SWDEV-492607: Adding ATT wrapper (#40)

* Adding att parser wrapper

* Adding ATT tests as optional

* Adding decoder API for query capability

* Removed samples

* Formatting

* adding new line

* Removed perfetto and moved to static library

* using default search for lib

* Updated to SDK

* Namespace changes

* Added tests

* Small refactor

* Updated API to receive agent_id

* Fixing tests

* Tidy fixes

* Not write to file

* Switch to filesystem.hpp

* Compilation fixes

* Formatting

* Tidy fix

* Removed likely

* Adding tests

* Added gfx9 test

* Adding gfx12 tests

* Formatting

* Enable tidy

* Fix tests

* Fix deadlock on agent test

* Workaround ASAN

* Moving query outside class.

* Fix standalone tool

* Addressing comments

* Formatting

* Change query name

* Fixed some tests. Updated PR comments.

* Formatting

* Improved coverage

* Formatting

* Fix for comments

* Formatting

* Adding some description. Fix error type.

---------

Co-authored-by: Giovanni Baraldi <gbaraldi@amd.com>
Este commit está contenido en:
Baraldi, Giovanni
2024-12-19 03:53:32 +01:00
cometido por GitHub
padre 9de284a568
commit 2c8e88a76b
Se han modificado 59 ficheros con 4250 adiciones y 1086 borrados
+92 -161
Ver fichero
@@ -25,217 +25,148 @@
# undef NDEBUG
#endif
#include <rocprofiler-sdk/amd_detail/thread_trace.h>
#include <rocprofiler-sdk/fwd.h>
#include <rocprofiler-sdk/registration.h>
#include <rocprofiler-sdk/rocprofiler.h>
#include "trace_callbacks.hpp"
#include <rocprofiler-sdk/cxx/codeobj/code_printing.hpp>
#include "common.hpp"
#include <atomic>
#ifdef ENABLE_ATT_FILES
# include <nlohmann/json.hpp>
#endif
#include <unistd.h>
#include <cassert>
#include <cstdint>
#include <cstdlib>
#include <iostream>
#include <map>
#include <mutex>
#include <sstream>
#include <string>
#include <string_view>
#include <unordered_map>
#include <fstream>
namespace ATTTest
{
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;
using CodeobjAddressTranslate = rocprofiler::sdk::codeobj::disassembly::CodeobjAddressTranslate;
using Instruction = rocprofiler::sdk::codeobj::disassembly::Instruction;
CodeobjAddressTranslate* codeobjTranslate = nullptr;
struct trace_data_t
{
int64_t id;
uint8_t* data;
uint64_t size;
ToolData* tool;
};
void
tool_codeobj_tracing_callback(rocprofiler_callback_tracing_record_t record,
rocprofiler_user_data_t* /* user_data */,
void* callback_data)
void* userdata)
{
C_API_BEGIN
if(record.kind != ROCPROFILER_CALLBACK_TRACING_CODE_OBJECT) return;
if(record.phase != ROCPROFILER_CALLBACK_PHASE_LOAD) return;
assert(callback_data && "Shader callback passed null!");
ToolData& tool = *reinterpret_cast<ToolData*>(callback_data);
assert(userdata && "Dispatch callback passed null!");
auto& tool = *reinterpret_cast<Callbacks::ToolData*>(userdata);
if(record.operation == ROCPROFILER_CODE_OBJECT_DEVICE_KERNEL_SYMBOL_REGISTER)
{
std::unique_lock<std::mutex> lg(tool.isa_map_mut);
auto* data = static_cast<kernel_symbol_data_t*>(record.payload);
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);
if(!data || !data->uri) return;
std::unique_lock<std::mutex> lg(tool.isa_map_mut);
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)
{
codeobjTranslate->addDecoder(
data->uri, data->code_object_id, data->load_delta, data->load_size);
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());
}
else
{
codeobjTranslate->addDecoder(reinterpret_cast<const void*>(data->memory_base),
data->memory_size,
data->code_object_id,
data->load_delta,
data->load_size);
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});
C_API_END
}
void
get_trace_data(rocprofiler_att_parser_data_type_t type, void* att_data, void* userdata)
{
C_API_BEGIN
assert(userdata && "ISA callback passed null!");
trace_data_t& trace_data = *reinterpret_cast<trace_data_t*>(userdata);
assert(trace_data.tool && "ISA callback passed null!");
ToolData& tool = *reinterpret_cast<ToolData*>(trace_data.tool);
std::unique_lock<std::mutex> lk(tool.isa_map_mut);
if(type == ROCPROFILER_ATT_PARSER_DATA_TYPE_OCCUPANCY)
{
const auto& ev = *reinterpret_cast<const rocprofiler_att_data_type_occupancy_t*>(att_data);
tool.wave_start_locations.insert({ev.offset, ev.marker_id});
if(ev.enabled)
tool.waves_started.fetch_add(1);
else
tool.waves_ended.fetch_add(1);
}
if(type != ROCPROFILER_ATT_PARSER_DATA_TYPE_ISA) return;
auto& event = *reinterpret_cast<rocprofiler_att_data_type_isa_t*>(att_data);
pcInfo pc{event.offset, event.marker_id};
auto it = tool.isa_map.find(pc);
if(it == tool.isa_map.end())
{
auto ptr = std::make_unique<TrackedIsa>();
try
{
auto unique_inst = codeobjTranslate->get(pc.marker_id, pc.addr);
if(unique_inst == nullptr) return;
ptr->inst = unique_inst->inst;
} catch(...)
{
return;
}
it = tool.isa_map.emplace(pc, std::move(ptr)).first;
}
it->second->hitcount.fetch_add(event.hitcount, std::memory_order_relaxed);
it->second->latency.fetch_add(event.latency, std::memory_order_relaxed);
C_API_END
}
uint64_t
copy_trace_data(int* seid, uint8_t** buffer, uint64_t* buffer_size, void* userdata)
{
trace_data_t& data = *reinterpret_cast<trace_data_t*>(userdata);
*seid = data.id;
*buffer_size = data.size;
*buffer = data.data;
data.size = 0;
return *buffer_size;
}
rocprofiler_status_t
isa_callback(char* isa_instruction,
uint64_t* isa_memory_size,
uint64_t* isa_size,
uint64_t marker_id,
uint64_t offset,
void* userdata)
{
C_API_BEGIN
assert(userdata && "ISA callback passed null!");
trace_data_t& trace_data = *reinterpret_cast<trace_data_t*>(userdata);
assert(trace_data.tool && "ISA callback passed null!");
ToolData& tool = *reinterpret_cast<ToolData*>(trace_data.tool);
std::unique_ptr<Instruction> instruction;
try
{
std::unique_lock<std::mutex> unique_lock(tool.isa_map_mut);
instruction = codeobjTranslate->get(marker_id, offset);
} catch(...)
{
return ROCPROFILER_STATUS_ERROR;
}
if(!instruction.get()) return ROCPROFILER_STATUS_ERROR_INVALID_ARGUMENT;
{
size_t tmp_isa_size = *isa_size;
*isa_size = instruction->inst.size();
if(*isa_size > tmp_isa_size) return ROCPROFILER_STATUS_ERROR_OUT_OF_RESOURCES;
}
memcpy(isa_instruction, instruction->inst.data(), *isa_size);
*isa_memory_size = instruction->size;
auto ptr = std::make_unique<TrackedIsa>();
ptr->inst = instruction->inst;
tool.isa_map.emplace(pcInfo{offset, marker_id}, std::move(ptr));
return ROCPROFILER_STATUS_SUCCESS;
C_API_END
return ROCPROFILER_STATUS_ERROR;
}
void
shader_data_callback(int64_t se_id,
shader_data_callback(rocprofiler_agent_id_t agent,
int64_t se_id,
void* se_data,
size_t data_size,
rocprofiler_user_data_t userdata)
{
C_API_BEGIN
assert(userdata.ptr && "Shader callback passed null!");
ToolData& tool = *reinterpret_cast<ToolData*>(userdata.ptr);
trace_data_t data{.id = se_id, .data = (uint8_t*) se_data, .size = data_size, .tool = &tool};
auto status = rocprofiler_att_parse_data(copy_trace_data, get_trace_data, isa_callback, &data);
if(status != ROCPROFILER_STATUS_SUCCESS)
std::cerr << "shader_data_callback failed with status " << status << std::endl;
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
}
void
callbacks_init()
finalize_json(void* userdata)
{
codeobjTranslate = new CodeobjAddressTranslate();
}
void
callbacks_fini()
{
delete codeobjTranslate;
assert(userdata && "Dispatch callback passed null!");
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
}
} // namespace Callbacks
} // namespace ATTTest