Thread trace and Trace Decoder API tests and samples (#416)

* Adding test and samples to decoder

* Fix sample

* Formatting

* Fix multi test

* Disable sample

* Fix tests

* Format

* Version fix

* Locking the decoder

* Add atomic

* Review comments

* Format

* Adding readme

* merge conflict and adding PCS+ATT test

* Review comments

* Properly disable PCS test

* Update tests/rocprofv3/advanced-thread-trace/CMakeLists.txt

* Adding back env var test

* Name fix

* Preload sample

* Addressing review comments

* Update docs

---------

Co-authored-by: Giovanni Baraldi <gbaraldi@amd.com>
Αυτή η υποβολή περιλαμβάνεται σε:
Baraldi, Giovanni
2025-07-23 03:08:12 +02:00
υποβλήθηκε από GitHub
γονέας bf0fad1d54
υποβολή e898079a13
22 αρχεία άλλαξαν με 982 προσθήκες και 404 διαγραφές
@@ -26,11 +26,6 @@
#endif
#include "trace_callbacks.hpp"
#include <rocprofiler-sdk/cxx/codeobj/code_printing.hpp>
#ifdef ENABLE_ATT_FILES
# include <nlohmann/json.hpp>
#endif
#include <unistd.h>
#include <cassert>
@@ -38,135 +33,78 @@
namespace Callbacks
{
using code_obj_load_data_t = rocprofiler_callback_tracing_code_object_load_data_t;
using kernel_symbol_data_t = rocprofiler_callback_tracing_code_object_kernel_symbol_register_data_t;
rocprofiler_thread_trace_decoder_id_t decoder{};
std::atomic<size_t> latency{0};
void
tool_codeobj_tracing_callback(rocprofiler_callback_tracing_record_t record,
rocprofiler_user_data_t* /* user_data */,
void* userdata)
void* /* userdata */)
{
C_API_BEGIN
if(record.kind != ROCPROFILER_CALLBACK_TRACING_CODE_OBJECT) return;
if(record.phase != ROCPROFILER_CALLBACK_PHASE_LOAD) return;
assert(userdata && "Dispatch callback passed null!");
auto& tool = *reinterpret_cast<Callbacks::ToolData*>(userdata);
if(record.operation == ROCPROFILER_CODE_OBJECT_DEVICE_KERNEL_SYMBOL_REGISTER)
{
auto* data = static_cast<kernel_symbol_data_t*>(record.payload);
tool.kernel_id_to_kernel_name.emplace(data->kernel_id, data->kernel_name);
}
if(record.operation != ROCPROFILER_CODE_OBJECT_LOAD) return;
auto* data = static_cast<code_obj_load_data_t*>(record.payload);
auto* data = static_cast<rocprofiler_callback_tracing_code_object_load_data_t*>(record.payload);
if(data->storage_type == ROCPROFILER_CODE_OBJECT_STORAGE_TYPE_FILE) return;
static std::atomic<int> filecnt{0};
std::string name = "codeobj_" + std::to_string(filecnt.fetch_add(1)) + ".out";
#ifdef ENABLE_ATT_FILES
if(std::string_view(data->uri).find("file:///") == 0)
if(record.phase != ROCPROFILER_CALLBACK_PHASE_LOAD)
{
rocprofiler::sdk::codeobj::disassembly::CodeObjectBinary binary(data->uri);
std::ofstream file(tool.out_dir + name, std::ios::binary);
assert(file.is_open() && "Could not open codeobj file for writing");
file.write((char*) binary.buffer.data(), binary.buffer.size());
DECODER_CALL(
rocprofiler_thread_trace_decoder_codeobj_unload(decoder, data->code_object_id));
return;
}
else
{
std::ofstream file(tool.out_dir + name, std::ios::binary);
file.write((char*) data->memory_base, data->memory_size);
}
#endif
auto _lk = std::unique_lock{tool.mut};
tool.codeobjs.push_back(
{data->load_delta, data->load_size, data->code_object_id, name, data->uri});
DECODER_CALL(rocprofiler_thread_trace_decoder_codeobj_load(
decoder,
data->code_object_id,
data->load_delta,
data->load_size,
reinterpret_cast<const void*>(data->memory_base),
data->memory_size));
}
C_API_END
typedef void (*rocprofiler_thread_trace_decoder_callback_t)(
rocprofiler_thread_trace_decoder_record_type_t record_type_id,
void* trace_events,
uint64_t trace_size,
void* userdata);
void
shader_data_callback(rocprofiler_agent_id_t /* agent */,
int64_t /* se_id */,
void* se_data,
size_t data_size,
rocprofiler_user_data_t /* userdata */)
{
auto parse = [](rocprofiler_thread_trace_decoder_record_type_t record_type_id,
void* trace_events,
uint64_t trace_size,
void* /* userdata */) {
if(record_type_id != ROCPROFILER_THREAD_TRACE_DECODER_RECORD_WAVE) return;
for(size_t w = 0; w < trace_size; w++)
{
auto* wave = static_cast<rocprofiler_thread_trace_decoder_wave_t*>(trace_events);
for(size_t i = 0; i < wave->instructions_size; i++)
latency += wave->instructions_array[i].duration;
}
};
DECODER_CALL(rocprofiler_trace_decode(decoder, parse, se_data, data_size, nullptr));
}
void
shader_data_callback(rocprofiler_agent_id_t agent,
int64_t se_id,
void* se_data,
size_t data_size,
rocprofiler_user_data_t userdata)
init()
{
C_API_BEGIN
assert(userdata.ptr && "Dispatch callback passed null!");
auto& tool = *reinterpret_cast<Callbacks::ToolData*>(userdata.ptr);
std::string name = "agent_" + std::to_string(agent.handle) + "_shader_engine_" +
std::to_string(se_id) + "_" + std::to_string(agent.handle) + ".att";
#ifdef ENABLE_ATT_FILES
{
std::ofstream file(tool.out_dir + name, std::ios::binary);
assert(file.is_open() && "Could not open ATT file for writing");
file.write((char*) se_data, data_size);
}
#endif
assert(se_data);
assert(data_size);
auto _lk = std::unique_lock{tool.mut};
tool.att_files.push_back(name);
C_API_END
// const char* decoder_lib = std::getenv("ROCPROF_TRACE_DECODER_PATH");
DECODER_CALL(rocprofiler_thread_trace_decoder_create(&decoder, "/opt/rocm/lib"));
}
void
finalize_json(void* userdata)
finalize(void* /* tool_data */)
{
assert(userdata && "Dispatch callback passed null!");
rocprofiler_thread_trace_decoder_destroy(decoder);
auto& tool = *reinterpret_cast<Callbacks::ToolData*>(userdata);
auto _lk = std::unique_lock{tool.mut};
assert(!tool.att_files.empty());
#ifdef ENABLE_ATT_FILES
nlohmann::json att_json;
for(auto& file : tool.att_files)
att_json.push_back(file);
nlohmann::json codeobj_json;
nlohmann::json snapshot_json;
for(auto& file : tool.codeobjs)
{
nlohmann::json codeobj;
codeobj["code_object_id"] = file.id;
codeobj["load_delta"] = file.addr;
codeobj["load_size"] = file.size;
codeobj["uri"] = file.uri;
codeobj["filename"] = file.filename;
codeobj_json.push_back(codeobj);
nlohmann::json pair_json;
pair_json["key"] = file.id;
pair_json["value"] = file.filename;
snapshot_json.push_back(pair_json);
}
nlohmann::json tool_json;
tool_json["strings"]["att_files"] = att_json;
tool_json["code_objects"] = codeobj_json;
tool_json["strings"]["code_object_snapshot_files"] = snapshot_json;
nlohmann::json array;
array.push_back(tool_json);
nlohmann::json sdk_json;
sdk_json["rocprofiler-sdk-tool"] = array;
std::ofstream json_file(tool.out_dir + (std::to_string(getpid()) + "_results.json"));
assert(json_file.is_open() && "Could not open json file for writing!");
json_file << sdk_json;
#endif
if(latency.load() == 0) std::cerr << "Error: No latency was assigned to the trace!";
}
} // namespace Callbacks