sycl: simplify bin_bcast_kernel #13383
Open
+121
−232
Conversation
This file contains hidden or 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
github-actions
bot
added
ggml
s-Nick
reviewed
May 12, 2025
s-Nick
reviewed
May 12, 2025
|
|
||
| std::size_t num_dst_elements = static_cast<std::size_t>(ne0) * static_cast<std::size_t>(ne1) * | ||
| static_cast<std::size_t>(ne2) * static_cast<std::size_t>(ne3); | ||
| std::size_t local_range = 256; |
There was a problem hiding this comment.
is there a specific reason for using 256 and not something else like 128?
There was a problem hiding this comment.
No reason, just a high enough default value.
AD2605
force-pushed
the
ad/simplify_binbcast
branch
from
0aa9325 to
6c2091f
Compare
Rbiessy
approved these changes
May 14, 2025
qnixsynapse
approved these changes
May 15, 2025
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Outdated suggestions cannot be applied.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 PR simplifies the
bin-bcastkernel, and adds a special code-path when the all the inputs are contiguous, thus avoiding the need for unnecessary index calculations.The current bin-bcast launches a 3D grid or a 1D grid, and the former often limiting in the number of workitems it can accomodate.
This PR completely flattens the kernel, which also makes it easier to check for contiguous memory accesses, and the separate contiguous path also opens the possibility of vectorization later on, though in my current testing, it did not prove to bring about meaningful difference to performance.
This PR also bring minor but consistent improvement of around 1 tk/s on some models.
Performance compared with the following parameters (with
-mmp 0 -ngl 99 -t 8)Intel Lunar Lake 140V iGPU
Intel Data Center Max 1100