ROCpd support [Part 2] (#109)

* Rocpd part 2, caching

* Fix shadowed variables

* backward compatibility

* Fixed designated initializers

* Fix timemory include

* Remove benchmark & Fix build issues for rhel

* Add missing bracket

* Fix shadowing and pedantic

* Fix pedantic pt2

* Fix duplicated SDK calls

* Add decay in get_size_impl

* Rename sample cache to trace cache

* Add cache storage supported types

* Resolving track naming in sampling module

* fix sampling of flushing thread

* fix sampling of flushing thread 2

* throw exception upon store while buffer storage is not running

* Prevent fork crashing

* Fix rebase issue

* Applied suggestions from code review

* Change flushing thread to use PTL

* Fix agent creation order

* Fix stream id ci throw

* Remove force setup of rocprofiler-sdk

* Code cleanup

* Change initialization for agent

* Add missing namespace

* Fix the mismatch within the tool_agent->device_id

* Switch from using handle to use agent type index

* Fix pmc info comparator in metadata registry

---------

Co-authored-by: Aleksandar <aleksandar.djordjevic@amd.com>
Co-authored-by: Milan Radosavljevic <milan.radosavljevic@amd.com>
Co-authored-by: Marjan Antic <marantic@amd.com>
This commit is contained in:
authored and GitHub committed 2025-08-19 22:01:04 -04:00
1 parent 351d598869
commit 1f86010ca2
40 files changed
+3432 -1184

No files matched your search

@@ -23,6 +23,7 @@
#include "library/components/backtrace_metrics.hpp"
#include "core/agent.hpp"
#include "core/agent_manager.hpp"
#include "core/common.hpp"
#include "core/components/fwd.hpp"
#include "core/config.hpp"
#include "core/debug.hpp"
@@ -167,7 +168,141 @@ get_enabled(tim::type_list<Tp...>)
(_v.set(_n++, trait::runtime_enabled<Tp>::get()), ...);
return _v;
}
void
rocpd_init_categories()
{
static bool _is_initialized = false;
if(_is_initialized) return;
auto& data_processor = get_data_processor();
data_processor.insert_string(trait::name<category::thread_cpu_time>::value);
data_processor.insert_string(trait::name<category::thread_peak_memory>::value);
data_processor.insert_string(trait::name<category::thread_context_switch>::value);
data_processor.insert_string(trait::name<category::thread_page_fault>::value);
data_processor.insert_string(trait::name<category::thread_hardware_counter>::value);
_is_initialized = true;
}
template <typename Category>
void
rocpd_init_tracks(int64_t _tid)
{
auto& data_processor = get_data_processor();
auto& n_info = node_info::get_instance();
const auto& t_info = thread_info::get(_tid, SequentTID);
auto _tid_name = JOIN("", '[', _tid, ']');
auto thread_idx = data_processor.insert_thread_info(
n_info.id, getppid(), getpid(), t_info->index_data->system_value,
JOIN(" ", "Thread", _tid).c_str(), t_info->get_start(), t_info->get_stop(), "{}");
if constexpr(std::is_same_v<Category, category::thread_hardware_counter>)
{
// Initialize hw_counter_tracks and create one track for each hardware counter
auto _hw_cnt_labels = *get_papi_labels(_tid);
for(auto& itr : _hw_cnt_labels)
{
std::string _desc = tim::papi::get_event_info(itr).short_descr;
if(_desc.empty()) _desc = itr;
ROCPROFSYS_CI_THROW(_desc.empty(), "Empty description for %s\n", itr.c_str());
std::string track_name = JOIN(' ', "Thread", _desc, _tid_name, "(S)");
data_processor.insert_track(track_name.c_str(), n_info.id, getpid(),
thread_idx, "{}");
}
}
else
data_processor.insert_track(
JOIN('_', trait::name<Category>::value, _tid_name).c_str(), n_info.id,
getpid(), thread_idx, "{}");
}
template <typename Category>
void
rocpd_initialize_backtrace_metrics_pmc(size_t dev_id, const char* units, int64_t _tid)
{
auto& data_processor = get_data_processor();
auto _tid_name = JOIN("", '[', _tid, ']');
size_t EVENT_CODE = 0;
size_t INSTANCE_ID = 0;
const char* LONG_DESCRIPTION = "";
const char* COMPONENT = "";
const char* BLOCK = "";
const char* EXPRESSION = "";
auto ni = node_info::get_instance();
const auto TARGET_ARCH = "CPU";
auto& agent_mngr = agent_manager::get_instance();
auto base_id = agent_mngr.get_agent_by_id(dev_id, agent_type::CPU).base_id;
if constexpr(std::is_same_v<Category, category::thread_hardware_counter>)
{
auto _hw_cnt_labels = *get_papi_labels(_tid);
for(auto& itr : _hw_cnt_labels)
{
std::string _desc = tim::papi::get_event_info(itr).short_descr;
if(_desc.empty()) _desc = itr;
ROCPROFSYS_CI_THROW(_desc.empty(), "Empty description for %s\n", itr.c_str());
std::string track_name = JOIN(' ', "Thread", _desc, _tid_name, "(S)");
data_processor.insert_pmc_description(
ni.id, getpid(), base_id, TARGET_ARCH, EVENT_CODE, INSTANCE_ID,
track_name.c_str(), trait::name<Category>::value,
trait::name<Category>::description, LONG_DESCRIPTION, COMPONENT, units,
"ABS", BLOCK, EXPRESSION, 0, 0);
}
}
else
data_processor.insert_pmc_description(
ni.id, getpid(), base_id, TARGET_ARCH, EVENT_CODE, INSTANCE_ID,
JOIN("_", trait::name<Category>::value, _tid_name).c_str(),
trait::name<Category>::value, trait::name<Category>::description,
LONG_DESCRIPTION, COMPONENT, units, "ABS", BLOCK, EXPRESSION, 0, 0);
}
template <typename Category, typename Value>
void
rocpd_process_backtrace_metrics_events(const uint32_t device_id, uint64_t timestamp,
Value value, int64_t _tid)
{
auto& data_processor = get_data_processor();
auto _tid_name = JOIN("", '[', _tid, ']');
auto string_primary_key = data_processor.insert_string(trait::name<Category>::value);
auto event_id = data_processor.insert_event(string_primary_key, 0, 0, 0);
auto& agent_mngr = agent_manager::get_instance();
auto base_id = agent_mngr.get_agent_by_id(device_id, agent_type::CPU).base_id;
auto insert_event_and_sample = [&](const char* _name, double _value) {
data_processor.insert_pmc_event(event_id, base_id, _name, _value);
data_processor.insert_sample(_name, timestamp, event_id);
};
if constexpr(std::is_same_v<Category, category::thread_hardware_counter>)
{
auto _hw_cnt_labels = *get_papi_labels(_tid);
const auto& hw_counters =
static_cast<backtrace_metrics::hw_counter_data_t>(value);
for(size_t i = 0; i < _hw_cnt_labels.size() && i < hw_counters.size(); ++i)
{
std::string _desc = tim::papi::get_event_info(_hw_cnt_labels[i]).short_descr;
if(_desc.empty()) _desc = _hw_cnt_labels[i];
std::string track_name = JOIN(' ', "Thread", _desc, _tid_name, "(S)");
insert_event_and_sample(track_name.c_str(), hw_counters.at(i));
}
}
else
insert_event_and_sample(
JOIN("_", trait::name<Category>::value, _tid_name).c_str(), value);
}
} // namespace
void
backtrace_metrics::sample(int)
{
@@ -337,146 +472,6 @@ backtrace_metrics::fini_perfetto(int64_t _tid, valid_array_t _valid)
}
}
void
rocpd_init_categories()
{
static bool _is_initialized = false;
if(_is_initialized) return;
get_data_processor().insert_category(
category_enum_id<category::thread_cpu_time>::value,
trait::name<category::thread_cpu_time>::value);
get_data_processor().insert_category(
category_enum_id<category::thread_peak_memory>::value,
trait::name<category::thread_peak_memory>::value);
get_data_processor().insert_category(
category_enum_id<category::thread_context_switch>::value,
trait::name<category::thread_context_switch>::value);
get_data_processor().insert_category(
category_enum_id<category::thread_page_fault>::value,
trait::name<category::thread_page_fault>::value);
get_data_processor().insert_category(
category_enum_id<category::thread_hardware_counter>::value,
trait::name<category::thread_hardware_counter>::value);
_is_initialized = true;
}
template <typename Category>
void
rocpd_init_tracks(int64_t _tid)
{
auto& data_processor = get_data_processor();
auto& n_info = node_info::get_instance();
const auto& t_info = thread_info::get(_tid, SequentTID);
auto _tid_name = JOIN("", '[', _tid, ']');
auto thread_idx = data_processor.insert_thread_info(
n_info.id, getppid(), getpid(), t_info->index_data->system_value,
JOIN(" ", "Thread", _tid).c_str(), t_info->get_start(), t_info->get_stop(), "{}");
if constexpr(std::is_same_v<Category, category::thread_hardware_counter>)
{
// Initialize hw_counter_tracks and create one track for each hardware counter
auto _hw_cnt_labels = *get_papi_labels(_tid);
for(auto& itr : _hw_cnt_labels)
{
std::string _desc = tim::papi::get_event_info(itr).short_descr;
if(_desc.empty()) _desc = itr;
ROCPROFSYS_CI_THROW(_desc.empty(), "Empty description for %s\n", itr.c_str());
std::string track_name = JOIN(' ', "Thread", _desc, _tid_name, "(S)");
data_processor.insert_track(track_name.c_str(), n_info.id, getpid(),
thread_idx, "{}");
}
}
else
data_processor.insert_track(
JOIN('_', trait::name<Category>::value, _tid_name).c_str(), n_info.id,
getpid(), thread_idx, "{}");
}
template <typename Category>
void
rocpd_initialize_backtrace_metrics_pmc(size_t dev_id, const char* units, int64_t _tid)
{
auto& data_processor = get_data_processor();
auto _tid_name = JOIN("", '[', _tid, ']');
size_t EVENT_CODE = 0;
size_t INSTANCE_ID = 0;
const char* LONG_DESCRIPTION = "";
const char* COMPONENT = "";
const char* BLOCK = "";
const char* EXPRESSION = "";
auto ni = node_info::get_instance();
const auto* TARGET_ARCH = "CPU";
auto& _agent_manager = agent_manager::get_instance();
auto _base_id = _agent_manager.get_agent_by_id(dev_id, agent_type::CPU).base_id;
if constexpr(std::is_same_v<Category, category::thread_hardware_counter>)
{
auto _hw_cnt_labels = *get_papi_labels(_tid);
for(auto& itr : _hw_cnt_labels)
{
std::string _desc = tim::papi::get_event_info(itr).short_descr;
if(_desc.empty()) _desc = itr;
ROCPROFSYS_CI_THROW(_desc.empty(), "Empty description for %s\n", itr.c_str());
std::string track_name = JOIN(' ', "Thread", _desc, _tid_name, "(S)");
data_processor.insert_pmc_description(
ni.id, getpid(), _base_id, TARGET_ARCH, EVENT_CODE, INSTANCE_ID,
track_name.c_str(), trait::name<Category>::value,
trait::name<Category>::description, LONG_DESCRIPTION, COMPONENT, units,
"ABS", BLOCK, EXPRESSION, 0, 0);
}
}
else
data_processor.insert_pmc_description(
ni.id, getpid(), _base_id, TARGET_ARCH, EVENT_CODE, INSTANCE_ID,
JOIN("_", trait::name<Category>::value, _tid_name).c_str(),
trait::name<Category>::value, trait::name<Category>::description,
LONG_DESCRIPTION, COMPONENT, units, "ABS", BLOCK, EXPRESSION, 0, 0);
}
template <typename Category, typename Value>
void
rocpd_process_backtrace_metrics_events(const uint32_t device_id, uint64_t timestamp,
Value value, int64_t _tid)
{
auto& data_processor = get_data_processor();
auto _tid_name = JOIN("", '[', _tid, ']');
auto event_id =
data_processor.insert_event(category_enum_id<Category>::value, 0, 0, 0);
auto& agent_mngr = agent_manager::get_instance();
auto base_id = agent_mngr.get_agent_by_id(device_id, agent_type::CPU).base_id;
auto insert_event_and_sample = [&](const char* name, double _value) {
data_processor.insert_pmc_event(event_id, base_id, name, _value);
data_processor.insert_sample(name, timestamp, event_id);
};
if constexpr(std::is_same_v<Category, category::thread_hardware_counter>)
{
auto _hw_cnt_labels = *get_papi_labels(_tid);
const auto& _hw_counters =
static_cast<backtrace_metrics::hw_counter_data_t>(value);
for(size_t i = 0; i < _hw_cnt_labels.size() && i < _hw_counters.size(); ++i)
{
std::string _desc = tim::papi::get_event_info(_hw_cnt_labels[i]).short_descr;
if(_desc.empty()) _desc = _hw_cnt_labels[i];
std::string track_name = JOIN(' ', "Thread", _desc, _tid_name, "(S)");
insert_event_and_sample(track_name.c_str(), _hw_counters.at(i));
}
}
else
insert_event_and_sample(
JOIN("_", trait::name<Category>::value, _tid_name).c_str(), value);
}
void
backtrace_metrics::init_rocpd(int64_t _tid, valid_array_t _valid)
{
@@ -641,8 +636,10 @@ backtrace_metrics::post_process_perfetto(int64_t _tid, uint64_t _ts) const
}
void
backtrace_metrics::post_process_rocpd(int64_t _tid, uint64_t _ts) const
backtrace_metrics::post_process_rocpd([[maybe_unused]] int64_t _tid,
[[maybe_unused]] uint64_t _ts) const
{
#if ROCPROFSYS_USE_ROCM > 0
auto is_category_enabled = [&](const auto& _category) { return (*this)(_category); };
if(is_category_enabled(category::thread_cpu_time{}))
@@ -675,6 +672,7 @@ backtrace_metrics::post_process_rocpd(int64_t _tid, uint64_t _ts) const
hw_counter_data_t>(0, _ts, m_hw_counter,
_tid);
}
#endif
}
} // namespace component
} // namespace rocprofsys
@@ -26,7 +26,8 @@
#include "core/config.hpp"
#include "core/node_info.hpp"
#include "core/perfetto.hpp"
#include "core/rocpd/data_processor.hpp"
#include "core/trace_cache/cache_manager.hpp"
#include "core/trace_cache/sample_type.hpp"
#include "library/tracing.hpp"
#include <timemory/units.hpp>
@@ -76,41 +77,25 @@ write_perfetto_counter_track(uint64_t _val)
namespace
{
rocpd::data_processor&
get_data_processor()
{
return rocpd::data_processor::get_instance();
}
void
rocpd_initialize_comm_data_categories()
metadata_initialize_comm_data_categories()
{
static bool _is_initialized = false;
if(_is_initialized) return;
get_data_processor().insert_category(category_enum_id<category::comm_data>::value,
trait::name<category::comm_data>::value);
#if defined(ROCPROFSYS_USE_MPI)
get_data_processor().insert_category(category_enum_id<category::mpi>::value,
trait::name<category::mpi>::value);
#endif
#if defined(ROCPROFSYS_USE_RCCL)
get_data_processor().insert_category(category_enum_id<category::rocm_rccl>::value,
trait::name<category::rocm_rccl>::value);
#endif
trace_cache::get_metadata_registry().add_string(
trait::name<category::comm_data>::value);
trace_cache::get_metadata_registry().add_string(trait::name<category::mpi>::value);
_is_initialized = true;
}
template <typename Track>
void
rocpd_initialize_track()
metadata_initialize_track()
{
auto& n_info = node_info::get_instance();
auto thread_id = std::nullopt;
auto _init_track = [&](const char* label) {
ROCPROFSYS_VERBOSE(3, "INSERT_TRACK label: %s, node ID: %d, Process ID: %d",
label, n_info.id, getpid());
get_data_processor().insert_track(label, n_info.id, getpid(), thread_id);
auto _init_track = [&](const char* label) {
trace_cache::get_metadata_registry().add_track({ label, std::nullopt, "{}" });
};
static std::once_flag _once{};
@@ -118,9 +103,8 @@ rocpd_initialize_track()
}
void
rocpd_initialize_comm_data_pmc()
metadata_initialize_comm_data_pmc()
{
[[maybe_unused]] auto& data_processor = get_data_processor();
// find the proper values for a following definitions
[[maybe_unused]] size_t EVENT_CODE = 0;
[[maybe_unused]] size_t INSTANCE_ID = 0;
@@ -131,56 +115,28 @@ rocpd_initialize_comm_data_pmc()
[[maybe_unused]] constexpr const char* MSG = "bytes";
[[maybe_unused]] constexpr const auto* TARGET_ARCH = "CPU";
auto ni = node_info::get_instance();
constexpr const auto DEVICE_ID = 0; // Assuming CPU device ID is 0
auto& _agent_manager = agent_manager::get_instance();
[[maybe_unused]] auto base_id =
_agent_manager.get_agent_by_id(DEVICE_ID, agent_type::CPU).base_id;
[[maybe_unused]] constexpr const auto DEVICE_ID = 0; // Assuming CPU device ID is 0
#if defined(ROCPROFSYS_USE_MPI)
data_processor.insert_pmc_description(
ni.id, getpid(), base_id, TARGET_ARCH, EVENT_CODE, INSTANCE_ID,
comm_data::mpi_send::label, "Tracks MPI Send communication data sizes",
trait::name<category::mpi>::description, LONG_DESCRIPTION, COMPONENT, MSG, "ABS",
BLOCK, EXPRESSION, 0, 0);
data_processor.insert_pmc_description(
ni.id, getpid(), base_id, TARGET_ARCH, EVENT_CODE, INSTANCE_ID,
comm_data::mpi_recv::label, "Tracks MPI Receive communication data sizes",
trait::name<category::mpi>::description, LONG_DESCRIPTION, COMPONENT, MSG, "ABS",
BLOCK, EXPRESSION, 0, 0);
#endif
#if defined(ROCPROFSYS_USE_RCCL)
data_processor.insert_pmc_description(
ni.id, getpid(), base_id, TARGET_ARCH, EVENT_CODE, INSTANCE_ID, rccl_send::label,
"Tracks RCCL Send communication data sizes",
trait::name<category::rocm_rccl>::description, LONG_DESCRIPTION, COMPONENT, MSG,
"ABS", BLOCK, EXPRESSION, 0, 0);
data_processor.insert_pmc_description(
ni.id, getpid(), base_id, TARGET_ARCH, EVENT_CODE, INSTANCE_ID, rccl_recv::label,
"Tracks RCCL Receive communication data sizes",
trait::name<category::rocm_rccl>::description, LONG_DESCRIPTION, COMPONENT, MSG,
"ABS", BLOCK, EXPRESSION, 0, 0);
trace_cache::get_metadata_registry().add_pmc_info(
{ agent_type::CPU, DEVICE_ID, TARGET_ARCH, EVENT_CODE, INSTANCE_ID,
comm_data::mpi_send::label, "Tracks MPI communication data sizes",
trait::name<category::mpi>::description, LONG_DESCRIPTION, COMPONENT, MSG,
rocprofsys::trace_cache::ABSOLUTE, BLOCK, EXPRESSION, 0, 0 });
trace_cache::get_metadata_registry().add_pmc_info(
{ agent_type::CPU, DEVICE_ID, TARGET_ARCH, EVENT_CODE, INSTANCE_ID,
comm_data::mpi_recv::label, "Tracks MPI communication data sizes",
trait::name<category::mpi>::description, LONG_DESCRIPTION, COMPONENT, MSG,
rocprofsys::trace_cache::ABSOLUTE, BLOCK, EXPRESSION, 0, 0 });
#endif
}
template <typename Track>
void
rocpd_process_cpu_usage_events(const uint32_t device_id, int bytes)
cache_comm_data_events(const uint32_t device_id, int bytes)
{
auto& data_processor = get_data_processor();
auto event_id = data_processor.insert_event(
category_enum_id<category::comm_data>::value, 0, 0, 0);
auto& agents = agent_manager::get_instance();
auto agent = agents.get_agent_by_id(device_id, agent_type::CPU);
auto insert_event_and_sample = [&](const char* name, uint64_t timestamp,
double value) {
data_processor.insert_pmc_event(event_id, agent.device_id, name, value);
data_processor.insert_sample(name, timestamp, event_id);
};
auto agent = agents.get_agent_by_type_index(device_id, agent_type::CPU);
static std::mutex _mutex{};
static uint64_t value = 0;
@@ -190,8 +146,20 @@ rocpd_process_cpu_usage_events(const uint32_t device_id, int bytes)
_now = rocprofsys::tracing::now<uint64_t>();
bytes = (value += bytes);
}
const std::string track_name = Track::label;
const size_t timestamp_ns = _now;
const std::string event_metadata = "{}";
const size_t stack_id = 0;
const size_t parent_stack_id = 0;
const size_t correlation_id = 0;
const std::string call_stack = "{}";
const std::string line_info = "{}";
const size_t agent_handle = agent.handle;
insert_event_and_sample(Track::label, _now, bytes);
trace_cache::get_buffer_storage().store(
trace_cache::entry_type::pmc_event_with_sample, track_name.c_str(), timestamp_ns,
event_metadata.c_str(), stack_id, parent_stack_id, correlation_id,
call_stack.c_str(), line_info.c_str(), agent_handle, track_name.c_str(), value);
}
} // namespace
@@ -199,10 +167,14 @@ rocpd_process_cpu_usage_events(const uint32_t device_id, int bytes)
void
comm_data::start()
{
if(get_use_rocpd())
{
rocpd_initialize_comm_data_categories();
rocpd_initialize_comm_data_pmc();
metadata_initialize_comm_data_categories();
metadata_initialize_comm_data_pmc();
#if defined(ROCPROFSYS_USE_MPI)
metadata_initialize_track<mpi_send>();
metadata_initialize_track<mpi_recv>();
#endif
}
}
@@ -248,10 +220,8 @@ comm_data::audit(const gotcha_data& _data, audit::incoming, const void*, int cou
write_perfetto_counter_track<mpi_send>(count * _size);
if(get_use_rocpd())
{
rocpd_initialize_track<mpi_send>();
rocpd_process_cpu_usage_events<mpi_send>(0, count * _size);
cache_comm_data_events<mpi_send>(0, count * _size);
}
if(rocprofsys::get_use_timemory())
@@ -276,10 +246,8 @@ comm_data::audit(const gotcha_data& _data, audit::incoming, void*, int count,
if(get_use_perfetto()) write_perfetto_counter_track<mpi_recv>(count * _size);
if(get_use_rocpd())
{
rocpd_initialize_track<mpi_recv>();
rocpd_process_cpu_usage_events<mpi_recv>(0, count * _size);
cache_comm_data_events<mpi_recv>(0, count * _size);
}
if(rocprofsys::get_use_timemory())
@@ -304,10 +272,8 @@ comm_data::audit(const gotcha_data& _data, audit::incoming, const void*, int cou
if(get_use_perfetto()) write_perfetto_counter_track<mpi_send>(count * _size);
if(get_use_rocpd())
{
rocpd_initialize_track<mpi_send>();
rocpd_process_cpu_usage_events<mpi_send>(0, count * _size);
cache_comm_data_events<mpi_send>(0, count * _size);
}
if(rocprofsys::get_use_timemory())
@@ -332,10 +298,8 @@ comm_data::audit(const gotcha_data& _data, audit::incoming, void*, int count,
if(get_use_perfetto()) write_perfetto_counter_track<mpi_recv>(count * _size);
if(get_use_rocpd())
{
rocpd_initialize_track<mpi_recv>();
rocpd_process_cpu_usage_events<mpi_recv>(0, count * _size);
cache_comm_data_events<mpi_recv>(0, count * _size);
}
if(rocprofsys::get_use_timemory())
@@ -360,10 +324,8 @@ comm_data::audit(const gotcha_data& _data, audit::incoming, void*, int count,
if(get_use_perfetto()) write_perfetto_counter_track<mpi_send>(count * _size);
if(get_use_rocpd())
{
rocpd_initialize_track<mpi_send>();
rocpd_process_cpu_usage_events<mpi_send>(0, count * _size);
cache_comm_data_events<mpi_send>(0, count * _size);
}
if(rocprofsys::get_use_timemory())
@@ -389,12 +351,9 @@ comm_data::audit(const gotcha_data& _data, audit::incoming, const void*, void*,
write_perfetto_counter_track<mpi_send>(count * _size);
}
if(get_use_rocpd())
{
rocpd_initialize_track<mpi_send>();
rocpd_initialize_track<mpi_recv>();
rocpd_process_cpu_usage_events<mpi_recv>(0, count * _size);
rocpd_process_cpu_usage_events<mpi_send>(0, count * _size);
cache_comm_data_events<mpi_recv>(0, count * _size);
cache_comm_data_events<mpi_send>(0, count * _size);
}
if(rocprofsys::get_use_timemory()) add(_data, count * _size);
@@ -416,10 +375,9 @@ comm_data::audit(const gotcha_data& _data, audit::incoming, const void*, int sen
write_perfetto_counter_track<mpi_recv>(recvcount * _recv_size);
}
if(get_use_rocpd())
{
rocpd_process_cpu_usage_events<mpi_send>(0, sendcount * _send_size);
rocpd_process_cpu_usage_events<mpi_recv>(0, recvcount * _send_size);
cache_comm_data_events<mpi_send>(0, sendcount * _send_size);
cache_comm_data_events<mpi_recv>(0, recvcount * _send_size);
}
if(rocprofsys::get_use_timemory())
@@ -467,10 +425,9 @@ comm_data::audit(const gotcha_data& _data, audit::incoming, const void*, int sen
write_perfetto_counter_track<mpi_recv>(recvcount * _recv_size);
}
if(get_use_rocpd())
{
rocpd_process_cpu_usage_events<mpi_send>(0, sendcount * _send_size);
rocpd_process_cpu_usage_events<mpi_recv>(0, recvcount * _send_size);
cache_comm_data_events<mpi_send>(0, sendcount * _send_size);
cache_comm_data_events<mpi_recv>(0, recvcount * _send_size);
}
if(rocprofsys::get_use_timemory())
@@ -501,10 +458,9 @@ comm_data::audit(const gotcha_data& _data, audit::incoming, const void*, int sen
write_perfetto_counter_track<mpi_recv>(recvcount * _recv_size);
}
if(get_use_rocpd())
{
rocpd_process_cpu_usage_events<mpi_send>(0, sendcount * _send_size);
rocpd_process_cpu_usage_events<mpi_recv>(0, recvcount * _recv_size);
cache_comm_data_events<mpi_send>(0, sendcount * _send_size);
cache_comm_data_events<mpi_recv>(0, recvcount * _recv_size);
}
if(rocprofsys::get_use_timemory())