- Sampling duration config values
- OMNITRACE_SAMPLING_DURATION
- OMNITRACE_PROCESS_SAMPLING_DURATION
- Disables sampling after this time (in seconds) has elapsed
- Sampling thread-id config values
- OMNITRACE_SAMPLING_TIDS
- OMNITRACE_SAMPLING_CPUTIME_TIDS
- OMNITRACE_SAMPLING_REALTIME_TIDS
- Allows user to select certain threads for sampling
- Miscellaneous
- Tweaked the finalization verbosity messages
- moved sampling-on-child-threads into runtime.hpp and runtime.cpp
- fixed submodule dyninst header install
[ROCm/rocprofiler-systems commit: e67afd33eb]
## Overview
This is a significant PR which has 3 very notable characteristics:
1. Omnitrace colorizes most of it's logging
2. Completely reworked the sampling
- Samples now record the current instruction pointers instead of strings
- This _dramatically_ decreases the overhead of taking a sample
- The collection of metrics during a sample are split out into another component, enabling that data collection to be disabled -- which decreases the sampling overhead even further
- When both `OMNITRACE_SAMPLING_CPUTIME` and `OMNITRACE_SAMPLING_REALTIME` are ON:
- `OMNITRACE_SAMPLING_CPUTIME_FREQ` and `OMNITRACE_SAMPLING_REALTIME_FREQ` can be used to individually control the sampling frequency
- `OMNITRACE_SAMPLING_CPUTIME_DELAY` and `OMNITRACE_SAMPLING_REALTIME_DELAY` can be used to individually control the delay time before starting
- Now, omnitrace does not start a real-time sampler on the main thread unless `OMNITRACE_SAMPLING_REALTIME` is ON
- In the future, an `OMNITRACE_SAMPLING_TIDS` (and real-time, cpu-time variants) configuration variable(s) will allow you to select which threads will be sampled
3. Files produced by `omnitrace` exe -- `available-instr.txt`, `instrumented-instr.txt`, etc. -- now no longer has `-instr` suffix and are placed in `instrumentation/` subfolder, i.e. `available-instr.txt` -> instrumentation/available.txt`
- This helped de-clutter the output folder
Most of the other edits were reorganization (e.g. internal namespace changes), cleanup, and splitting up functionality.
## Bug Fixes
There is a bug fix with respect to the HSA callbacks which disabled sampling on child threads when an HSA API call was made
## Details
- created thread_info struct for mapping different thread IDs
- reorganized file structure significantly
- added categories.hpp, concepts.hpp
- moved around name trait definitions
- moved all omnitrace components into `omnitrace::component` namespace
- there was a lot of inconsistency b/t using `tim::component` in some places and `omnitrace::component`
- added macros like OMNITRACE_DECLARE_COMPONENT in lieu of TIMEMORY_DECLARE_COMPONENT
- OMNITRACE_CRITICAL_TRACE_NUM_THREADS -> OMNITRACE_THREAD_POOL_SIZE
- roctracer and critical_trace use same thread pool
- critical_trace functions do not lock anymore bc of thread-local TaskGroup
- added `component::local_category_region` to support using `component::category_region` without explicitly passing in name
- removed `component::omnitrace` (unused)
- migrated KokkosP and OMPT to use `component::local_category_region`
- removed `component::user_region` as a result
- migrated omnitrace_{push,pop}_{trace,region}_hidden to use component::category_region
- removed `component::functors` as a result
- migrated some ppdefs
- `api::omnitrace` -> `project::omnitrace`
- `api::(...)` -> `category::(...)`
- improved recording the execution time of threads
- migrated this functionality out of pthread_create_gotcha and into thread_info
- moved mpi_gotcha, fork_gotcha, exit_gotcha, rcclp into omnitrace::component namespace
- split backtrace up into backtrace, backtrace_metrics, backtrace_timestamp components
- sampling.cpp handles setup and post-processing that was formerly in backtrace
- updated logging to use colors
- `OMNITRACE_COLORIZED_LOG` config variable
- updated docs on JSON output from timemory
- instrumentation info in instrumentation subfolder
- added testing for KokkosP entries
- added testing for ompt entries
- add_critical_trace function defined in critical_trace.hpp
- disable push_thread_state and pop_thread_state when thread state is Disabled or Completed
- add comp::page_rss to main bundle
- thread_data supports std::optional instead of std::unique_ptr
- thread_data supports tim::identity<T> to avoid unique_ptr or optional
- tracing::record_thread_start_time()
- tracing::push_timemory and tracing::pop_timemory are templated on CategoryT
- removed anonymous namespace from omnitrace::utility
- sampling backtrace stores instruction pointers instead of strings
- component::category_region updates
- handle disabled thread state
- handle finalized state
- fewer debug messages
- invoke thread_init()
- invoke thread_init_sampling()
- handle push/pop count based on category
- push/pop count only modified when used
- component::cpu_freq
- components/ensure_storage.hpp
- reworked the pthread_create replacement function
- updated parallel-overhead example to report # of times locked
- OMNITRACE_MAX_UNWIND_DEPTH build option
- update timemory submodule
[ROCm/rocprofiler-systems commit: 808ea7dfa7]
* Generic comm_data component
- moved rccl_comm_data to comm_data
- comm_data includes communication data for MPI
* fix timemory include with quotes
* Only support MPI comm data with full MPI support
* Increase timeouts + kill perfetto
* Update timemory submodule
* Fix missing command killall
* set +e in Kill Perfetto workflow step
* Updated MPI example to include MPI_Send and MPI_Recv calls
* Update timemory submodule with storage merge fix
* Perfetto comm data
- tracing::now<T>() function
* Fix timemory header include
[ROCm/rocprofiler-systems commit: 0dd8f52292]
* Enable OMNITRACE_TRACE_THREAD_{RW,SPIN}_LOCKS by default
- updates timemory submodule with updated GOTCHA submodule
- fix to GOTCHA library which defaults to not wrapping dlopen and dlsym prevents deadlock
* Bump version to 1.4.0
[ROCm/rocprofiler-systems commit: 040da3fc6a]
- configure known background threads to start indexing down from TIMEMORY_MAX_THREADS
- invoke sampling::shutdown() instead of just blocking signals where possible
[ROCm/rocprofiler-systems commit: 1343c6722a]
Adds advanced category
- advanced category hides less relevant configuration options
- omnitrace-avail has new '--advanced' option which shows these flags
- increase verbosity level to print issue with reading ppid children
- OMNITRACE_ROCTRACER_HSA_ACTIVITY defaults to ON
- OMNITRACE_ROCTRACER_HSA_API defaults to ON
[ROCm/rocprofiler-systems commit: afa3df8523]
* RPATH to rocprofiler_LIBRARY_DIR for ROCm < v5.2
- until v5.2 only librocprofiler64.so was symlinked in /opt/rocm. Thus linker using SOVERSION caused issues finding librocprofiler64.so.1
* Test ROCm w/ CMAKE_INSTALL_RPATH_USE_LINK_PATH=OFF
* INSTALL_RPATH_USE_LINK_PATH for omnitrace exe
[ROCm/rocprofiler-systems commit: b79ce10fee]
* Initial support for RCCL
* OMNITRACE_USE_RCCLP + sampling tweaks
- also OMNITRACE_SAMPLING_KEEP_INTERNAL option
- minor modifications to sampling to use keep internal option + discard funlockfile
* Update docker and workflows to download RCCL
* Update CPack DEB with rocprofiler dependency
* Rework rccl into library and library/components folder
- add tpls/rccl/rccl/rccl.h
* Fix timemory includes
* rcclp inline definitions when disabled
* Tweaks to ubuntu-focal-external-rocm
- disable ompt
- enable building testing
* Tweaks to ubuntu-focal-external-rocm
- ctest exclude
* Tweak ubuntu-focal.yml
- remove source /.../setup-env.sh, replace with $GITHUB_ENV
* Fix ubuntu-focal-rocm + OMPI + root
* Improved rocm-smi error handling
- Recover from rocm-smi errors
- Disabling rocm-smi after recovering from errors
- Werror in developer mode
- Remove State::DelayedInit
- Add State::Disabled
* formatting
* Fix merge of OMNITRACE_SAMPLING_KEEP_INTERNAL
* Update RCCL include directory
- based on ROCm version we need with <rccl/rccl.h> or <rccl.h>
* RCCL Testing
- updated tests to use configuration files
- many tests generate a configuration file
- tests how have GPU option
- enable ncclCommCount, disable ncclGetVersion
- add testing for RCCLP via rccl-tests
- working directory of tests is PROJECT_BINARY_DIR
- add nccl/rccl functions to get_whole_function_names
- some clang compiler fixes
* Handle RCCL include w/o HIP
* RCCL requires HIP
* Update OMNITRACE_SAMPLING_CPUS for testing
* Update tests/CMakeLists.txt
* Debug settings
* Install MPI even when USE_MPI=OFF
* exclude printf
* skip mpi tests w/o USE_MPI or USE_MPI_HEADERS
* update ubuntu rocm workflow
* Fix configure env step for ubuntu rocm
[ROCm/rocprofiler-systems commit: 45be03906a]
* exit gotcha + remove DelayedInit state + cleanup
- exit gotcha which wraps exit, quick_exit, abort
- minor refactor of mpi gotchas
- removed some redundant code in omnitrace_finalized_hidden
- exclude instrumenting functions starting with dlopen and dlsym
- exclude instrumenting exit, quick_exit, and abort functions
- update timemory submodule with support for new gotcha_invoker with (gotcha_data, <function pointer>, args...)
* Improved rocm_smi error handling
[ROCm/rocprofiler-systems commit: 99da25ea80]
* Fix reliability when KOKKOS_PROFILE_LIBRARY is set in env
- in certain situations, an exe using kokkos may be instrumented
- this will cause libomnitrace to be dlopened via libomnitrace-dl
- if KOKKOS_PROFILE_LIBRARY is set to libomnitrace and not libomnitrace-dl, you will end up with different instances of libomnitrace trying to collect data
* Set OMNITRACE_MAX_THREADS=32 in CI
[ROCm/rocprofiler-systems commit: bcdec188eb]
* Added new tests validating gotcha wrappers
* Update MPI example to use thread
* Tweaks to mpi-flat test and mpi_gotcha
- enabled MPI_Comm_size and MPI_Comm_rank in mpip so disabled them at runtime
- set test to collapse threads and processes
* Tweak to test and example
- mpi test sets GOTCHA_DEBUG=1 in env
- removed checking for MPI_{Comm_dup,Comm_group,Group_incl}
- tweaked tests so pthread_join is where it is expected
[ROCm/rocprofiler-systems commit: b1191b2831]
* tweak starting sampling on main thread
- rename omnitrace::common::utility to omnitrace::common::path
- tracing::thread_init_sampling does not start sampling on main thread
* Cancel sampling itimers + timemory submodule update
- updates timemory submodule with support for cancelling itimer and SIGRTMIN through SIGRTMAX
[ROCm/rocprofiler-systems commit: 400e5078ed]
* fix omnitrace print-* with libraries
* timemory submodule update
* Update workflows to use ./bin/omnitrace instead of ./omnitrace
* cmake format
* update timemory submodule
- fix ODR violations in utility/procfs
* cmake updates
- uniform find_package for all ROCm-based libraries
* tweak transpose example
- throw exception instead of std::exit
* Inspect cmdv name before assuming not exe
- some ELF execs "think" they are libraries so only assume rewrite + simulate + all-functions if filename looks like library
- adds some test for --print-available -- <library>
* Fix _has_lib_prefix when command is < 3
* Updates and reverts to omnitrace exe
- update module_function operator< and operator==
- add function_signature operator<
- refactor module_function ctor
- revert some previous changes w.r.t. simulate and include_unninstr
* Fix source/bin/tests to use same output dir as tests
* cmake format
* Segfault mitigation + refactor + modify function iteration
- refactor module_function ctor to avoid segfaults
- string_t -> std::string
- replace std::string with std::string_view in some places
- get_name(module_t*)
- get_name(procedure_t*)
- disable using both app_modules and app_functions
- new option: --parse-all-modules to iterate over app_modules
- removed some unused code w.r.t. debug info
* Disable module_function address range for uninstrumentable functions
* Disable module_function address range for uninstrumentable functions
* Refactored getting file/line info and init/fini
- use dyninst insertInitCallback and insertFiniCallback if main not found
- fixed all issues with segmentation faults in --simulate --all-functions
* revert changes to Findrocprofiler.cmake
[ROCm/rocprofiler-systems commit: d04cbe862e]
- mpi_gotcha::shutdown() calls mpi_gotcha::update()
- fixes the MPI rank not being available to output files when `OMNITRACE_USE_PID=ON`
[ROCm/rocprofiler-systems commit: 8f8aa43d1a]
- fix reporting units and statistics type for rocm_data_tracker
- use indexes for feature names instead of strings
[ROCm/rocprofiler-systems commit: 7745c12417]
* Improved sampling performance
* Sampling tweaks
- backtrace::get() returns vector of string_view
- further performance improvements
- tweaked _use_label
- wrapped samples in perfetto "samples [omnitrace]" block
- samples in TID=0 in sampling mode are in separate thread row
* Fix empty HW counter desc + category for sampling
- fallback to to metric name if papi event info description not found
- add perfetto sampling category
* Limit the SIGALRM frequency
[ROCm/rocprofiler-systems commit: 0ee32b3755]
- found when using ROCm-enabled OpenMPI with rocHPL
- when wrapping pthread_rwlock_rdlock, pthread_rwlock_wrlock, and pthread_rwlock_unlock, omnitrace has been found to deadlock for some unknown reason
- New configuration variable: OMNITRACE_TRACE_THREAD_RW_LOCKS which defaults to false
[ROCm/rocprofiler-systems commit: e099c84640]
- handle OMNITRACE_ENABLED=OFF by disabling everything
- use set_data to get wrappee in pthread_create_gotcha
- clear roctracer_data storage if roctracer not initialized
[ROCm/rocprofiler-systems commit: f82845388a]
* HIP API perfetto args + updated perfetto categories
- Support for HIP API args field in perfetto
- PERFETTO_CATEGORIES -> OMNITRACE_PERFETTO_CATEGORIES
- Changed perfetto categories for several trace events and trace counters
- migrated several TRACE_EVENT_* to use omnitrace::tracing::{push,pop}_perfetto_ts(...)
* Tweaked category_region to encode the type of args as well as value
- Affects MPI args field in perfetto
* Improved testing in ubuntu-focal.yml
- "Test Install" step sources setup-env.sh
- "Test Install" step tests python support
- "Test Install" step tests reading ~/.omnitrace.cfg
- Avoid installing boost and tbb libs when building from submodule
* validate-perfetto-proto.py accepts -m / --categories
* Remove reference from category_region typeids
* Tweak opensuse action name
* Tweak the "Test Install" Step of ubuntu-focal
[ROCm/rocprofiler-systems commit: 2e1fd5a3c4]
* Rework submodule installation
- use add_subdirectory(... EXCLUDE_FROM_ALL) + explicit installation of deps
- install all library deps to lib/omnitrace
- internal builds of dyninst use libomnitrace-rt for binary rewriting
- support libdyninstAPI_RT not in LD_LIBRARY_PATH when dyninst built internally
* Update ubuntu-focal to test full dyninst install
* Use RelWithDebInfo because Dyninst segfaults with MinSizeRel
* Fix ubuntu-focal.yml install step
[ROCm/rocprofiler-systems commit: fadcfa36da]
* Config updates
- See PR #69 for details
- change type of OMNITRACE_DL_VERBOSE
- add "deprecated" category to OMNITRACE_ROCM_SMI_DEVICES
- reduce size of perfetto shared memory size hint
- deprecate OMNITRACE_OUTPUT_FILE in favor of OMNITRACE_PERFETTO_FILE
- set papi event choices
- read config file after reading command line
- fix update of OMNITRACE_DL_VERBOSE
- mark several settings as hidden
- timemory update support hidden attribute for settings
- rework get_perfetto_output_filename()
- Hide settings from not available backends
* Rework omnitrace-avail to support dumping configurations
* Overwrite query, tests, output flag
- Support using -O flag when dumping config
- Support checking before overwriting existing config
- Support --force to overwrite existing config
- Fix get_component_info not including omnitrace components
- Testing for dumping config
* Update documentation on omnitrace-avail
* Fix issue with timemory format + "/__w/"
* Update output prefix keys docs
* Rename --dump-config to --generate-config
* Hide MPI related options
- OMNITRACE_PERFETTO_COMBINE_TRACES and OMNITRACE_COLLAPSE_PROCESSES are hidden w/o MPI support
[ROCm/rocprofiler-systems commit: 1877ebf47b]
* Fix to pid via mpi_gotcha
* OMNITRACE_VERSION defines
* call perfetto on hsa_activity_callback thread
* Test label tweak
* Config fixes
- Change type of OMNITRACE_DL_VERBOSE
- Update OMNITRACE_DL_VERBOSE properly
- Add OMNITRACE_ROCM_SMI_DEVICES to deprecated group
- Set default_process_suffix
* metadata for OMNITRACE_VERSION and OMNITRACE_HIP_VERSION
[ROCm/rocprofiler-systems commit: efe1edd253]
tracing NS + category region component
- made library.cpp impl more broadly available
- support for perfetto args
- MPI wrappers encode args and return type
- new categories / perfetto categories
- omnitrace_library category -> libomnitrace
- omnitrace_dl_library -> libomnitrace-dl
[ROCm/rocprofiler-systems commit: 5105e2c94f]
* Remove reliance on MPI_Comm_rank
- read /proc/<PID>/tasks/<PID>/children of parent process to deduce the rank
- Old format relied on user calling MPI_Comm_rank(MPI_COMM_WORLD, ...)
- if MPI_Comm_rank called with subcommunicators only, multiple ranks would write to same file
* Tweak mpi example
[ROCm/rocprofiler-systems commit: 8eff363ed3]
Jonathan R. Madsen
