Releases: ROCm/omnitrace
Releases · ROCm/omnitrace
v1.1.0: Bug Fixes, Strict Configs, Perfetto Label Overhaul
What's Changed
- Standalone build examples + testing workflow updates by @jrmadsen in #15
- CMake updates/fixes + parallel-overhead updates by @jrmadsen in #16
- New documentation page with youtube links to tutorials by @jrmadsen in #23
- Update setup-env.sh and modulefile by @jrmadsen in #26
- Fix perfetto_counter_track string lifetime by @jrmadsen in #28
- Set OMNITRACE_USE_THREAD_SAMPLING=ON for several tests by @jrmadsen in #29
- Fix category regex + new features by @jrmadsen in #25
- Support strict settings option in timemory + expanded config syntax by @jrmadsen in #31
- Rework sampling trace counter names + new trace counters by @jrmadsen in #30
- Fix loop-level instrumentation + more by @jrmadsen in #32
- Fix sampling counter time scales by @jrmadsen in #33
- Implements --label option for python profiler by @jrmadsen in #34
- export libomnitrace-dl.so to OMP_TOOL_LIBRARIES by @jrmadsen in #27
Full Changelog: v1.0.0...v1.1.0
v1.0.0: Initial Release of Omnitrace
Overview
- Please refer to the documentation about the capabilities of omnitrace.
- The binary installers attached to this release only have external dependencies on the ROCm version noted in the script name, support for PAPI, OMPT, etc. are built-in
Full Changelog: v0.0.1...v1.0.0
Instructions for installing binary releases
See the documentation here to determine whether your OS supports installation via the pre-built installation scripts. It is possible to use these scripts on similar Linux flavors, e.g. the Ubuntu 20.04 (focal fossa) installer is compatible with Debian 11 (bullseye/sid).
- Download the binary for your OS and with the desired ROCm compatibility (if any)
- The supported Python versions for all installers in this release are 3.6, 3.7, 3.8, 3.9, and 3.10
- Install the dependencies as needed (if not already installed)
- All binary installers require installing OpenMP, e.g.,
apt-get install libgomp1
on Ubuntu - Packages with ROCm in the name require installing ROCm: instructions can be found here
- PAPI and OMPT support have no runtime dependencies and thus require no additional installations.
- All installations have partial MPI support
- If you do not have Python installed on your system and do not intend to use the Python capabilities, installing Python is not necessary.
- All binary installers require installing OpenMP, e.g.,
- Create the installation directory for omnitrace, e.g.
mkdir /opt/omnitrace
- Run the installer script (see example below)
- Recommendation: use
--exclude-subdir
option
- Recommendation: use
- Setup the environment via
setup-env.sh
or environment-modules
a. Source thesetup-env.sh
script in<prefix>/share/omnitrace
, e.g.source /opt/omnitrace/share/omnitrace/setup-env.sh
b.module use <prefix>/share/modulefiles
andmodule load omnitrace/1.0.0
- Verify
which omnitrace
andwhich omnitrace-avail
return<prefix>/bin/omnitrace
and<prefix>/bin/omnitrace-avail
Example for omnitrace-1.0.0-ubuntu-20.04-ROCm-50000-PAPI-OMPT-Python3.sh
$ mkdir ${HOME}/omnitrace
$ ./omnitrace-1.0.0-ubuntu-20.04-ROCm-50000-PAPI-OMPT-Python3.sh --prefix=/opt/omnitrace --skip-license --exclude-subdir
omnitrace Installer Version: 1.0.0, Copyright (c) Advanced Micro Devices, Inc.
This is a self-extracting archive.
The archive will be extracted to: /opt/omnitrace
Using target directory: /opt/omnitrace
Extracting, please wait...
Unpacking finished successfully
$ source /opt/omnitrace/share/omnitrace/setup-env.sh
$ which omnitrace
/opt/omnitrace/bin/omnitrace
$ which omnitrace-avail
/opt/omnitrace/bin/omnitrace-avail
Enabling CPU Hardware Counters
In order to enable collecting CPU hardware counters, the value of /proc/sys/kernel/perf_event_paranoid
may need to be changed.
The default value is 2. To update /proc/sys/kernel/perf_event_paranoid
run:
echo <VALUE> | sudo tee /proc/sys/kernel/perf_event_paranoid
Value | CPU Hardware Counter Capabilities |
---|---|
-1 | Allow use of (almost) all events by all users. Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK |
>=0 | Disallow ftrace function tracepoint by users without CAP_SYS_ADMIN. Disallow raw tracepoint access by users without CAP_SYS_ADMIN |
>=1 | Disallow CPU event access by users without CAP_SYS_ADMIN |
>=2 | Disallow kernel profiling by users without CAP_SYS_ADMIN |