AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove Critical Trace Support (#327)

Browse Source 
* Delete core critical-trace files

* Update docs and README

* Update workflows

* Update testing

* Update cmake

* Remove critical trace usage in source code

* Update source/docs/critical_trace.md

- fix spelling

* Formatting

* Update bin/omnitrace-avail/avail.cpp

- statically allocate shared pointers for timemory manager and hash id/aliases to prevent use-after-free errors
This commit is contained in:
Jonathan R. Madsen
2024-04-23 09:35:44 -05:00
committed by GitHub
parent b81db80926
commit 9499e2f521
45 changed files with 55 additions and 2954 deletions
Download Patch File Download Diff File Expand all files Collapse all files
source/lib/omnitrace/library.cpp
-52
View File
@@ -48,7 +48,6 @@
#include "library/components/pthread_gotcha.hpp"
#include "library/components/rocprofiler.hpp"
#include "library/coverage.hpp"
#include "library/critical_trace.hpp"
#include "library/ompt.hpp"
#include "library/process_sampler.hpp"
#include "library/ptl.hpp"
@@ -203,9 +202,6 @@ ensure_finalization(bool _static_init = false)
return scope::destructor{ []() { omnitrace_finalize_hidden(); } };
}
using Device = critical_trace::Device;
using Phase = critical_trace::Phase;
template <typename... Tp>
struct fini_bundle
{
@@ -402,11 +398,6 @@ omnitrace_init_library_hidden()
if(_debug_init) config::set_setting_value("OMNITRACE_DEBUG", _debug_value);
} };
// below will effectively do:
// get_cpu_cid_stack(0)->emplace_back(-1);
// plus query some env variables
add_critical_trace<Device::CPU, Phase::NONE>(0, -1, 0, 0, 0, 0, 0, 0, 0, -1, 0);
tim::trait::runtime_enabled<comp::roctracer>::set(get_use_roctracer());
tim::trait::runtime_enabled<comp::roctracer_data>::set(get_use_roctracer() &&
get_use_timemory());
@@ -920,55 +911,12 @@ omnitrace_finalize_hidden(void)
causal::finish_experimenting();
}
if(get_use_critical_trace() || (get_use_rocm_smi() && get_use_roctracer()))
{
OMNITRACE_VERBOSE_F(1, "Generating the critical trace...\n");
for(size_t i = 0; i < thread_info::get_peak_num_threads(); ++i)
{
using critical_trace_hash_data =
thread_data<critical_trace::hash_ids, critical_trace::id>;
if(i < critical_trace_hash_data::get()->size() &&
critical_trace_hash_data::get()->at(i))
{
OMNITRACE_DEBUG_F("Copying the hash id data for thread %zu...\n", i);
critical_trace::add_hash_id(*critical_trace_hash_data::get()->at(i));
}
}
for(size_t i = 0; i < thread_info::get_peak_num_threads(); ++i)
{
using critical_trace_chain_data = thread_data<critical_trace::call_chain>;
if(i < critical_trace_chain_data::get()->size() &&
critical_trace_chain_data::get()->at(i))
{
OMNITRACE_DEBUG_F(
"Updating the critical trace call-chains for thread %zu...\n", i);
critical_trace::update(i); // launch update task
}
}
OMNITRACE_VERBOSE_F(1, "Waiting on critical trace updates...\n");
tasking::join();
}
if(get_use_process_sampling())
{
OMNITRACE_VERBOSE_F(1, "Post-processing the system-level samples...\n");
process_sampler::post_process();
}
if(get_use_critical_trace())
{
// launch compute task
OMNITRACE_VERBOSE_F(1, "Launching critical trace compute task...\n");
critical_trace::compute();
OMNITRACE_VERBOSE_F(1, "Waiting on critical trace computation...\n");
tasking::join();
}
// shutdown tasking before timemory is finalized, especially the roctracer thread-pool
OMNITRACE_VERBOSE_F(1, "Shutting down thread-pools...\n");
tasking::shutdown();