ファイル
rocm-systems/tests/source/thread-limit.cpp
T
Jonathan R. Madsen 518c83e0f9 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
2023-10-16 18:04:47 -05:00

87 行
2.2 KiB
C++

#include <atomic>
#include <chrono>
#include <cstdio>
#include <cstdlib>
#include <iostream>
#include <mutex>
#include <pthread.h>
#include <random>
#include <ratio>
#include <string>
#include <thread>
#include <vector>
long
fib(long n)
{
return (n < 2) ? n : fib(n - 1) + fib(n - 2);
}
#if !defined(MAX_THREADS)
# define MAX_THREADS 4000
#endif
auto total_duration = std::chrono::duration<long, std::nano>{};
int
main(int argc, char** argv)
{
std::string _name = argv[0];
auto _pos = _name.find_last_of('/');
if(_pos != std::string::npos) _name = _name.substr(_pos + 1);
size_t nthread = 2 * MAX_THREADS;
size_t concurrency = std::thread::hardware_concurrency();
long nfib = 35;
if(argc > 1) nfib = atol(argv[1]);
if(argc > 2) concurrency = atol(argv[2]);
if(argc > 3) nthread = atol(argv[3]);
printf("\n[%s] Threads: %zu\n[%s] concurrency: %zu\n[%s] fibonacci(%li)\n",
_name.c_str(), nthread, _name.c_str(), concurrency, _name.c_str(), nfib);
auto threads = std::vector<std::thread>{};
auto _sync = [_name, &threads]() {
std::this_thread::yield();
for(auto& itr : threads)
itr.join();
threads.clear();
};
threads.reserve(concurrency);
for(size_t i = 0; i < nthread; ++i)
{
if(i > MAX_THREADS - 8)
{
printf("[%s] launching thread %zu (max: %d)...\n", _name.c_str(), i,
MAX_THREADS);
fflush(stdout);
}
threads.emplace_back(
[](auto n) {
auto t0 = std::chrono::steady_clock::now();
n = fib(n);
(void) n;
auto diff = (std::chrono::steady_clock::now() - t0);
static auto _mutex = std::mutex{};
_mutex.lock();
total_duration += diff;
_mutex.unlock();
},
nfib);
if(i % concurrency == (concurrency - 1)) _sync();
}
_sync();
printf("[%s] ... completed with an average of %.3f msec per thread\n", _name.c_str(),
std::chrono::duration_cast<std::chrono::milliseconds>(total_duration).count() *
(1.0 / nthread));
return 0;
}