profiling: add Task/GC/Inference/Codegen signposts for Tracy #49140
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
topolarity
force-pushed
the
tracy-experimental
branch
from
2ddd16f
to
09def6a
Compare
This includes a few separate enhancements: 1. `growable` is added to allow for fixed-size buffers 2. `ios_t` struct layout is optimized for better packing 3. inline buffer is increased to 83 bytes from 54 This change grows the size of ios_t by 16 bytes (160 -> 172 bytes), but the internal buffer was increased by 29 bytes thanks to re-arranging the fields in the struct to be packed more efficiently.
topolarity
force-pushed
the
tracy-experimental
branch
4 times, most recently
from
8d97522
to
5fd96e1
Compare
gbaraldi
reviewed
Apr 5, 2023
topolarity
force-pushed
the
tracy-experimental
branch
from
8b7adce
to
2cc62d0
Compare
This adds a variety of metadata to send to Tracy during profiling, in particular method instance signatures, the Heap size, and a couple of other small pieces of relevant metadata. It also adds full support for Task execution, tracking these as "Fibers" on the Tracy side. This does incur some overhead until an enhancement is implemented on the Tracy-side to avoid a globally synchronous Fiber event queue, but most of the events that we're tracking now are relatively coarse so it should not be a major issue.
This environmental variable instructs the Julia runtime to wait until it receives a connection from the Tracy profiler before starting. This allows us to get a complete trace, without having to accumulate a large queue of events to send over the wire later.
This functionality is currently only exposed on the master branch of Tracy, but it's quite useful so let's carry the patch to expose it until the next Tracy release.
This is useful for Tracy, which can receive method signature strings over the write. For performance reasons, we want to keep those strings as small as possible, typically less than 80 characters long. This change adds basic "configuration" support for `jl_static_show`, including only a "quiet" parameter for now which can be used to avoid printing DataType parameters.
This replaces the `ENABLE_TIMINGS` #define in `options.h` Since we now support multiple back-ends for timing events, each back-end is expected to be enabled with a `WITH_*` Makefile flag, and then the JL_TIMING API is automatically enabled if any back-end is available. Turning on multiple back-ends at the same time is also supported.
Cap the number of method instances we report to the profiler for a single LLVM_MODULE_FINISH to 10.
topolarity
force-pushed
the
tracy-experimental
branch
from
2cc62d0
to
1526534
Compare
kpamnany
reviewed
Apr 12, 2023
kpamnany
reviewed
Apr 12, 2023
kpamnany
reviewed
Apr 12, 2023
kpamnany
approved these changes
Apr 12, 2023
staticfloat
reviewed
Apr 12, 2023
This enables broadcast messages so that Tracy processes can be listed by the profiler, which is a nice quality-of-life feature.
This commit adds the following environment variables:
- `JULIA_TIMING_SUBSYSTEMS` which can be set to, e.g.,
"+INFERENCE,-GC,METHOD_MATCH" to enable INFERENCE
and METHOD_MATCH and disable GC.
- `JULIA_TIMING_METADATA_PRINT_LIMIT` which defaults
to 10 and determines how many metadata items to add to
a single timing zone before truncating
This commit also includes other miscellaneous changes to incorporate
review feedback.
topolarity
force-pushed
the
tracy-experimental
branch
from
ff7f893
to
8e0cba5
Compare
Great idea - I've added that to my list for the follow-up PR (which is targeting locks, etc.) One other detail that I noticed today is that Any ideas about how we should handle that? |
|
I would just delete that environment variable by adding a mapping to |
We'll eventually want to provide some kind of useful interface to turn this on for a particular pre-compilation, but for now we should just prevent causing pre-compilation to hang forever.
staticfloat
approved these changes
Apr 17, 2023
staticfloat
added
the
status:merge me
|
Great stuff! However, I noticed that this PR doesn't add any documentation for the new features. I think that would be a very valuable addition. |
oscardssmith
added
profiler
and removed
status:merge me
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This adds a variety of metadata to send to Tracy during profiling, in particular method instance signatures, the Heap size, GC Mark/Sweep/Stop-the-World, and a couple of other small pieces of relevant metadata.
It also adds full support for Tasks, tracking these as "Fibers" on the Tracy side:
Check out an example trace here: https://topolarity.github.io/tracy-traces/tracy-experimental/
Loose ends
Before this is ready for final review, I still need to:
TRACY_ON_DEMANDand Tracy's Fiber support