Skip to content

Pull requests: google/XNNPACK

New pull request New
191 Open 8,849 Closed
191 Open 8,849 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Simplify dot by separating the float and integer paths
#9341 opened Dec 22, 2025 by copybara-service bot Loading…
Simplify dot by separating the float and integer paths
#9340 opened Dec 22, 2025 by copybara-service bot Loading…
For avx512, combine f, bw, vl, dq into one target
#9339 opened Dec 20, 2025 by copybara-service bot Loading…
Upgrade handling of partial loads/stores of avx-sized vectors using avx512 instructions
#9337 opened Dec 20, 2025 by copybara-service bot Loading…
Try to store output of stencil_copy and static_pad at the innermost loop level.
#9336 opened Dec 19, 2025 by copybara-service bot Loading…
1. Adjust AVX reductions to match AVX512
#9331 opened Dec 19, 2025 by copybara-service bot Loading…
Copybara import of the project:
#9309 opened Dec 15, 2025 by copybara-service bot Loading…
Added bf16x8->f32x4x2 convert for arm neon.
#9299 opened Dec 13, 2025 by copybara-service bot Loading…
Update scalar GEMM kernels to use unaligned.h
#9293 opened Dec 12, 2025 by copybara-service bot Loading…
Rewrite transpose and stencil_copy
#9281 opened Dec 9, 2025 by copybara-service bot Loading…
Added convert<bf16x8, f32xXx2> to arm_neon and x86 to decrease intrinsics usage in reduce kernels.
#9277 opened Dec 8, 2025 by copybara-service bot Loading…
SME1 support for int4 GEMM and GEMMV kernels
#9257 opened Dec 5, 2025 by qti-vaiskv • Draft
2
Use unaligned.h loads for qb4w scalar ksum loads
#9227 opened Dec 2, 2025 by GregoryComer Loading…
2
build: add experimental support for GN
#9218 opened Nov 28, 2025 by copybara-service bot Loading…
Force split_n to be at least 64 in dot subgraph.
#9194 opened Nov 21, 2025 by copybara-service bot Loading…
Add Hexagon transpose kernels to YNNPACK
#9177 opened Nov 20, 2025 by copybara-service bot Loading…
Use the XNNPack packing fingerprints to invalidate the weight cache.
#9174 opened Nov 19, 2025 by copybara-service bot Loading…
Replace small constant tensors with static values.
#9149 opened Nov 13, 2025 by copybara-service bot Loading…
Be more strict with elided arithmetic op shapes.
#9139 opened Nov 12, 2025 by copybara-service bot Loading…
FIx logic in xnn_use_nr2 to match the previous behaviour.
#9138 opened Nov 12, 2025 by copybara-service bot Loading…
Internal code change
#9119 opened Nov 10, 2025 by copybara-service bot Loading…
Convert batch-matrix-multiply nodes with 2D weights to fully-connected nodes.
#9114 opened Nov 7, 2025 by copybara-service bot Loading…
Split large fully-connected nodes along k.
#9078 opened Nov 2, 2025 by copybara-service bot Loading…
Clean up unnecessary codepaths from operator-run
#9075 opened Nov 2, 2025 by copybara-service bot Loading…
Detect the number of SMCUs on Linux/Android
#9028 opened Oct 22, 2025 by doviethoa-at-work Loading…
ProTip! Follow long discussions with comments:>50.