Skip to content

JuliaPerf/IntelITT.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

IntelITT.jl

Julia interface for the Intel Instrumentation and Tracing APIs

IntelITT.jl provides a high-level Julia interface to the Intel Instrumentation and Tracing Technology APIs APIs. These make it possible to control and enrich the collection of trace data when profiling Julia code with Intel tools such as VTune.

When importing IntelITT.jl, a library that provides stub implementations of the IntelITT APIs will be loaded. In the case your application is being profiled, these stubs will be replaced by an actual implementation. This means that you should always be able to use IntelITT.jl without having to worry about whether you are being profiled or not.

Collection control

When actually using Intel tools to profile your application, it's recommended to start the profiler in paused mode, and use IntelITT.jl to only instrument the parts of your code that you are interested in. This can be done as follows:

using IntelITT

# warm-up, compile code, etc

IntelITT.resume()
# do interesting things here
IntelITT.pause()

# convenience macro for the above:
IntelITT.@collect begin
    # ...
end

Trace instrumentation

IntelITT.jl can also be used to enrich the trace data with custom information. For example, using tasks you can annotate a logical unit of work (e.g., a function call):

dom = Domain("MyApplication")
task_begin(dom, "my task")
# ...
task_end(dom)

# or, using a convenience macro:
IntelITT.@task "my task" begin
    # ...
end

Events can be used to observe when demarcated events occur in your application, or to identify how long it takes to execute demarcated regions of code:

ev = Event("my event")
start(ev)
# ...
stop(ev)

If you need to keep track of a value, you can use counters:

ctr = Counter{Float64}("my counter", "MyApplication")
ctr[] = 0.0
# ...
ctr[] = 1.0

# when using an UInt64 counter, you can increment/decrement it:
ctr = Counter{UInt64}("my counter", "MyApplication")
ctr[] = 0
# ...
increment!(ctr, #=delta=1=#)
decrement!(ctr, #=delta=1=#)

Attaching to launched processes

If you want to attach a profiler to a running process, you need to take extra care, as the profiler will not be able to override the stubs provided by IntelITT.jl. To work around this, you need to specify beforehand which ITT API collector to use. For example, if you know you'll be attaching using VTune, find the collector library that VTune uses:

$ INTEL_LIBITTNOTIFY64=~/intel/oneapi/vtune/latest/lib64/runtime/libittnotify_collector.so \
  julia

Running Julia under VTune

In addition to the instrumentation added using IntelITT.jl, it is also possible to have Julia's JIT compiler emit instrumentation that tools like VTune can use. This requires starting Julia with the environment variable ENABLE_JITPROFILING set to 1. On older versions of Julia, pre-1.9, you also need to re-compile Julia from source with the build option USE_INTEL_JITEVENTS set to 1.

More information, including, e.g., a Intel VTune remote usage example can be found in the Julia On HPC Clusters notes.

Acknowledgments