Files
rocm-systems/projects/rocprofiler-systems/source/lib/omnitrace/library/runtime.cpp
T
Jonathan R. Madsen a1b11b94f0 Dynamic expansion of thread data (#294)
* Tests for exceeding OMNITRACE_MAX_THREADS

- tests which exceeds OMNITRACE_MAX_THREADS value for thread creation

* CMake Formatting.cmake update

- include source files in /tests/source directory

* Add unknown-hash= to OMNITRACE_ABORT_FAIL_REGEX

- fail if a timemory hash is not resolved to a name

* Tests for exceeding OMNITRACE_MAX_THREADS

- update

* omnitrace-sample update

- remove env disabling of critical-trace and process-sampling

* core library update

- make_unique in concepts.hpp
- add OMNITRACE_USE_ROCM_SMI to "process_sampling" category
- remove forced disabling of critical-trace in sampling mode
- parentheses for OMNITRACE_PREFER
- use tim::get_hash_id instead of tim::get_combined_hash_id

* core library update (containers)

- added aligned_static_vector.hpp
  - similar to static_vector.hpp but attempts to align to cache line size
- alignment template parameter for stable_vector
- added missing aliases in static_vector
  - consistent with aligned_static_vector aliases

* thread_info update

- track the peak number of threads created
- thread_info::get_peak_num_threads() returns the peak number of threads

* thread_data update

- generic thread_data inherits from base_thread_data
- thread_data reworked to support dynamic expansion
- base_thread_data updated to invoke private_instance() function
- thread_data<optional<T>> uses stable_vector aligned to cache line width
- thread_data<identity<T>> uses stable_vector aligned to cache line width
- thread_data for optional and identity provide private private_instance function + friend to base_thread_data
- component_bundle_cache<T> is now thread_data<component_bundle_cache_impl<T>>

* causal update

- thread_data<T>::instances -> thread_data<T>::instance(construct_on_thread{ ... })
- loop over max_supported_threads (constexpr) -> loop over thread_info::get_peak_num_threads()
- tim::get_combined_hash_id -> tim::get_hash_id
- update progress_bundle usage to new thread_data API

* backtrace/backtrace_metrics component update

- backtrace_metrics update
  - update to new thead_data API
  - add thread CPU time row in perfetto
  - fix potential bug when rusage categories are disabled
  - fix bug in operator-= not subtracting cpu time of rhs
- backtrace update
  - skip all child call-stack below 'tim::openmp::' if sampling_keep_internal = false

* pthread_gotcha component update

- pthread_gotcha::shutdown() invokes pthread_create_gotcha::shutdown()

* pthread_create_gotcha component update

- minor tweak to {start,stop}_bundle functions: pass in thread id
- update to new thread_data API
- track native handles of internal threads
- implement system with pthread_kill to stop dangling bundles

* rocprofiler/roctracer component update

- update to new thread_data API
- loop over max_supported_threads (constexpr) -> loop over thread_info::get_peak_num_threads()

* critical trace (library) update

- update to new thread_data API
- tim::get_combined_hash_id -> tim::get_hash_id

* coverage update

- update to new thread_data API

* tasking update

- update to new thread_data API
- loop over max_supported_threads (constexpr) -> loop over thread_info::get_peak_num_threads()

* roctracer update

- update to new thread_data API
- loop over max_supported_threads (constexpr) -> loop over thread_info::get_peak_num_threads()

* rocm_smi update

- update to new thread_data API

* runtime.cpp update

- update to new thread_data API

* sampling.cpp update

- update to new thread_data API
- loop over max_supported_threads (constexpr) -> loop over thread_info::get_peak_num_threads()

* ompt.cpp update

- invoke pthread_gotcha::shutdown before invoking OMPT finalize function
  - this prevents signals from being delivered to OpenMP threads

* tracing.hpp and tracing.cpp update

- replace get_timemory_hash_{ids,aliases} functions with copy_timemory_hash_ids function
- update to new thread_data API
- loop over max_supported_threads (constexpr) -> loop over thread_info::get_peak_num_threads()
- tim::get_combined_hash_id -> tim::get_hash_id
- improvements to + error checking in thread_init function

* library.cpp update

- move copying timemory hash id/aliases to tracing.cpp
- update to new thread_data API
- loop over max_supported_threads (constexpr) -> loop over thread_info::get_peak_num_threads()

* Update BuildSettings.cmake

- add -Wno-interference-size to suppress warning about use of std::hardware_destructive_interference

* Update fork example

- improve scheme for waiting on child processes via waitpid instead of wait
- support running main routine multiple times
- push/pop regions in child process

* Update lib/common/defines.h.in

- allow use to specify misc values via -D <name>=<value>
  - OMNITRACE_CACHELINE_SIZE
  - OMNITRACE_CACHELINE_SIZE_MIN
  - OMNITRACE_ROCM_MAX_COUNTERS
- remove unused defines
  - OMNITRACE_ROCM_LOOK_AHEAD
  - OMNITRACE_MAX_ROCM_QUEUES

* Update rocprofiler.hpp

- OMNITRACE_MAX_ROCM_COUNTERS -> OMNITRACE_ROCM_MAX_COUNTERS

* Update aligned_static_vector

- set cacheline_align_v from max of OMNITRACE_CACHELINE_SIZE and OMNITRACE_CACHELINE_SIZE_MIN

* Update tracing.cpp

- acquire locks for updating main hash ids/aliases
- only propagate ids/aliases when finalizing

* Update pthread_create_gotcha.cpp

- make sure hash for "start_thread" exists on main thread

* Update causal end to end tests

- if OMNITRACE_BUILD_NUMBER is 1, set OMNITRACE_VERBOSE=0

[ROCm/rocprofiler-systems commit: 518c83e0f9]
2023-10-16 18:04:47 -05:00

282 lines
8.1 KiB
C++

// MIT License
//
// Copyright (c) 2022 Advanced Micro Devices, Inc. All Rights Reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in all
// copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
// SOFTWARE.
#include "library/runtime.hpp"
#include "api.hpp"
#include "core/config.hpp"
#include "core/debug.hpp"
#include "core/defines.hpp"
#include "core/utility.hpp"
#include "library/thread_data.hpp"
#include "library/thread_info.hpp"
#include <timemory/backends/dmp.hpp>
#include <timemory/backends/mpi.hpp>
#include <timemory/backends/process.hpp>
#include <timemory/backends/threading.hpp>
#include <timemory/components/rusage/backends.hpp>
#include <timemory/environment.hpp>
#include <timemory/process/process.hpp>
#include <timemory/sampling/allocator.hpp>
#include <timemory/settings.hpp>
#include <timemory/settings/types.hpp>
#include <timemory/utility/argparse.hpp>
#include <timemory/utility/declaration.hpp>
#include <timemory/utility/signals.hpp>
#include <array>
#include <csignal>
#include <cstdint>
#include <cstdlib>
#include <numeric>
#include <ostream>
#include <string>
#include <unistd.h>
namespace omnitrace
{
namespace
{
auto root_process_id = get_env<pid_t>("OMNITRACE_ROOT_PROCESS", process::get_id(), false);
auto&
get_sampling_on_child_threads_history(int64_t _idx = utility::get_thread_index())
{
static auto _v = utility::get_filled_array<OMNITRACE_MAX_THREADS>(
[]() { return utility::get_reserved_vector<bool>(32); });
if(_idx >= OMNITRACE_MAX_THREADS)
{
static thread_local auto _tl_v = utility::get_reserved_vector<bool>(32);
return _tl_v;
}
return _v.at(_idx);
}
bool&
sampling_on_child_threads()
{
static const auto& _thr_info = thread_info::get();
// if the thread is offset, disable by default
// if the global state is not active or the thread state is not enabled, disable by
// default if there is no history, disable by default (first thread) otherwise,
// inherit the last state
static thread_local bool _v =
(_thr_info) ? !_thr_info->is_offset
: (get_state() != State::Active || get_thread_state() != ThreadState::Enabled)
? false
: (get_sampling_on_child_threads_history().empty()
? false
: get_sampling_on_child_threads_history().back());
return _v;
}
} // namespace
std::atomic<uint64_t>&
get_cpu_cid()
{
static std::atomic<uint64_t> _v{ 0 };
return _v;
}
unique_ptr_t<std::vector<uint64_t>>&
get_cpu_cid_stack(int64_t _tid, int64_t _parent)
{
struct omnitrace_cpu_cid_stack
{};
using init_data_t = thread_data<bool, omnitrace_cpu_cid_stack>;
using thread_data_t = thread_data<std::vector<uint64_t>, omnitrace_cpu_cid_stack>;
auto& _v_tid = thread_data_t::instance(construct_on_thread{ _tid });
auto& _b_tid = init_data_t::instance(construct_on_thread{ _tid }, false);
if(_b_tid && !(*_b_tid))
{
*_b_tid = true;
auto _parent_tid = _parent;
auto& _p_tid = thread_data_t::instance(construct_on_thread{ _parent_tid });
// if tid != parent and there is not a valid pointer for the provided parent
// thread id set it to zero since that will always be valid
if(_tid != _parent_tid && !_p_tid) _parent_tid = 0;
// copy over the thread ids from the parent if tid != parent
if(_tid != _parent_tid) *_v_tid = *_p_tid;
}
return _v_tid;
}
unique_ptr_t<cpu_cid_parent_map_t>&
get_cpu_cid_parents(int64_t _tid)
{
struct omnitrace_cpu_cid_stack
{};
using thread_data_t = thread_data<cpu_cid_parent_map_t, omnitrace_cpu_cid_stack>;
return thread_data_t::instance(construct_on_thread{ _tid }, cpu_cid_parent_map_t{});
}
std::tuple<uint64_t, uint64_t, uint32_t>
create_cpu_cid_entry(int64_t _tid)
{
using tim::auto_lock_t;
OMNITRACE_SCOPED_THREAD_STATE(ThreadState::Internal);
// unique lock for _tid
auto& _mtx = get_cpu_cid_stack_lock(_tid);
auto_lock_t _lk{ _mtx, std::defer_lock };
if(!_lk.owns_lock()) _lk.lock();
int64_t _p_idx = (get_cpu_cid_stack(_tid)->empty()) ? 0 : _tid;
auto& _p_mtx = get_cpu_cid_stack_lock(_p_idx);
auto_lock_t _p_lk{ _p_mtx, std::defer_lock };
if(!_p_lk.owns_lock()) _p_lk.lock();
auto&& _cid = get_cpu_cid()++;
auto&& _parent_cid = get_cpu_cid_stack(_p_idx)->back();
uint32_t&& _depth = get_cpu_cid_stack(_p_idx)->size() - ((_p_idx == _tid) ? 1 : 0);
get_cpu_cid_parents(_tid)->emplace(_cid, std::make_tuple(_parent_cid, _depth));
return std::make_tuple(_cid, _parent_cid, _depth);
}
cpu_cid_pair_t
get_cpu_cid_entry(uint64_t _cid, int64_t _tid)
{
return get_cpu_cid_parents(_tid)->at(_cid);
}
tim::mutex_t&
get_cpu_cid_stack_lock(int64_t _tid)
{
struct cpu_cid_stack_s
{};
return tim::type_mutex<cpu_cid_stack_s, project::omnitrace, max_supported_threads>(
_tid);
}
namespace
{
void
setup_gotchas()
{
static bool _initialized = false;
if(_initialized) return;
_initialized = true;
OMNITRACE_BASIC_DEBUG(
"Configuring gotcha wrapper around fork, MPI_Init, and MPI_Init_thread\n");
component::mpi_gotcha::configure();
component::exit_gotcha::configure();
component::fork_gotcha::configure();
}
} // namespace
std::unique_ptr<main_bundle_t>&
get_main_bundle()
{
static auto _v = []() {
auto _self = RUSAGE_SELF;
std::swap(_self, tim::get_rusage_type());
auto _tmp = std::make_unique<main_bundle_t>(
JOIN('/', "omnitrace/process", process::get_id()),
quirk::config<quirk::auto_start>{});
std::swap(_self, tim::get_rusage_type());
return _tmp;
}();
return _v;
}
std::unique_ptr<init_bundle_t>&
get_init_bundle()
{
static auto _v = std::make_unique<init_bundle_t>(
JOIN('/', "omnitrace/process", process::get_id()));
return _v;
}
std::unique_ptr<preinit_bundle_t>&
get_preinit_bundle()
{
static auto _v =
(setup_gotchas(), std::make_unique<preinit_bundle_t>(
JOIN('/', "omnitrace/process", process::get_id()),
quirk::config<quirk::auto_start>{}));
return _v;
}
bool
sampling_enabled_on_child_threads()
{
return sampling_on_child_threads();
}
bool
push_enable_sampling_on_child_threads(bool _v)
{
bool _last = sampling_on_child_threads();
sampling_on_child_threads() = _v;
auto& _hist = get_sampling_on_child_threads_history();
_hist.emplace_back(_last);
return _last;
}
bool
pop_enable_sampling_on_child_threads()
{
auto& _hist = get_sampling_on_child_threads_history();
if(!_hist.empty())
{
bool _restored = _hist.back();
_hist.pop_back();
sampling_on_child_threads() = _restored;
}
return sampling_on_child_threads();
}
void
set_sampling_on_all_future_threads(bool _v)
{
for(size_t i = 0; i < max_supported_threads; ++i)
get_sampling_on_child_threads_history(i).emplace_back(_v);
}
pid_t
get_root_process_id()
{
return root_process_id;
}
bool
is_root_process()
{
return (root_process_id == process::get_id());
}
bool
is_child_process()
{
return (root_process_id != process::get_id());
}
} // namespace omnitrace