Support for tracing mutex locking (#52)

* Parallel overhead example with locks

* Support tracing mutex locking + more

- support wrapping pthread_mutex_lock
- support wrapping pthread_mutex_unlock
- support wrapping pthread_mutex_trylock
- get_perfetto_combined_traces setting
- OMNITRACE_TRACE_THREAD_LOCKS option
- ThreadState
- critical trace includes queue id
- enabled/disabled settings in timemory
- fix OMNITRACE_TIMEMORY_COMPONENTS
- fix reading config
- fix setting categories
- applied ThreadState::Internal in various places
- utility::get_filled_array
- utility::get_reserved_vector
- utility::get_thread_index
- fork_gotcha messages about forks
- split out some pthread_gotcha functionality into pthread_create_gotcha
- handle queue id in roctracer callbacks

* Update timemory and PTL submodules

* Misc CMake updates

- Includes fix to omnitrace-static-lib{gcc,stdcxx}

* Misc cleanup to pthread_mutex_gotcha and backtrace

* Fix to duplicate field in module_function json

* Improvement to debug messages

* omnitrace-dl and common improvements

- tweak to delimit
- common::ignore message
- common::join quoting of strings
- omnitrace_set_env ignores if inited and active
- omnitrace_set_mpi ignores if inited and active

* nsync for transpose example

* Fix to thread_deleter<void> functor invoke

* Fix thread state and HIP stream enums

[ROCm/rocprofiler-systems commit: b208047741]
This commit is contained in:
Jonathan R. Madsen
2022-05-08 04:40:10 -05:00
committed by GitHub
parent 0094a471fd
commit 0d5f0fb9cf
50 changed files with 1736 additions and 483 deletions
@@ -46,8 +46,7 @@ omnitrace_mpi_set_attr()
return MPI_SUCCESS;
};
static auto _mpi_fini = [](MPI_Comm, int, void*, void*) {
OMNITRACE_CONDITIONAL_BASIC_PRINT(get_debug_env(),
"MPI Comm attribute finalize\n");
OMNITRACE_DEBUG("MPI Comm attribute finalize\n");
if(mpip_index != std::numeric_limits<uint64_t>::max())
comp::deactivate_mpip<tim::component_tuple<omnitrace::component::omnitrace>,
api::omnitrace>(mpip_index);
@@ -83,8 +82,7 @@ mpi_gotcha::configure()
void
mpi_gotcha::audit(const gotcha_data_t& _data, audit::incoming, int*, char***)
{
OMNITRACE_CONDITIONAL_BASIC_PRINT_F(get_debug_env(), "%s(int*, char***)\n",
_data.tool_id.c_str());
OMNITRACE_BASIC_DEBUG_F("%s(int*, char***)\n", _data.tool_id.c_str());
if(get_state() < ::omnitrace::State::Init) set_state(::omnitrace::State::PreInit);
@@ -98,8 +96,7 @@ mpi_gotcha::audit(const gotcha_data_t& _data, audit::incoming, int*, char***)
void
mpi_gotcha::audit(const gotcha_data_t& _data, audit::incoming, int*, char***, int, int*)
{
OMNITRACE_CONDITIONAL_BASIC_PRINT_F(get_debug_env(), "%s(int*, char***, int, int*)\n",
_data.tool_id.c_str());
OMNITRACE_BASIC_DEBUG_F("%s(int*, char***, int, int*)\n", _data.tool_id.c_str());
if(get_state() < ::omnitrace::State::Init) set_state(::omnitrace::State::PreInit);
@@ -113,7 +110,7 @@ mpi_gotcha::audit(const gotcha_data_t& _data, audit::incoming, int*, char***, in
void
mpi_gotcha::audit(const gotcha_data_t& _data, audit::incoming)
{
OMNITRACE_CONDITIONAL_BASIC_PRINT_F(get_debug_env(), "%s()\n", _data.tool_id.c_str());
OMNITRACE_BASIC_DEBUG_F("%s()\n", _data.tool_id.c_str());
if(mpip_index != std::numeric_limits<uint64_t>::max())
comp::deactivate_mpip<tim::component_tuple<omnitrace::component::omnitrace>,
@@ -130,7 +127,7 @@ mpi_gotcha::audit(const gotcha_data_t& _data, audit::incoming)
void
mpi_gotcha::audit(const gotcha_data_t& _data, audit::incoming, comm_t, int* _val)
{
OMNITRACE_CONDITIONAL_BASIC_PRINT_F(get_debug_env(), "%s()\n", _data.tool_id.c_str());
OMNITRACE_BASIC_DEBUG_F("%s()\n", _data.tool_id.c_str());
omnitrace_push_trace_hidden(_data.tool_id.c_str());
if(_data.tool_id == "MPI_Comm_rank")
@@ -151,8 +148,7 @@ mpi_gotcha::audit(const gotcha_data_t& _data, audit::incoming, comm_t, int* _val
void
mpi_gotcha::audit(const gotcha_data_t& _data, audit::outgoing, int _retval)
{
OMNITRACE_CONDITIONAL_BASIC_PRINT_F(get_debug_env(), "%s() returned %i\n",
_data.tool_id.c_str(), (int) _retval);
OMNITRACE_BASIC_DEBUG_F("%s() returned %i\n", _data.tool_id.c_str(), (int) _retval);
if(_retval == tim::mpi::success_v && _data.tool_id.find("MPI_Init") == 0)
{
@@ -164,8 +160,7 @@ mpi_gotcha::audit(const gotcha_data_t& _data, audit::outgoing, int _retval)
// were excluded via a regex expression)
if(get_use_mpip())
{
OMNITRACE_CONDITIONAL_BASIC_PRINT_F(get_debug_env() || get_verbose_env() > 0,
"Activating MPI wrappers...\n");
OMNITRACE_BASIC_VERBOSE_F(2, "Activating MPI wrappers...\n");
// use env vars OMNITRACE_MPIP_PERMIT_LIST and OMNITRACE_MPIP_REJECT_LIST
// to control the gotcha bindings at runtime
@@ -186,13 +181,12 @@ mpi_gotcha::audit(const gotcha_data_t& _data, audit::outgoing, int _retval)
tim::mpi::set_rank(m_rank);
tim::settings::default_process_suffix() = m_rank;
get_perfetto_output_filename().clear();
OMNITRACE_CONDITIONAL_BASIC_PRINT(
get_debug() || get_verbose() > 0, "[pid=%i] MPI rank: %i (%i)\n",
process::get_id(), tim::mpi::rank(), m_rank);
OMNITRACE_BASIC_VERBOSE(0, "[pid=%i] MPI rank: %i (%i)\n",
process::get_id(), tim::mpi::rank(), m_rank);
}
else
{
OMNITRACE_BASIC_PRINT_F("%s() returned %i :: nullptr to rank\n",
OMNITRACE_BASIC_VERBOSE(0, "%s() returned %i :: nullptr to rank\n",
_data.tool_id.c_str(), (int) _retval);
}
}
@@ -202,19 +196,19 @@ mpi_gotcha::audit(const gotcha_data_t& _data, audit::outgoing, int _retval)
{
m_size = std::max<int>(*m_size_ptr, m_size);
tim::mpi::set_size(m_size);
OMNITRACE_CONDITIONAL_BASIC_PRINT(
get_debug() || get_verbose() > 0, "[pid=%i] MPI size: %i (%i)\n",
process::get_id(), tim::mpi::size(), m_size);
OMNITRACE_BASIC_VERBOSE(0, "[pid=%i] MPI size: %i (%i)\n",
process::get_id(), tim::mpi::size(), m_size);
}
else
{
OMNITRACE_BASIC_PRINT_F("%s() returned %i :: nullptr to size\n",
OMNITRACE_BASIC_VERBOSE(0, "%s() returned %i :: nullptr to size\n",
_data.tool_id.c_str(), (int) _retval);
}
}
else
{
OMNITRACE_BASIC_PRINT_F("%s() returned %i :: unexpected function wrapper\n",
OMNITRACE_BASIC_VERBOSE(0,
"%s() returned %i :: unexpected function wrapper\n",
_data.tool_id.c_str(), (int) _retval);
}
}