Skip to content

Pull requests: google/XNNPACK

New pull request New
254 Open 9,903 Closed
254 Open 9,903 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

size both argmax pooling outputs before returning reallocation
#10471 opened Jun 13, 2026 by aizu-m Contributor Loading…
Fix attention benchmark to be more realistic and test longer sequences
#10469 opened Jun 12, 2026 by copybara-service Bot Loading…
softmax: reject input/output datatype mismatch
#10467 opened Jun 12, 2026 by foodlook Loading…
1
Use backward propagated extents to decide when loops should be fused.
#10465 opened Jun 11, 2026 by copybara-service Bot Loading…
Bounds-check data-derived axes and quantization params in tensor builder
#10464 opened Jun 11, 2026 by sin99xx Contributor Loading…
Fix out-of-bounds read when packing transposed qb4w weights
#10463 opened Jun 11, 2026 by sin99xx Contributor Loading…
Expand FullyConnected ATS tests and update XNNPACK exclusions.
#10462 opened Jun 11, 2026 by copybara-service Bot Loading…
max-pooling: clamp dilated indirection coordinates to avoid index underflow (OOB read)
#10457 opened Jun 11, 2026 by evilgensec Loading…
resize-bilinear: clamp source coordinate in legacy/align-corners mode (OOB read)
#10455 opened Jun 11, 2026 by evilgensec Loading…
x32-unpool: bound the index map before dereferencing the output pointers (OOB write)
#10454 opened Jun 11, 2026 by evilgensec Loading…
Reject convolution reshape when the kernel does not fit the padded input (OOB read)
#10453 opened Jun 11, 2026 by evilgensec Loading…
Enable optimized 4-bit weight packing (qs8_to_qu8_qc4w) on AVX VNNI.
#10443 opened Jun 10, 2026 by copybara-service Bot Loading…
Adds support for dynamic weights/bias, asymmetric quantization, and different weights formats to the FullyConnected Generator.
#10442 opened Jun 10, 2026 by copybara-service Bot Loading…
Refactor ATS output checking to support multiple conformance criteria.
#10440 opened Jun 10, 2026 by copybara-service Bot Loading…
Update Gemma3 example with slicing and layernorm adjustments.
#10431 opened Jun 9, 2026 by copybara-service Bot Loading…
Copybara import of the project:
#10428 opened Jun 9, 2026 by copybara-service Bot Loading…
F32-GEMM use 7x16 for faster dynamic weight packing
#10417 opened Jun 8, 2026 by copybara-service Bot Loading…
Remove fxdiv usages from XNNPACK, keeping it only for pthreadpool
#10410 opened Jun 5, 2026 by copybara-service Bot Loading…
Add Pad Generator to ATS.
#10408 opened Jun 5, 2026 by copybara-service Bot Loading…
Fix sqrt(+inf) evaluates to NaN under -ffast-math due to x * rsqrt(x) approximation
#10407 opened Jun 5, 2026 by Senthil455 Loading…
1
6
Enable SSSE3 GEMM microkernels for qs8_qc2w and qd8_f32_qc2w in XNNPACK.
#10406 opened Jun 5, 2026 by copybara-service Bot Loading…
Add Transpose Generator to ATS.
#10398 opened Jun 4, 2026 by copybara-service Bot Loading…
Add Concatenation Generator to ATS.
#10397 opened Jun 4, 2026 by copybara-service Bot Loading…
[gn] Update DEPS
#10393 opened Jun 4, 2026 by copybara-service Bot Loading…
Add a target in BUILD.gn that allows using the XNNPack export of the Tensor API.
#10391 opened Jun 4, 2026 by copybara-service Bot Loading…
ProTip! Add no:assignee to see everything that’s not assigned.