[SYCL][XPTI] Improvements to allow framework/app software level layers to provide code locations for sycl generated XPTI events. #15190
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
It is useful for framework software layers which uses sycl in their implementation (like IPEX) to provide framework level code location information for XPTI events generated by sycl. This allows a framework specific instrumentation tool to capture sycl xpti events with code location information coming from the framework level. This commit change all sycl entry points which tries to capture code location in TLS to query the TLS and use the code location that was already set in the TLS (if any) rather then using the code location tried to be set. framework level software can set code location in TLS by instatiating the existing sycl::detail::tls_code_loc_t object. Signed-off-by: Guy Zadicario <guy.zadicario@intel.com>
…payload. When user has captured a code location in TLS before calling queue.submit, use the captured code location function name if exist instead the kernel name for the command group event payload. Kernel name is still added as metadata, only the payload changes. If user did not capture code location in TLS then use the kernel name for backward compatability. Additional version for queue.submit_impl functions added in order to propegate this state without breaking ABI. Signed-off-by: Guy Zadicario <guy.zadicario@intel.com>
When graph is finalized kernels can be directly enqueued to a command buffer without using the scheduler, in this case XPTI events for the node_create/task_begin/task_end were missing. Save code location from tls when CGF is added to a graph through graph.add API. This captures user stored code location in TLS, if one was set before calling graph.add Signed-off-by: Guy Zadicario <guy.zadicario@intel.com>
gzadica1
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
Added small unittest case to check that user code location is captured in node_create event: NodeCreation.QueueParallelForWithUserCodeLoc Signed-off-by: Guy Zadicario <guy.zadicario@intel.com>
EwanC
reviewed
Aug 27, 2024
…ph API. Signed-off-by: Guy Zadicario <guy.zadicario@intel.com>
|
The changes look alright to me, but I'm not as familiar with XPTI. @KseniyaTikhomirova @tovinkere Could you please have a look? |
EwanC
approved these changes
Sep 16, 2024
cppchedy
approved these changes
Sep 16, 2024
gzadica1
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
sergey-semenov
approved these changes
Sep 19, 2024
Co-authored-by: Sergey Semenov <sergey.semenov@intel.com>
gzadica1
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
sergey-semenov
requested changes
Sep 20, 2024
gzadica1
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
gzadica1
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
gzadica1
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
- moved handler::MIsTopCodeLoc to handler_impl
- updated new symbols in sycl_symbols_{linux,windows}.dump
gzadica1
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
gzadica1
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
sergey-semenov
approved these changes
Sep 25, 2024
sergey-semenov
approved these changes
Oct 7, 2024
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.
It is useful for framework software layers which uses sycl in their
implementation (like IPEX) to provide framework level code location
information for XPTI events generated by sycl. This allows a framework
specific instrumentation tool to capture sycl XPTI events with code location
information coming from the framework level.
This allows the specific instrumentation tool, for example, to capture sycl
stream task_begin and task_end events and correlate the specific execution
with the upper layer graph node (or application level name of work) that this
task represents by querying the payload attached to the events.
The change does not require any new APIs or ABI change, to capture a
code location the framework software layer should instantiate the existing
sycl::detail::tls_code_loc_t object before calling a sycl entry point (usually
queue.submit or graph.add).
There are 3 commits in this PR:
code location that is already set in TLS, if one is set. Instead of passing on
the entry point code location at any case.
function name from code location. This changes this behavior in case that the
upper layer software has captured code location in TLS before calling sycl.