feat: [NPP02-5848] sync with Google XNNPACK upstream#1
Merged
Conversation
PiperOrigin-RevId: 931261281
This generalizes the LUT operation to more types, and allows multiple dimensions in the "LUT" (now the input). This should perform well in both the gather dimension 0 case (provided we have an appropriate LUT kernel), and for other dimensions, because `slinky::copy` should boil down to a memcpy of large chunks in most cases. PiperOrigin-RevId: 931267865
- Fix batching for QD8 case - Fix QD8 parameters differing from F32 - Most of the intermediate tensor sizes are wrong. (This doesn't matter because XNNPACK overwrites these with inferred shapes.) PiperOrigin-RevId: 931269422
PiperOrigin-RevId: 931286477
PiperOrigin-RevId: 931300962
…nnected. PiperOrigin-RevId: 931318618
Also switches off AVX-512 assembly kernels for now on this platform. Takes care of another cluster of Chromium failures related to Windows. https://chromium-review.googlesource.com/c/chromium/src/+/7615883 PiperOrigin-RevId: 931424107
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
nota-github/XNNPACKwith currentgoogle/XNNPACKmaster.f0b3a2996d6fff82cd3f004c0b29f5ccb0ab947dso downstream ExecuTorch can point at a fork that includes the upstream inline-LHS IGEMM tile scheduling fix directly.Background
The first ExecuTorch PR moved the XNNPACK submodule to the Nota fork, but pinned a manual backport commit on the old XNNPACK base. This PR makes the fork match Option 2: use the Nota fork, synchronized with latest Google upstream.
The upstream fix needed for NPP02-5848 is already included in Google XNNPACK as
e0e024ee8a6d9ab6abcac7a876f5d96b32a72074/ PR google#9079.Test plan