[SYCL][Graph] Add specification for kernel binary updates #14896
+493
−80
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
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 10, 2024
Implement Dynamic Command-Group feature for intel#14896 to enable updating `ur_kernel_handle_t` objects in graph nodes between executions.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 10, 2024
Implement Dynamic Command-Group feature for intel#14896 to enable updating `ur_kernel_handle_t` objects in graph nodes between executions.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 11, 2024
Implement Dynamic Command-Group feature for intel#14896 to enable updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and nd-range of node.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 11, 2024
Implement Dynamic Command-Group feature for intel#14896 to enable updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and nd-range of node.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 14, 2024
Implement Dynamic Command-Group feature for intel#14896 to enable updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and nd-range of node.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 14, 2024
Implement Dynamic Command-Group feature for intel#14896 to enable updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and nd-range of node.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 14, 2024
Implement Dynamic Command-Group feature for intel#14896 to enable updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and nd-range of node.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 15, 2024
Implement Dynamic Command-Group feature for intel#14896 to enable updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and nd-range of node.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 15, 2024
Implement Dynamic Command-Group feature for intel#14896 to enable updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and nd-range of node.
EwanC
force-pushed
the
fabio/kernel_binary_update
branch
from
4637510
to
fcf08a6
Compare
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 16, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 17, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 17, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 18, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 18, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 21, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 22, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 22, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 23, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
Adds the kernel binary update feature to the sycl graph specification. This introduces a new dynamic_command_group class which can be used to update the command-group function of a kernel nodes in graphs.
EwanC
force-pushed
the
fabio/kernel_binary_update
branch
from
fcf08a6
to
119ba28
Compare
reble
reviewed
Oct 30, 2024
reble
reviewed
Oct 30, 2024
reble
reviewed
Oct 30, 2024
Co-authored-by: Pablo Reble <[email protected]>
Co-authored-by: Pablo Reble <[email protected]>
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 31, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
EwanC
added a commit
to reble/llvm
that referenced
this pull request
Oct 31, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](intel#14896) This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. This functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels from the E2E tests. The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. See the addition to the design doc for further details on the implementation.
martygrant
pushed a commit
that referenced
this pull request
Nov 8, 2024
Implement Dynamic Command-Group feature specified in PR [[SYCL][Graph] Add specification for kernel binary updates](#14896). This feature enables updating `ur_kernel_handle_t` objects in graph nodes between executions as well as parameters and execution range of nodes. Points to note in this change: * The functionality is currently supported on CUDA & HIP which are used for testing in the new E2E tests. Level Zero support will follow shortly, resulting in the removal of the `XFAIL` labels with tracker number from the E2E tests. * The code for adding nodes to a graph has been refactored to split out verification of edges, and marking memory objects used in a node, as separate helper functions. This allows path for adding a command-group node to do this functions over each CG in the list before creating the node itself. * The `dynamic_parameter_impl` code has also been refactored so the code is shared for updating a dynamic parameter used in both a regular kernel node and a dynamic command-group node. * There is now no need for the `handler::setNDRangeUsed()` API now that graph kernel nodes can update between kernels using `sycl::nd_range` and `sycl::range`. The functionality in this method has be turned into a no-op, however removing the method is an ABI breaking change, so it remains guarded by the `__INTEL_PREVIEW_BREAKING_CHANGES` macro. See the addition to the design doc for further details on the implementation.
EwanC
approved these changes
Nov 28, 2024
hdelan
reviewed
Nov 28, 2024
Bensuo
approved these changes
Nov 28, 2024
Comment on lines
572
to
574
| as active. When a dynamic command-group node is executed, the kernel of the active | ||
| command-group function will be run and all the other command-group functions in | ||
| `cgfList` will be ignored. |
There was a problem hiding this comment.
Slight concern that this reads like the currently active command group will always be the one that is executed (regardless of whether the graph has been updated) rather than the CGF that was active when the graph was finalized (or updated).
Comment on lines
583
to
584
| Dynamic command-groups can only be used to update kernels. Trying to update a command-group | ||
| function that contains other operations will result in an error. |
There was a problem hiding this comment.
To me this reads like: you can have CGFs which are not kernel operations in a dynamic command group, but you can't update them. This error actually happens in the constructor for the dynamic command group, so should this read more like:
Dynamic command-groups can only contain kernel operations. Trying to construct a dynamic command-group with functions that contain other operations will result in an error.
There was a problem hiding this comment.
Updated to use your suggested wording
Comment on lines
911
to
917
| Note that the execution range is tied to the command-group, therefore updating | ||
| the range of a node which uses a dynamic command-group shared by another node | ||
| will also update the execution range of the nodes sharing the dynamic | ||
| command-group. Activating a command-group with `set_active_cgf` to a | ||
| command-group that previously had its execution range updated with | ||
| `node::update_range()` or `node::update_nd_range()` will not reset the execution | ||
| range to the original, but instead use the most recently updated value. |
There was a problem hiding this comment.
Could we mention explicitly that updating the range of a node using a dynamic command-group will only update the range of the currently active CGF? It is sort of implied by talking about the fact it won't be reset, but I think it would be good for clarity to specify it explicitly first.
There was a problem hiding this comment.
Good shout, mentioned this explicitly
hdelan
approved these changes
Nov 28, 2024
sycl/doc/extensions/experimental/sycl_ext_oneapi_graph.asciidoc
Outdated
Show resolved
Hide resolved
Also refine spec wording based on PR feedback
EwanC
force-pushed
the
fabio/kernel_binary_update
branch
from
33ae1f4
to
3edf870
Compare
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.
Adds the kernel binary update feature to the sycl graph specification. This introduces a new dynamic_command_group class which can be used to update the command-group function of a kernel nodes in graphs.
Implemented in: