AI/rocm-systems
2
0
Forka 0
Codice Problemi Pull Request Actions Pacchetti Progetti Rilasci Wiki Attività

PC sampling: online partial PC sampling decoding (#1004)

Sfoglia il codice sorgente 
* PC sampling: online partial PC sampling decoding

PC sampling service decodes a PC sample partially
by replacing the PC with an id of the loaded code object instance
containing PC and the offset of the PC within that code object instance.

* PC sampling: marker records removed

* PC sampling parser: minor doc update in mock

* PC sampling: introducing rocprofiler_pc_t

* NULL value of the code object id introduced.

* Clarifying documenation related to PC offset.

* PC offset documentation improvement

* PC sampling parser benchmark: Reducing the number of samples to recreate half of performance.
This commit is contained in:
Vladimir Indic
2024-09-05 18:35:46 +02:00
committato da GitHub
parent fa91169479
commit 93e82663d9
12 ha cambiato i file con 160 aggiunte e 129 eliminazioni
Scarica il file Patch Scarica il file Diff Expand all files Collapse all files
source/lib/rocprofiler-sdk/pc_sampling/code_object.cpp
+28 -48
Vedi File
@@ -26,6 +26,7 @@
# include "lib/common/container/operators.hpp"
# include "lib/common/logging.hpp"
# include "lib/common/static_object.hpp"
# include "lib/rocprofiler-sdk/code_object/code_object.hpp"
# include "lib/rocprofiler-sdk/pc_sampling/service.hpp"
@@ -44,6 +45,13 @@ namespace pc_sampling
{
namespace code_object
{
CodeobjTableTranslatorSynchronized*
get_code_object_translator()
{
static auto*& _v = common::static_object<CodeobjTableTranslatorSynchronized>::construct();
return _v;
}
namespace
{
auto&
@@ -74,14 +82,13 @@ get_destroy_function()
* @param [in] code_object - loaded/unloaded code object.
*/
void
flush_buffers_generate_marker_record(rocprofiler_callback_phase_t phase,
const rocprofiler::code_object::hsa::code_object& code_object)
flush_pc_sampling_buffers(const rocprofiler::code_object::hsa::code_object& code_object)
{
auto agent_id = code_object.rocp_data.rocp_agent;
if(!is_pc_sample_service_configured(agent_id)) return;
// The PC sampling service is configured on the agent.
// Find the agent's buffer and place marker record.
// The PC sampling service is configured on the agent,
// so flush its PC sampling buffer
// TODO: Creating a function that gives the buffer_id based on the agent_id?
const auto* pcs_service = get_configured_pc_sampling_service().load();
const auto* agent_session = pcs_service->agent_sessions.at(agent_id).get();
@@ -89,46 +96,6 @@ flush_buffers_generate_marker_record(rocprofiler_callback_phase_t
// flush internal PC sampling buffers
flush_internal_agent_buffers(agent_buffer_id);
auto* buff = rocprofiler::buffer::get_buffer(agent_buffer_id);
// create code object load/unload marker record and emplace it into the SDK's PC SAMPLING
// buffer.
if(phase == ROCPROFILER_CALLBACK_PHASE_LOAD)
{
auto marker =
common::init_public_api_struct(rocprofiler_pc_sampling_code_object_load_marker_t{});
marker.code_object_id = code_object.rocp_data.code_object_id;
// emplace marker to the SDK's PC sampling buffer
buff->emplace(ROCPROFILER_BUFFER_CATEGORY_PC_SAMPLING,
ROCPROFILER_PC_SAMPLING_RECORD_CODE_OBJECT_LOAD_MARKER,
marker);
}
else
{
auto marker =
common::init_public_api_struct(rocprofiler_pc_sampling_code_object_unload_marker_t{});
marker.code_object_id = code_object.rocp_data.code_object_id;
// emplace marker to the SDK's PC sampling buffer
buff->emplace(ROCPROFILER_BUFFER_CATEGORY_PC_SAMPLING,
ROCPROFILER_PC_SAMPLING_RECORD_CODE_OBJECT_UNLOAD_MARKER,
marker);
}
// Assuming that the `rocprofiler_pc_sampling_code_object_load_marker_t` and
// `rocprofiler_pc_sampling_code_object_unload_marker_t` share the same content,
// we could replace the previous if else with the following
/*
auto marker =
common::init_public_api_struct(rocprofiler_pc_sampling_code_object_load_marker_t{});
marker.code_object_id = code_object.rocp_data.code_object_id;
// emplace marker to the SDK's PC sampling buffer
buff->emplace(ROCPROFILER_BUFFER_CATEGORY_PC_SAMPLING,
(phase == ROCPROFILER_CALLBACK_PHASE_LOAD) ?
ROCPROFILER_PC_SAMPLING_RECORD_CODE_OBJECT_LOAD_MARKER
: ROCPROFILER_PC_SAMPLING_RECORD_CODE_OBJECT_UNLOAD_MARKER,
marker);
*/
}
hsa_status_t
@@ -141,7 +108,14 @@ executable_freeze(hsa_executable_t executable, const char* options)
rocprofiler::code_object::iterate_loaded_code_objects(
[&](const rocprofiler::code_object::hsa::code_object& code_object) {
if(code_object.hsa_executable == executable)
flush_buffers_generate_marker_record(ROCPROFILER_CALLBACK_PHASE_LOAD, code_object);
{
const auto& code_object_rocp = code_object.rocp_data;
get_code_object_translator()->insert(
address_range_t{code_object_rocp.load_base,
code_object_rocp.load_size,
code_object_rocp.code_object_id});
flush_pc_sampling_buffers(code_object);
}
});
return HSA_STATUS_SUCCESS;
@@ -153,8 +127,14 @@ executable_destroy(hsa_executable_t executable)
rocprofiler::code_object::iterate_loaded_code_objects(
[&](const rocprofiler::code_object::hsa::code_object& code_object) {
if(code_object.hsa_executable == executable)
flush_buffers_generate_marker_record(ROCPROFILER_CALLBACK_PHASE_UNLOAD,
code_object);
{
flush_pc_sampling_buffers(code_object);
const auto& code_object_rocp = code_object.rocp_data;
get_code_object_translator()->remove(
address_range_t{code_object_rocp.load_base,
code_object_rocp.load_size,
code_object_rocp.code_object_id});
}
});
// Call underlying function
@@ -183,7 +163,7 @@ finalize()
{
rocprofiler::code_object::iterate_loaded_code_objects(
[&](const rocprofiler::code_object::hsa::code_object& code_object) {
flush_buffers_generate_marker_record(ROCPROFILER_CALLBACK_PHASE_UNLOAD, code_object);
flush_pc_sampling_buffers(code_object);
});
}