hpp | 12 +- .../source/include/rocprofiler-sdk/fwd.h | 4 + .../rocprofiler-sdk/rocdecode/api_args.h | 1 + .../source/lib/output/buffered_output.hpp | 2 +- .../source/lib/output/generateCSV.cpp | 8 +- .../source/lib/output/generateCSV.hpp | 8 +- .../source/lib/output/generateJSON.cpp | 2 +- .../source/lib/output/generateJSON.hpp | 2 +- .../source/lib/output/generateOTF2.cpp | 25 ++- .../source/lib/output/generateOTF2.hpp | 2 +- .../source/lib/output/generatePerfetto.cpp | 16 +- .../source/lib/output/generatePerfetto.hpp | 2 +- .../source/lib/output/generateStats.cpp | 4 +- .../source/lib/output/generateStats.hpp | 6 +- .../source/lib/rocprofiler-sdk-tool/tool.cpp | 18 +- .../lib/rocprofiler-sdk/buffer_tracing.cpp | 11 + .../lib/rocprofiler-sdk/callback_tracing.cpp | 12 +- .../lib/rocprofiler-sdk/rocdecode/defines.hpp | 18 +- .../rocprofiler-sdk/rocdecode/rocdecode.cpp | 189 +++++++++++++----- .../rocdecode/rocdecode.def.cpp | 20 +- .../rocprofiler-sdk/rocdecode/rocdecode.hpp | 17 +- .../tests/rocdecode/validate.py | 30 ++- .../rocprofv3/rocdecode-trace/CMakeLists.txt | 12 +- .../rocprofv3/rocdecode-trace/conftest.py | 8 +- .../rocprofv3/rocdecode-trace/validate.py | 39 ++-- .../rocprofiler-sdk/tests/tools/json-tool.cpp | 37 +++- 28 files changed, 391 insertions(+), 146 deletions(-) diff --git a/projects/rocprofiler-sdk/source/docs/how-to/using-rocprofv3.rst b/projects/rocprofiler-sdk/source/docs/how-to/using-rocprofv3.rst index ff0c3268bd..3912840c95 100644 --- a/projects/rocprofiler-sdk/source/docs/how-to/using-rocprofv3.rst +++ b/projects/rocprofiler-sdk/source/docs/how-to/using-rocprofv3.rst @@ -503,6 +503,8 @@ Here are the contents of ``rocdecode_api_trace.csv`` file: :widths: 10,10,10,10,10,20,20 :header-rows: 1 +Perfetto will also show rocDeocde API arguments. Pointers will not be dereferenced and only the address will be displayed. + rocJPEG trace +++++++++++++++ diff --git a/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/buffer_tracing.h b/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/buffer_tracing.h index 5f6ccde9ec..24da823d8f 100644 --- a/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/buffer_tracing.h +++ b/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/buffer_tracing.h @@ -27,6 +27,8 @@ #include #include #include +#include +#include #include @@ -213,7 +215,7 @@ typedef struct rocprofiler_buffer_tracing_rocdecode_api_record_t { uint64_t size; ///< size of this struct rocprofiler_buffer_tracing_kind_t kind; - rocprofiler_tracing_operation_t operation; + rocprofiler_rocdecode_api_id_t operation; rocprofiler_correlation_id_t correlation_id; ///< correlation ids for record rocprofiler_timestamp_t start_timestamp; ///< start time in nanoseconds rocprofiler_timestamp_t end_timestamp; ///< end time in nanoseconds @@ -225,6 +227,29 @@ typedef struct rocprofiler_buffer_tracing_rocdecode_api_record_t /// @brief Specification of the API function, e.g., ::rocprofiler_rocdecode_api_id_t } rocprofiler_buffer_tracing_rocdecode_api_record_t; +/** + * @brief An extended ROCProfiler rocDecode API Tracer Record which includes function arguments. + * Pointers are not dereferenced. + */ +typedef struct rocprofiler_buffer_tracing_rocdecode_api_ext_record_t +{ + uint64_t size; ///< size of this struct + rocprofiler_buffer_tracing_kind_t kind; + rocprofiler_rocdecode_api_id_t operation; + rocprofiler_correlation_id_t correlation_id; ///< correlation ids for record + rocprofiler_timestamp_t start_timestamp; ///< start time in nanoseconds + rocprofiler_timestamp_t end_timestamp; ///< end time in nanoseconds + rocprofiler_thread_id_t thread_id; ///< id for thread generating this record + rocprofiler_rocdecode_api_args_t args; ///< arguments of function call + rocprofiler_rocdecode_api_retval_t retval; ///< return value of function call + + /// @var kind + /// @brief ::ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT + /// @var operation + /// @brief Specification of the API function (@see + /// ::rocprofiler_rocdecode_api_id_t) +} rocprofiler_buffer_tracing_rocdecode_api_ext_record_t; + /** * @brief ROCProfiler Buffer rocJPEG API Record. */ @@ -241,7 +266,8 @@ typedef struct rocprofiler_buffer_tracing_rocjpeg_api_record_t /// @var kind /// @brief ::ROCPROFILER_CALLBACK_TRACING_ROCJPEG_API /// @var operation - /// @brief Specification of the API function, e.g., ::rocprofiler_rocjpeg_api_id_t + /// @brief Specification of the API function (@see + /// ::rocprofiler_rocjpeg_api_id_t) } rocprofiler_buffer_tracing_rocjpeg_api_record_t; /** diff --git a/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/cxx/serialization.hpp b/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/cxx/serialization.hpp index d744643340..ced0371a85 100644 --- a/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/cxx/serialization.hpp +++ b/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/cxx/serialization.hpp @@ -469,7 +469,7 @@ template void save(ArchiveT& ar, rocprofiler_rocdecode_api_retval_t data) { - ROCP_SDK_SAVE_DATA_FIELD(rocDecStatus_retval); + ROCP_SDK_SAVE_DATA_FIELD(uint64_t_retval); } template @@ -605,6 +605,16 @@ save(ArchiveT& ar, rocprofiler_buffer_tracing_rocdecode_api_record_t data) save_buffer_tracing_api_record(ar, data); } +template +void +save(ArchiveT& ar, rocprofiler_buffer_tracing_rocdecode_api_ext_record_t data) +{ + save_buffer_tracing_api_record(ar, data); + auto args = sdk::serialization::get_buffer_tracing_args(data); + ROCP_SDK_SAVE_VALUE("args", args); + ROCP_SDK_SAVE_DATA_FIELD(retval); +} + template void save(ArchiveT& ar, rocprofiler_buffer_tracing_rocjpeg_api_record_t data) diff --git a/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/fwd.h b/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/fwd.h index 7cda8e77c1..70f59c055d 100644 --- a/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/fwd.h +++ b/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/fwd.h @@ -211,6 +211,7 @@ typedef enum rocprofiler_buffer_tracing_kind_t // NOLINT(performance-enum-size) ROCPROFILER_BUFFER_TRACING_HIP_STREAM, ///< @see ::rocprofiler_hip_stream_operation_t ROCPROFILER_BUFFER_TRACING_HIP_RUNTIME_API_EXT, ROCPROFILER_BUFFER_TRACING_HIP_COMPILER_API_EXT, + ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT, ROCPROFILER_BUFFER_TRACING_LAST, /// @var ROCPROFILER_BUFFER_TRACING_HIP_RUNTIME_API_EXT @@ -219,6 +220,9 @@ typedef enum rocprofiler_buffer_tracing_kind_t // NOLINT(performance-enum-size) /// @var ROCPROFILER_BUFFER_TRACING_HIP_COMPILER_API_EXT /// @brief Similar to ROCPROFILER_BUFFER_TRACING_HIP_COMPILER_API except the buffer record /// contains the function argument(s) and return value + /// @var ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT + /// @brief Similar to ROCPROFILER_BUFFER_TRACING_ROCDECODE_API except the buffer record + /// contains the function argument(s) and return value } rocprofiler_buffer_tracing_kind_t; /** diff --git a/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/rocdecode/api_args.h b/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/rocdecode/api_args.h index 13e0378d71..2082da9087 100644 --- a/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/rocdecode/api_args.h +++ b/projects/rocprofiler-sdk/source/include/rocprofiler-sdk/rocdecode/api_args.h @@ -51,6 +51,7 @@ typedef struct rocprofiler_rocdecode_api_no_args typedef union rocprofiler_rocdecode_api_retval_t { + uint64_t uint64_t_retval; int32_t rocDecStatus_retval; const char* const_charp_retval; } rocprofiler_rocdecode_api_retval_t; diff --git a/projects/rocprofiler-sdk/source/lib/output/buffered_output.hpp b/projects/rocprofiler-sdk/source/lib/output/buffered_output.hpp index 90de2d4aa7..ef48819374 100644 --- a/projects/rocprofiler-sdk/source/lib/output/buffered_output.hpp +++ b/projects/rocprofiler-sdk/source/lib/output/buffered_output.hpp @@ -160,7 +160,7 @@ using pc_sampling_host_trap_buffered_output_t = buffered_output; using rocdecode_buffered_output_t = - buffered_output; + buffered_output; using rocjpeg_buffered_output_t = buffered_output; using kernel_dispatch_buffered_output_with_stream_t = diff --git a/projects/rocprofiler-sdk/source/lib/output/generateCSV.cpp b/projects/rocprofiler-sdk/source/lib/output/generateCSV.cpp index 78581d86c0..9b04320c5d 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generateCSV.cpp +++ b/projects/rocprofiler-sdk/source/lib/output/generateCSV.cpp @@ -735,10 +735,10 @@ generate_csv(const output_config& cfg, } void -generate_csv(const output_config& cfg, - const metadata& tool_metadata, - const generator& data, - const stats_entry_t& stats) +generate_csv(const output_config& cfg, + const metadata& tool_metadata, + const generator& data, + const stats_entry_t& stats) { if(data.empty()) return; diff --git a/projects/rocprofiler-sdk/source/lib/output/generateCSV.hpp b/projects/rocprofiler-sdk/source/lib/output/generateCSV.hpp index 29df079987..7910d68545 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generateCSV.hpp +++ b/projects/rocprofiler-sdk/source/lib/output/generateCSV.hpp @@ -88,10 +88,10 @@ generate_csv(const output_config& cfg, const stats_entry_t& stats); void -generate_csv(const output_config& cfg, - const metadata& tool_metadata, - const generator& data, - const stats_entry_t& stats); +generate_csv(const output_config& cfg, + const metadata& tool_metadata, + const generator& data, + const stats_entry_t& stats); void generate_csv(const output_config& cfg, diff --git a/projects/rocprofiler-sdk/source/lib/output/generateJSON.cpp b/projects/rocprofiler-sdk/source/lib/output/generateJSON.cpp index 7bc279ba34..89f38765ee 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generateJSON.cpp +++ b/projects/rocprofiler-sdk/source/lib/output/generateJSON.cpp @@ -196,7 +196,7 @@ write_json(json_output& json_ar, generator scratch_memory_gen, generator rccl_api_gen, generator memory_allocation_gen, - generator rocdecode_api_gen, + generator rocdecode_api_gen, generator rocjpeg_api_gen, generator pc_sampling_host_trap_gen, generator pc_sampling_stochastic_gen) diff --git a/projects/rocprofiler-sdk/source/lib/output/generateJSON.hpp b/projects/rocprofiler-sdk/source/lib/output/generateJSON.hpp index b9db97a3c9..97d9fbab0b 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generateJSON.hpp +++ b/projects/rocprofiler-sdk/source/lib/output/generateJSON.hpp @@ -94,7 +94,7 @@ write_json(json_output& j generator scratch_memory_gen, generator rccl_api_gen, generator memory_allocation_gen, - generator rocdecode_api_gen, + generator rocdecode_api_gen, generator rocjpeg_api_gen, generator pc_sampling_host_trap_gen, generator pc_sampling_stochastic_gen); diff --git a/projects/rocprofiler-sdk/source/lib/output/generateOTF2.cpp b/projects/rocprofiler-sdk/source/lib/output/generateOTF2.cpp index 737663e7c9..37f18c0813 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generateOTF2.cpp +++ b/projects/rocprofiler-sdk/source/lib/output/generateOTF2.cpp @@ -368,7 +368,7 @@ write_otf2( std::deque* /*scratch_memory_data*/, std::deque* rccl_api_data, std::deque* memory_allocation_data, - std::deque* rocdecode_api_data, + std::deque* rocdecode_api_data, std::deque* rocjpeg_api_data) { namespace sdk = ::rocprofiler::sdk; @@ -618,10 +618,31 @@ 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{}); add_event_data(rocjpeg_api_data, sdk::category::rocjpeg_api{}); } + for(auto itr : *rocdecode_api_data) + { + auto name = buffer_names.at(itr.kind, itr.operation); + _hash_data.emplace( + get_hash_id(name), + region_info{std::string{name}, OTF2_REGION_ROLE_FUNCTION, OTF2_PARADIGM_USER}); + + auto& _evt_info = thread_event_info.at(itr.thread_id); + _evt_info.event_count += 1; + + _data.emplace_back(evt_data{ROCPROFILER_CALLBACK_PHASE_ENTER, + name, + _evt_info.get_location(), + itr.start_timestamp, + get_attr(sdk::category::rocdecode_api{})}); + _data.emplace_back(evt_data{ROCPROFILER_CALLBACK_PHASE_EXIT, + name, + _evt_info.get_location(), + itr.end_timestamp, + nullptr}); + } + for(auto itr : *memory_copy_data) { auto name = buffer_names.at(itr.kind, itr.operation); diff --git a/projects/rocprofiler-sdk/source/lib/output/generateOTF2.hpp b/projects/rocprofiler-sdk/source/lib/output/generateOTF2.hpp index aa65dea91e..f358c567f9 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generateOTF2.hpp +++ b/projects/rocprofiler-sdk/source/lib/output/generateOTF2.hpp @@ -48,7 +48,7 @@ write_otf2( std::deque* scratch_memory_data, std::deque* rccl_api_data, std::deque* memory_allocation_data, - std::deque* rocdecode_api_data, + std::deque* rocdecode_api_data, std::deque* rocjpeg_api_data); } // namespace tool } // namespace rocprofiler diff --git a/projects/rocprofiler-sdk/source/lib/output/generatePerfetto.cpp b/projects/rocprofiler-sdk/source/lib/output/generatePerfetto.cpp index 296d69a3e2..ea3aefc171 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generatePerfetto.cpp +++ b/projects/rocprofiler-sdk/source/lib/output/generatePerfetto.cpp @@ -77,7 +77,7 @@ write_perfetto( const generator& /*scratch_memory_gen*/, const generator& rccl_api_gen, const generator& memory_allocation_gen, - const generator& rocdecode_api_gen, + const generator& rocdecode_api_gen, const generator& rocjpeg_api_gen) { namespace sdk = ::rocprofiler::sdk; @@ -493,8 +493,9 @@ write_perfetto( 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); + auto name = buffer_names.at(itr.kind, itr.operation); + auto& track = thread_tracks.at(itr.thread_id); + auto rocdecode_args = sdk::serialization::get_buffer_tracing_args(itr); TRACE_EVENT_BEGIN(sdk::perfetto_category::name, ::perfetto::StaticString(name.data()), @@ -516,7 +517,14 @@ write_perfetto( "corr_id", itr.correlation_id.internal, "ancestor_id", - itr.correlation_id.ancestor); + itr.correlation_id.ancestor, + [&](::perfetto::EventContext ctx) { + for(const auto& rocdecode_arg : rocdecode_args) + { + sdk::add_perfetto_annotation( + ctx, rocdecode_arg.name, rocdecode_arg.value); + } + }); TRACE_EVENT_END(sdk::perfetto_category::name, track, itr.end_timestamp); diff --git a/projects/rocprofiler-sdk/source/lib/output/generatePerfetto.hpp b/projects/rocprofiler-sdk/source/lib/output/generatePerfetto.hpp index cf26ef3433..c4626c5327 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generatePerfetto.hpp +++ b/projects/rocprofiler-sdk/source/lib/output/generatePerfetto.hpp @@ -49,7 +49,7 @@ write_perfetto( const generator& scratch_memory_gen, const generator& rccl_api_gen, const generator& memory_allocation_gen, - const generator& rocdecode_api_gen, + const generator& rocdecode_api_gen, const generator& rocjpeg_api_gen); } // namespace tool } // namespace rocprofiler diff --git a/projects/rocprofiler-sdk/source/lib/output/generateStats.cpp b/projects/rocprofiler-sdk/source/lib/output/generateStats.cpp index 98f80783b5..302ce33dc4 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generateStats.cpp +++ b/projects/rocprofiler-sdk/source/lib/output/generateStats.cpp @@ -230,8 +230,8 @@ generate_stats(const output_config& /*cfg*/, stats_entry_t generate_stats(const output_config& /*cfg*/, - const metadata& tool_metadata, - const generator& data) + const metadata& tool_metadata, + const generator& data) { auto rocdecode_stats = stats_map_t{}; for(auto ditr : data) diff --git a/projects/rocprofiler-sdk/source/lib/output/generateStats.hpp b/projects/rocprofiler-sdk/source/lib/output/generateStats.hpp index ef7ae20aba..87790d2a87 100644 --- a/projects/rocprofiler-sdk/source/lib/output/generateStats.hpp +++ b/projects/rocprofiler-sdk/source/lib/output/generateStats.hpp @@ -77,9 +77,9 @@ generate_stats(const output_config& cfg, const generator& data); stats_entry_t -generate_stats(const output_config& cfg, - const metadata& tool_metadata, - const generator& data); +generate_stats(const output_config& cfg, + const metadata& tool_metadata, + const generator& data); stats_entry_t generate_stats(const output_config& cfg, diff --git a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk-tool/tool.cpp b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk-tool/tool.cpp index c24b7a2243..9e08420ad1 100644 --- a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk-tool/tool.cpp +++ b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk-tool/tool.cpp @@ -962,9 +962,9 @@ buffered_tracing_callback(rocprofiler_context_id_t /*context*/, tool::write_ring_buffer(*record, domain_type::RCCL); } - else if(header->kind == ROCPROFILER_BUFFER_TRACING_ROCDECODE_API) + else if(header->kind == ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT) { - auto* record = static_cast( + auto* record = static_cast( header->payload); tool::write_ring_buffer(*record, domain_type::ROCDECODE); @@ -1747,13 +1747,13 @@ tool_init(rocprofiler_client_finalize_t fini_func, void* tool_data) &get_buffers().rocdecode_api_trace), "buffer creation"); - ROCPROFILER_CALL( - rocprofiler_configure_buffer_tracing_service(get_client_ctx(), - ROCPROFILER_BUFFER_TRACING_ROCDECODE_API, - nullptr, - 0, - get_buffers().rocdecode_api_trace), - "buffer tracing service for ROCDecode api configure"); + ROCPROFILER_CALL(rocprofiler_configure_buffer_tracing_service( + get_client_ctx(), + ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT, + nullptr, + 0, + get_buffers().rocdecode_api_trace), + "buffer tracing service for ROCDecode api configure"); } if(tool::get_config().rocjpeg_api_trace) diff --git a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/buffer_tracing.cpp b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/buffer_tracing.cpp index fc0606f603..1dad9cf6bd 100644 --- a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/buffer_tracing.cpp +++ b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/buffer_tracing.cpp @@ -102,6 +102,7 @@ ROCPROFILER_BUFFER_TRACING_KIND_STRING(ROCJPEG_API) ROCPROFILER_BUFFER_TRACING_KIND_STRING(HIP_STREAM) ROCPROFILER_BUFFER_TRACING_KIND_STRING(HIP_RUNTIME_API_EXT) ROCPROFILER_BUFFER_TRACING_KIND_STRING(HIP_COMPILER_API_EXT) +ROCPROFILER_BUFFER_TRACING_KIND_STRING(ROCDECODE_API_EXT) template std::pair @@ -302,6 +303,7 @@ rocprofiler_query_buffer_tracing_kind_operation_name(rocprofiler_buffer_tracing_ return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED; } case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API: + case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT: { val = rocprofiler::rocdecode::name_by_id(operation); @@ -451,6 +453,7 @@ rocprofiler_iterate_buffer_tracing_kind_operations( return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED; } case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API: + case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT: { ops = rocprofiler::rocdecode::get_ids(); break; @@ -518,6 +521,14 @@ rocprofiler_iterate_buffer_tracing_record_args( _payload->operation, _payload->args, callback, user_data); return ROCPROFILER_STATUS_SUCCESS; } + case ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT: + { + auto* _payload = + static_cast(record.payload); + rocprofiler::rocdecode::iterate_args( + _payload->operation, _payload->args, callback, user_data); + return ROCPROFILER_STATUS_SUCCESS; + } } return ROCPROFILER_STATUS_ERROR_NOT_IMPLEMENTED; diff --git a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/callback_tracing.cpp b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/callback_tracing.cpp index e48cc32500..abd552a0ae 100644 --- a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/callback_tracing.cpp +++ b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/callback_tracing.cpp @@ -574,6 +574,17 @@ rocprofiler_iterate_callback_tracing_kind_operation_args( user_data); return ROCPROFILER_STATUS_SUCCESS; } + case ROCPROFILER_CALLBACK_TRACING_ROCDECODE_API: + { + rocprofiler::rocdecode::iterate_args( + record.operation, + static_cast(record.payload) + ->args, + callback, + max_deref, + user_data); + return ROCPROFILER_STATUS_SUCCESS; + } case ROCPROFILER_CALLBACK_TRACING_SCRATCH_MEMORY: case ROCPROFILER_CALLBACK_TRACING_CODE_OBJECT: case ROCPROFILER_CALLBACK_TRACING_KERNEL_DISPATCH: @@ -581,7 +592,6 @@ rocprofiler_iterate_callback_tracing_kind_operation_args( case ROCPROFILER_CALLBACK_TRACING_MEMORY_ALLOCATION: case ROCPROFILER_CALLBACK_TRACING_RCCL_API: case ROCPROFILER_CALLBACK_TRACING_RUNTIME_INITIALIZATION: - case ROCPROFILER_CALLBACK_TRACING_ROCDECODE_API: case ROCPROFILER_CALLBACK_TRACING_ROCJPEG_API: case ROCPROFILER_CALLBACK_TRACING_HIP_STREAM: { diff --git a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/defines.hpp b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/defines.hpp index affe75ec33..5deaa04316 100644 --- a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/defines.hpp +++ b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/defines.hpp @@ -94,9 +94,14 @@ return &base_type::functor; \ } \ \ - static std::vector as_arg_addr(callback_data_type) { return std::vector{}; } \ + static std::vector as_arg_addr(rocprofiler_rocdecode_api_args_t) \ + { \ + return std::vector{}; \ + } \ \ - static std::vector as_arg_list(callback_data_type, int32_t) \ + static std::vector as_arg_list( \ + rocprofiler_rocdecode_api_args_t, \ + int32_t) \ { \ return {}; \ } \ @@ -175,16 +180,15 @@ return &base_type::functor; \ } \ \ - static std::vector as_arg_addr(callback_data_type trace_data) \ + static std::vector as_arg_addr(rocprofiler_rocdecode_api_args_t args) \ { \ return std::vector{ \ - GET_ADDR_MEMBER_FIELDS(get_api_data_args(trace_data.args), __VA_ARGS__)}; \ + GET_ADDR_MEMBER_FIELDS(get_api_data_args(args), __VA_ARGS__)}; \ } \ - static auto as_arg_list(callback_data_type trace_data, int32_t max_deref) \ + static auto as_arg_list(rocprofiler_rocdecode_api_args_t args, int32_t max_deref) \ { \ return utils::stringize( \ - max_deref, \ - GET_NAMED_MEMBER_FIELDS(get_api_data_args(trace_data.args), __VA_ARGS__)); \ + max_deref, GET_NAMED_MEMBER_FIELDS(get_api_data_args(args), __VA_ARGS__)); \ } \ }; \ } \ diff --git a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.cpp b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.cpp index 288815ee00..10fd9291b3 100644 --- a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.cpp +++ b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.cpp @@ -22,11 +22,11 @@ #include "lib/rocprofiler-sdk/rocdecode/rocdecode.hpp" #include "lib/common/defines.hpp" +#include "lib/common/mpl.hpp" #include "lib/common/static_object.hpp" +#include "lib/common/string_entry.hpp" #include "lib/rocprofiler-sdk/buffer.hpp" #include "lib/rocprofiler-sdk/context/context.hpp" -#include "lib/rocprofiler-sdk/hip/hip.hpp" -#include "lib/rocprofiler-sdk/hip/utils.hpp" #include "lib/rocprofiler-sdk/registration.hpp" #include "lib/rocprofiler-sdk/tracing/tracing.hpp" @@ -35,11 +35,6 @@ #include #include -#include -#include -// must be included after runtime api -#include - #include #include #include @@ -101,6 +96,23 @@ template auto* get_table(); +template +decltype(auto) +convert_arg(Tp&& val) +{ + using data_type = common::mpl::unqualified_type_t; + if constexpr(std::is_same::value) + { + return common::get_string_entry(val)->c_str(); + } + else + { + static_assert(!common::mpl::is_string_type::value, + "argument type is a string type. preceding if constexpr is incorrect"); + return std::forward(val); + } +} + } // namespace template @@ -149,6 +161,7 @@ rocdecode_api_impl::functor(Args... args) using info_type = rocdecode_api_info; using callback_api_data_t = typename rocdecode_domain_info::callback_data_type; using buffered_api_data_t = typename rocdecode_domain_info::buffer_data_type; + using buffered_ext_data_t = typename rocdecode_domain_info::buffered_ext_data_type; constexpr auto external_corr_id_domain_idx = rocdecode_domain_info::external_correlation_id_domain_idx; @@ -166,6 +179,7 @@ rocdecode_api_impl::functor(Args... args) auto thr_id = common::get_tid(); auto callback_contexts = tracing::callback_context_data_vec_t{}; auto buffered_contexts = tracing::buffered_context_data_vec_t{}; + auto extended_contexts = tracing::buffered_context_data_vec_t{}; auto external_corr_ids = tracing::external_correlation_id_map_t{}; tracing::populate_contexts(info_type::callback_domain_idx, @@ -174,8 +188,12 @@ rocdecode_api_impl::functor(Args... args) callback_contexts, buffered_contexts, external_corr_ids); + tracing::populate_contexts(info_type::buffered_ext_domain_idx, + info_type::operation_idx, + extended_contexts, + external_corr_ids); - if(callback_contexts.empty() && buffered_contexts.empty()) + if(callback_contexts.empty() && buffered_contexts.empty() && extended_contexts.empty()) { [[maybe_unused]] auto _ret = exec(info_type::get_table_func(), std::forward(args)...); if constexpr(!std::is_void::value) @@ -185,6 +203,7 @@ rocdecode_api_impl::functor(Args... args) } auto buffer_record = common::init_public_api_struct(buffered_api_data_t{}); + auto extended_record = common::init_public_api_struct(buffered_ext_data_t{}); auto tracer_data = common::init_public_api_struct(callback_api_data_t{}); auto* corr_id = tracing::correlation_service::construct(ref_count); auto internal_corr_id = corr_id->internal; @@ -196,11 +215,16 @@ rocdecode_api_impl::functor(Args... args) info_type::operation_idx, internal_corr_id); + // set the arguments + if(!callback_contexts.empty() || !extended_contexts.empty()) + { + set_data_args(info_type::get_api_data_args(tracer_data.args), + convert_arg(std::forward(args))...); + } + // invoke the callbacks if(!callback_contexts.empty()) { - set_data_args(info_type::get_api_data_args(tracer_data.args), std::forward(args)...); - tracing::execute_phase_enter_callbacks(callback_contexts, thr_id, internal_corr_id, @@ -216,7 +240,7 @@ rocdecode_api_impl::functor(Args... args) external_corr_ids, thr_id, external_corr_id_domain_idx); // record the start timestamp as close to the function call as possible - if(!buffered_contexts.empty()) + if(!buffered_contexts.empty() || !extended_contexts.empty()) { buffer_record.start_timestamp = common::timestamp_ns(); } @@ -227,15 +251,18 @@ rocdecode_api_impl::functor(Args... args) auto _ret = exec(info_type::get_table_func(), std::forward(args)...); // record the end timestamp as close to the function call as possible - if(!buffered_contexts.empty()) + if(!buffered_contexts.empty() || !extended_contexts.empty()) { buffer_record.end_timestamp = common::timestamp_ns(); } - if(!callback_contexts.empty()) + if(!callback_contexts.empty() || !extended_contexts.empty()) { set_data_retval(tracer_data.retval, _ret); + } + if(!callback_contexts.empty()) + { tracing::execute_phase_exit_callbacks(callback_contexts, external_corr_ids, info_type::callback_domain_idx, @@ -255,6 +282,23 @@ rocdecode_api_impl::functor(Args... args) buffer_record); } + if(!extended_contexts.empty()) + { + extended_record.start_timestamp = buffer_record.start_timestamp; + extended_record.end_timestamp = buffer_record.end_timestamp; + extended_record.args = tracer_data.args; + extended_record.retval = tracer_data.retval; + + tracing::execute_buffer_record_emplace(extended_contexts, + thr_id, + internal_corr_id, + external_corr_ids, + ancestor_corr_id, + info_type::buffered_ext_domain_idx, + info_type::operation_idx, + extended_record); + } + // decrement the reference count after usage in the callback/buffers corr_id->sub_ref_count(); @@ -323,13 +367,13 @@ get_names(std::vector& _name_list, std::index_sequence(_name_list, std::index_sequence{}); } -template +template void -iterate_args(const uint32_t id, - const DataT& data, - rocprofiler_callback_tracing_operation_args_cb_t func, - int32_t max_deref, - void* user_data, +iterate_args(const uint32_t id, + const DataT& data, + FuncT func, + int32_t max_deref, + void* user_data, std::index_sequence) { if(OpIdx == id) @@ -339,16 +383,42 @@ iterate_args(const uint32_t id, auto&& arg_addr = info_type::as_arg_addr(data); for(size_t i = 0; i < std::min(arg_list.size(), arg_addr.size()); ++i) { - auto ret = func(info_type::callback_domain_idx, // kind - id, // operation - i, // arg_number - arg_addr.at(i), // arg_value_addr - arg_list.at(i).indirection_level, // indirection - arg_list.at(i).type, // arg_type - arg_list.at(i).name, // arg_name - arg_list.at(i).value.c_str(), // arg_value_str - arg_list.at(i).dereference_count, // num deref in str - user_data); + using return_type = typename common::mpl::function_traits::result_type; + + auto ret = return_type{}; + if constexpr(std::is_same::value) + { + ret = func(info_type::callback_domain_idx, // kind + id, // operation + i, // arg_number + arg_addr.at(i), // arg_value_addr + arg_list.at(i).indirection_level, // indirection + arg_list.at(i).type, // arg_type + arg_list.at(i).name, // arg_name + arg_list.at(i).value.c_str(), // arg_value_str + arg_list.at(i).dereference_count, // num deref in str + user_data); + } + else if constexpr(std::is_same::value) + { + ret = func(info_type::buffered_ext_domain_idx, // kind + id, // operation + i, // arg_number + arg_addr.at(i), // arg_value_addr + arg_list.at(i).indirection_level, // indirection + arg_list.at(i).type, // arg_type + arg_list.at(i).name, // arg_name + arg_list.at(i).value.c_str(), // arg_value_str + user_data); + } + else + { + static_assert(common::mpl::assert_false::value, + "Error! unsupported callback type"); + } + if(ret != 0) break; } return; @@ -361,6 +431,7 @@ iterate_args(const uint32_t id, bool should_wrap_functor(rocprofiler_callback_tracing_kind_t _callback_domain, rocprofiler_buffer_tracing_kind_t _buffered_domain, + rocprofiler_buffer_tracing_kind_t _buffered_ext_domain, int _operation) { // we loop over all the *registered* contexts and see if any of them, at any point in time, @@ -378,6 +449,11 @@ should_wrap_functor(rocprofiler_callback_tracing_kind_t _callback_domain, if(itr->buffered_tracer && itr->buffered_tracer->domains(_buffered_domain) && itr->buffered_tracer->domains(_buffered_domain, _operation)) return true; + + // if there is a buffered tracer enabled for the given domain and op, we need to wrap + if(itr->buffered_tracer && itr->buffered_tracer->domains(_buffered_ext_domain) && + itr->buffered_tracer->domains(_buffered_ext_domain, _operation)) + return true; } return false; } @@ -436,8 +512,10 @@ update_table(Tp* _orig, std::integral_constant) if(_info.offset() >= _orig->size) return; // check to see if there are any contexts which enable this operation in the HIP API domain - if(!should_wrap_functor( - _info.callback_domain_idx, _info.buffered_domain_idx, _info.operation_idx)) + if(!should_wrap_functor(_info.callback_domain_idx, + _info.buffered_domain_idx, + _info.buffered_ext_domain_idx, + _info.operation_idx)) return; ROCP_TRACE << "updating table entry for " << _info.name; @@ -511,11 +589,11 @@ get_names() template void -iterate_args(uint32_t id, - const rocprofiler_callback_tracing_rocdecode_api_data_t& data, - rocprofiler_callback_tracing_operation_args_cb_t callback, - int32_t max_deref, - void* user_data) +iterate_args(uint32_t id, + const rocprofiler_rocdecode_api_args_t& data, + rocprofiler_callback_tracing_operation_args_cb_t callback, + int32_t max_deref, + void* user_data) { if(callback) iterate_args(id, @@ -526,6 +604,22 @@ iterate_args(uint32_t id, std::make_index_sequence::last>{}); } +template +void +iterate_args(uint32_t id, + const rocprofiler_rocdecode_api_args_t& data, + rocprofiler_buffer_tracing_operation_args_cb_t callback, + void* user_data) +{ + if(callback) + iterate_args(id, + data, + callback, + 0, + user_data, + std::make_index_sequence::last>{}); +} + template void copy_table(TableT* _orig, uint64_t _tbl_instance) @@ -547,18 +641,21 @@ update_table(TableT* _orig) std::make_index_sequence::last>{}); } -using rocdecode_api_data_t = rocprofiler_callback_tracing_rocdecode_api_data_t; +using rocdecode_api_data_t = rocprofiler_rocdecode_api_args_t; using rocdecode_op_args_cb_t = rocprofiler_callback_tracing_operation_args_cb_t; +using rocdecode_op_args_bf_t = rocprofiler_buffer_tracing_operation_args_cb_t; -#define INSTANTIATE_ROCDECODE_TABLE_FUNC(TABLE_TYPE, TABLE_IDX) \ - template void copy_table(TABLE_TYPE * _tbl, uint64_t _instv); \ - template void update_table(TABLE_TYPE * _tbl); \ - template const char* name_by_id(uint32_t); \ - template uint32_t id_by_name(const char*); \ - template std::vector get_ids(); \ - template std::vector get_names(); \ - template void iterate_args( \ - uint32_t, const rocdecode_api_data_t&, rocdecode_op_args_cb_t, int32_t, void*); +#define INSTANTIATE_ROCDECODE_TABLE_FUNC(TABLE_TYPE, TABLE_IDX) \ + template void copy_table(TABLE_TYPE * _tbl, uint64_t _instv); \ + template void update_table(TABLE_TYPE * _tbl); \ + template const char* name_by_id(uint32_t); \ + template uint32_t id_by_name(const char*); \ + template std::vector get_ids(); \ + template std::vector get_names(); \ + template void iterate_args( \ + uint32_t, const rocdecode_api_data_t&, rocdecode_op_args_cb_t, int32_t, void*); \ + template void iterate_args( \ + uint32_t, const rocdecode_api_data_t&, rocdecode_op_args_bf_t, void*); INSTANTIATE_ROCDECODE_TABLE_FUNC(rocdecode_api_func_table_t, ROCPROFILER_ROCDECODE_TABLE_ID_CORE) } // namespace rocdecode diff --git a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.def.cpp b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.def.cpp index 8f58dfde4b..e2282f5af5 100644 --- a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.def.cpp +++ b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.def.cpp @@ -36,21 +36,23 @@ namespace rocdecode template <> struct rocdecode_domain_info { - using args_type = rocprofiler_rocdecode_api_args_t; - using retval_type = rocprofiler_rocdecode_api_retval_t; - using callback_data_type = rocprofiler_callback_tracing_rocdecode_api_data_t; - using buffer_data_type = rocprofiler_buffer_tracing_rocdecode_api_record_t; + using args_type = rocprofiler_rocdecode_api_args_t; + using retval_type = rocprofiler_rocdecode_api_retval_t; + using callback_data_type = rocprofiler_callback_tracing_rocdecode_api_data_t; + using buffer_data_type = rocprofiler_buffer_tracing_rocdecode_api_record_t; + using buffered_ext_data_type = rocprofiler_buffer_tracing_rocdecode_api_ext_record_t; }; template <> struct rocdecode_domain_info : rocdecode_domain_info { - using enum_type = rocprofiler_marker_core_api_id_t; - static constexpr auto callback_domain_idx = ROCPROFILER_CALLBACK_TRACING_ROCDECODE_API; - static constexpr auto buffered_domain_idx = ROCPROFILER_BUFFER_TRACING_ROCDECODE_API; - static constexpr auto none = ROCPROFILER_ROCDECODE_API_ID_NONE; - static constexpr auto last = ROCPROFILER_ROCDECODE_API_ID_LAST; + using enum_type = rocprofiler_marker_core_api_id_t; + static constexpr auto callback_domain_idx = ROCPROFILER_CALLBACK_TRACING_ROCDECODE_API; + static constexpr auto buffered_domain_idx = ROCPROFILER_BUFFER_TRACING_ROCDECODE_API; + static constexpr auto buffered_ext_domain_idx = ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT; + static constexpr auto none = ROCPROFILER_ROCDECODE_API_ID_NONE; + static constexpr auto last = ROCPROFILER_ROCDECODE_API_ID_LAST; static constexpr auto external_correlation_id_domain_idx = ROCPROFILER_EXTERNAL_CORRELATION_REQUEST_ROCDECODE_API; }; diff --git a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.hpp b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.hpp index da212c0125..8870e9c3e2 100644 --- a/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.hpp +++ b/projects/rocprofiler-sdk/source/lib/rocprofiler-sdk/rocdecode/rocdecode.hpp @@ -98,11 +98,18 @@ get_ids(); template void -iterate_args(uint32_t id, - const rocprofiler_callback_tracing_rocdecode_api_data_t& data, - rocprofiler_callback_tracing_operation_args_cb_t callback, - int32_t max_deref, - void* user_data); +iterate_args(uint32_t id, + const rocprofiler_rocdecode_api_args_t& data, + rocprofiler_callback_tracing_operation_args_cb_t callback, + int32_t max_deref, + void* user_data); + +template +void +iterate_args(uint32_t id, + const rocprofiler_rocdecode_api_args_t& data, + rocprofiler_buffer_tracing_operation_args_cb_t callback, + void* user_data); template void diff --git a/projects/rocprofiler-sdk/tests/rocdecode/validate.py b/projects/rocprofiler-sdk/tests/rocdecode/validate.py index 43007c5ed2..c601d78789 100644 --- a/projects/rocprofiler-sdk/tests/rocdecode/validate.py +++ b/projects/rocprofiler-sdk/tests/rocdecode/validate.py @@ -32,12 +32,11 @@ def test_data_structure(input_data): node_exists("buffer_records", sdk_data) node_exists("names", sdk_data["callback_records"]) - # Uncomment when rocprofiler register mainline supports rocdecode - # node_exists("rocdecode_api_traces", sdk_data["callback_records"]) + node_exists("rocdecode_api_traces", sdk_data["callback_records"]) node_exists("names", sdk_data["buffer_records"]) - # Uncomment when rocprofiler register mainline supports rocdecode - # node_exists("rocdecode_api_traces", sdk_data["buffer_records"]) + node_exists("rocdecode_api_traces", sdk_data["buffer_records"]) + node_exists("rocdecode_api_ext_traces", sdk_data["buffer_records"]) def test_size_entries(input_data): @@ -88,6 +87,11 @@ def test_timestamps(input_data): for itr in sdk_data["buffer_records"][titr]: assert itr["start_timestamp"] <= itr["end_timestamp"] + for itr in sdk_data["buffer_records"]["rocdecode_api_ext_traces"]: + assert itr["start_timestamp"] < itr["end_timestamp"] + assert itr["start_timestamp"] > 0 + assert itr["end_timestamp"] > 0 + def test_internal_correlation_ids(input_data): """Assure correlation ids are unique""" @@ -135,12 +139,11 @@ def test_rocdecode_traces(input_data): rocdecode_bf_traces = sdk_data["buffer_records"]["rocdecode_api_traces"] rocdecode_api_bf_ops = get_operation(buffer_records, "ROCDECODE_API") assert len(rocdecode_api_bf_ops[1]) == 16 + assert len(rocdecode_bf_traces) > 0 rocdecode_cb_traces = sdk_data["callback_records"]["rocdecode_api_traces"] rocdecode_api_cb_ops = get_operation(callback_records, "ROCDECODE_API") - # If rocDecode tracing is not supported, end early - if len(rocdecode_bf_traces) == 0: - return pytest.skip("rocdecode tracing unavailable") + assert ( rocdecode_api_bf_ops[1] == rocdecode_api_cb_ops[1] and len(rocdecode_api_cb_ops[1]) == 16 @@ -177,6 +180,19 @@ def test_rocdecode_traces(input_data): ]: assert call in api_calls + rocdecode_ext_bf_traces = sdk_data["buffer_records"]["rocdecode_api_ext_traces"] + assert len(rocdecode_ext_bf_traces) > 0 + # Ensure that the input file path has the correct string saved + assert len(rocdecode_ext_bf_traces[0].args) > 1 + assert rocdecode_ext_bf_traces[0].args[1].name == "input_file_path" + assert ( + rocdecode_ext_bf_traces[0].args[1].value.split("/")[-1] + == "AMD_driving_virtual_20-H265.265" + ) + for api_call in rocdecode_ext_bf_traces: + assert len(api_call["args"]) > 0 + assert (api_call.retval.uint64_t_retval in [0, 1], "Unexpected return values") + def test_retired_correlation_ids(input_data): data = input_data diff --git a/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/CMakeLists.txt b/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/CMakeLists.txt index 2eec4a41bd..d5708fe830 100644 --- a/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/CMakeLists.txt +++ b/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/CMakeLists.txt @@ -30,8 +30,8 @@ add_test( NAME rocprofv3-test-rocdecode-tracing-execute COMMAND $ --rocdecode-trace -d - ${CMAKE_CURRENT_BINARY_DIR}/%tag%-trace -o out --output-format json csv - --log-level env -- + ${CMAKE_CURRENT_BINARY_DIR}/%tag%-trace -o out --output-format json csv pftrace + otf2 --log-level env -- $,$<$:$>,rocdecode-demo> -i ${ROCDECODE_VIDEO_FILE}) @@ -52,10 +52,10 @@ add_test( NAME rocprofv3-test-rocdecode-tracing-validate COMMAND ${Python3_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/validate.py --json-input - ${CMAKE_CURRENT_BINARY_DIR}/rocdecode-trace/out_results.json --otf2-input - ${CMAKE_CURRENT_BINARY_DIR}/rocdecode-trace/out_results.otf2 --pftrace-input - ${CMAKE_CURRENT_BINARY_DIR}/rocdecode-trace/out_results.pftrace --csv-input - ${CMAKE_CURRENT_BINARY_DIR}/rocdecode-trace/out_rocdecode_api_trace.csv) + ${CMAKE_CURRENT_BINARY_DIR}/rocdecode-demo-trace/out_results.json --otf2-input + ${CMAKE_CURRENT_BINARY_DIR}/rocdecode-demo-trace/out_results.otf2 --pftrace-input + ${CMAKE_CURRENT_BINARY_DIR}/rocdecode-demo-trace/out_results.pftrace --csv-input + ${CMAKE_CURRENT_BINARY_DIR}/rocdecode-demo-trace/out_rocdecode_api_trace.csv) set_tests_properties( rocprofv3-test-rocdecode-tracing-validate diff --git a/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/conftest.py b/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/conftest.py index e13f3450dc..5c14f4d9cd 100644 --- a/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/conftest.py +++ b/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/conftest.py @@ -15,25 +15,25 @@ def pytest_addoption(parser): parser.addoption( "--json-input", action="store", - default="rocdecode-tracing/out_results.json", + default="rocdecode-trace/rocdecode-demo-trace/out_results.json", help="Input JSON", ) parser.addoption( "--otf2-input", action="store", - default="rocdecode-tracing/out_results.otf2", + default="rocdecode-trace/rocdecode-demo-trace/out_results.otf2", help="Input OTF2", ) parser.addoption( "--pftrace-input", action="store", - default="rocdecode-tracing/out_results.pftrace", + default="rocdecode-trace/rocdecode-demo-trace/out_results.pftrace", help="Input pftrace file", ) parser.addoption( "--csv-input", action="store", - default="rocdecode-tracing/out_rocdecode_api_trace.csv", + default="rocdecode-trace/rocdecode-demo-trace/out_rocdecode_api_trace.csv", help="Input CSV", ) diff --git a/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/validate.py b/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/validate.py index 05d115be6d..dc6deb7ecd 100755 --- a/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/validate.py +++ b/projects/rocprofiler-sdk/tests/rocprofv3/rocdecode-trace/validate.py @@ -32,9 +32,14 @@ def test_rocdeocde(json_data): buffer_records = data["buffer_records"] rocdecode_data = buffer_records["rocdecode_api"] - # If rocDecode tracing is not supported, end early - if len(rocdecode_data) == 0: - return pytest.skip("rocdecode tracing unavailable") + assert len(rocdecode_data) > 0 + # Ensure that the input file path has the correct string saved + assert len(rocdecode_data[0].args) > 1 + assert rocdecode_data[0].args[1].name == "input_file_path" + assert ( + rocdecode_data[0].args[1].value.split("/")[-1] + == "AMD_driving_virtual_20-H265.265" + ) _, bf_op_names = get_operation(data, "ROCDECODE_API") @@ -50,14 +55,21 @@ def test_rocdeocde(json_data): assert "end_timestamp" in node assert "start_timestamp" in node assert "thread_id" in node + assert "args" in node + assert "retval" in node assert node.size > 0 assert node.thread_id > 0 assert node.start_timestamp > 0 assert node.end_timestamp > 0 assert node.start_timestamp < node.end_timestamp + assert len(node.args) > 0 + assert node.retval.uint64_t_retval in [0, 1], "Unexpected return values" - assert data.strings.buffer_records[node.kind].kind == "ROCDECODE_API" + assert ( + data.strings.buffer_records[node.kind].kind == "ROCDECODE_API" + or data.strings.buffer_records[node.kind].kind == "ROCDECODE_API_EXT" + ) assert ( data.strings.buffer_records[node.kind].operations[node.operation] in bf_op_names @@ -65,9 +77,6 @@ def test_rocdeocde(json_data): def test_csv_data(csv_data): - # If rocDecode tracing is not supported, end early - if len(csv_data) == 0: - return pytest.skip("rocdecode tracing unavailable") assert len(csv_data) > 0, "Expected non-empty csv data" api_calls = [] @@ -95,7 +104,7 @@ def test_csv_data(csv_data): api_calls.append(row["Function"]) - assert row["Domain"] == "ROCDECODE_API" + assert row["Domain"] in ("ROCDECODE_API", "ROCDECODE_API_EXT") assert int(row["Process_Id"]) > 0 assert int(row["Thread_Id"]) > 0 assert int(row["Start_Timestamp"]) > 0 @@ -122,13 +131,6 @@ def test_csv_data(csv_data): def test_perfetto_data(pftrace_data, json_data): import rocprofiler_sdk.tests.rocprofv3 as rocprofv3 - # If rocDecode tracing is not supported, end early - if ( - pftrace_data == None - or len(json_data["rocprofiler-sdk-tool"]["buffer_records"]["rocdecode_api"]) == 0 - ): - return pytest.skip("rocdecode tracing unavailable") - rocprofv3.test_perfetto_data( pftrace_data, json_data, @@ -139,13 +141,6 @@ def test_perfetto_data(pftrace_data, json_data): def test_otf2_data(otf2_data, json_data): import rocprofiler_sdk.tests.rocprofv3 as rocprofv3 - # If rocDecode tracing is not supported, end early - if ( - otf2_data == None - or len(json_data["rocprofiler-sdk-tool"]["buffer_records"]["rocdecode_api"]) == 0 - ): - return pytest.skip("rocdecode tracing unavailable") - rocprofv3.test_otf2_data( otf2_data, json_data, diff --git a/projects/rocprofiler-sdk/tests/tools/json-tool.cpp b/projects/rocprofiler-sdk/tests/tools/json-tool.cpp index be37fefbb9..44974d7017 100644 --- a/projects/rocprofiler-sdk/tests/tools/json-tool.cpp +++ b/projects/rocprofiler-sdk/tests/tools/json-tool.cpp @@ -908,8 +908,10 @@ auto corr_id_retire_records = std::deque{}; auto rccl_api_bf_records = std::deque{}; auto rocdecode_api_bf_records = std::deque{}; -auto rocjpeg_api_bf_records = std::deque{}; -auto ompt_bf_records = std::deque{}; +auto rocdecode_api_ext_bf_records = + std::deque{}; +auto rocjpeg_api_bf_records = std::deque{}; +auto ompt_bf_records = std::deque{}; void tool_tracing_buffered(rocprofiler_context_id_t /*context*/, @@ -1043,6 +1045,13 @@ tool_tracing_buffered(rocprofiler_context_id_t /*context*/, rocdecode_api_bf_records.emplace_back(*record); } + else if(header->kind == ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT) + { + auto* record = static_cast( + header->payload); + + rocdecode_api_ext_bf_records.emplace_back(*record); + } else if(header->kind == ROCPROFILER_BUFFER_TRACING_ROCJPEG_API) { auto* record = @@ -1150,6 +1159,7 @@ rocprofiler_context_id_t runtime_init_callback_ctx = {}; rocprofiler_context_id_t runtime_init_buffered_ctx = {}; rocprofiler_context_id_t rocdecode_api_callback_ctx = {0}; rocprofiler_context_id_t rocdecode_api_buffered_ctx = {0}; +rocprofiler_context_id_t rocdecode_api_ext_buffered_ctx = {0}; rocprofiler_context_id_t rocjpeg_api_callback_ctx = {0}; rocprofiler_context_id_t rocjpeg_api_buffered_ctx = {0}; @@ -1167,6 +1177,7 @@ rocprofiler_buffer_id_t scratch_memory_buffer = {}; rocprofiler_buffer_id_t corr_id_retire_buffer = {}; rocprofiler_buffer_id_t rccl_api_buffered_buffer = {}; rocprofiler_buffer_id_t rocdecode_api_buffer = {}; +rocprofiler_buffer_id_t rocdecode_api_ext_buffer = {}; rocprofiler_buffer_id_t rocjpeg_api_buffer = {}; rocprofiler_buffer_id_t ompt_buffered_buffer = {}; @@ -1195,12 +1206,13 @@ auto contexts = std::unordered_map{ {"RCCL_API_BUFFERED", &rccl_api_buffered_ctx}, {"ROCDECODE_API_CALLBACK", &rocdecode_api_callback_ctx}, {"ROCDECODE_API_BUFFERED", &rocdecode_api_buffered_ctx}, + {"ROCDECODE_API_EXT_BUFFERED", &rocdecode_api_ext_buffered_ctx}, {"ROCJPEG_API_CALLBACK", &rocjpeg_api_callback_ctx}, {"ROCJPEG_API_BUFFERED", &rocjpeg_api_buffered_ctx}, {"OMPT_BUFFERED", &ompt_buffered_ctx}, }; -auto buffers = std::array{&runtime_init_buffered_buffer, +auto buffers = std::array{&runtime_init_buffered_buffer, &hsa_api_buffered_buffer, &hip_api_buffered_buffer, &marker_api_buffered_buffer, @@ -1214,6 +1226,7 @@ auto buffers = std::array{&runtime_init_buffered_b &rccl_api_buffered_buffer, &ompt_buffered_buffer, &rocdecode_api_buffer, + &rocdecode_api_ext_buffer, &rocjpeg_api_buffer}; auto agents = std::vector{}; @@ -1526,6 +1539,14 @@ tool_init(rocprofiler_client_finalize_t fini_func, void* tool_data) tool_data, &rocdecode_api_buffer), "buffer creation"); + ROCPROFILER_CALL(rocprofiler_create_buffer(rocdecode_api_ext_buffered_ctx, + buffer_size, + watermark, + ROCPROFILER_BUFFER_POLICY_LOSSLESS, + tool_tracing_buffered, + tool_data, + &rocdecode_api_ext_buffer), + "buffer creation"); ROCPROFILER_CALL(rocprofiler_create_buffer(rocjpeg_api_buffered_ctx, buffer_size, @@ -1668,6 +1689,14 @@ tool_init(rocprofiler_client_finalize_t fini_func, void* tool_data) rocdecode_api_buffer), "buffer tracing service for rocdecode api configure"); + ROCPROFILER_CALL( + rocprofiler_configure_buffer_tracing_service(rocdecode_api_ext_buffered_ctx, + ROCPROFILER_BUFFER_TRACING_ROCDECODE_API_EXT, + nullptr, + 0, + rocdecode_api_ext_buffer), + "buffer tracing service for rocdecode ext api configure"); + ROCPROFILER_CALL( rocprofiler_configure_buffer_tracing_service(rocjpeg_api_buffered_ctx, ROCPROFILER_BUFFER_TRACING_ROCJPEG_API, @@ -1847,6 +1876,7 @@ tool_fini(void* tool_data) << ", counter_collection_value_records=" << counter_collection_bf_records.size() << ", rocdecode_api_callback_records=" << rocdecode_api_cb_records.size() << ", rocdecode_api_bf_records=" << rocdecode_api_bf_records.size() + << ", rocdecode_api_ext_bf_records=" << rocdecode_api_ext_bf_records.size() << ", rocjpeg_api_callback_records=" << rocjpeg_api_cb_records.size() << ", rocjpeg_api_bf_records=" << rocjpeg_api_bf_records.size() << "...\n" << std::flush; @@ -1973,6 +2003,7 @@ write_json(call_stack_t* _call_stack) json_ar(cereal::make_nvp("retired_correlation_ids", corr_id_retire_records)); json_ar(cereal::make_nvp("counter_collection", counter_collection_bf_records)); json_ar(cereal::make_nvp("rocdecode_api_traces", rocdecode_api_bf_records)); + json_ar(cereal::make_nvp("rocdecode_api_ext_traces", rocdecode_api_ext_bf_records)); json_ar(cereal::make_nvp("rocjpeg_api_traces", rocjpeg_api_bf_records)); } catch(std::exception& e) {