Skip to content

Pull requests: google/XNNPACK

New pull request New
258 Open 10,325 Closed
258 Open 10,325 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

Add support for SDPA op to YNNPACK delegate with tests and benchmarks
#10912 opened Aug 1, 2026 by copybara-service Bot Loading…
Support folding scalar arithmetic through layout transforms into unary operations.
#10911 opened Aug 1, 2026 by copybara-service Bot Loading…
use valid_batch_size to free zero buffers in xnn_destroy_operator
#10910 opened Jul 31, 2026 by aizu-m Contributor Loading…
space-to-depth: guard output_channels size against integer overflow
#10907 opened Jul 31, 2026 by prasanna8585 Contributor Loading…
[WIP] Implement optimized grouped_dot operator in YNNPACK subgraph.
#10905 opened Jul 31, 2026 by copybara-service Bot Loading…
Change XNNPACK RMSNorm to use Tensor API.
#10904 opened Jul 31, 2026 by copybara-service Bot Loading…
Implement optimized grouped_dot kernel in YNNPACK by reusing standard dot kernels.
#10903 opened Jul 31, 2026 by copybara-service Bot Loading…
deconvolution-nhwc: fix missing null check after xnn_allocate_simd_memory in compute_scale_params_qs8_qc8w
#10902 opened Jul 31, 2026 by destro4evr-rgb Contributor Loading…
QD8 qc2w AVXVNNI use biased 2-bit weights
#10901 opened Jul 31, 2026 by copybara-service Bot Loading…
Fix dimension folding in static transpose.
#10898 opened Jul 30, 2026 by copybara-service Bot Loading…
Disable unused variable warnings in XNNPACK for iOS.
#10893 opened Jul 30, 2026 by copybara-service Bot Loading…
QD8 qc2w SSSE3/AVX2/AVXVNNI use biased 2-bit weights
#10891 opened Jul 30, 2026 by copybara-service Bot Loading…
Add Qd8 bf16 qb4w avx512 gemm
#10888 opened Jul 30, 2026 by JakeStevens Contributor Loading…
Fix null pointer dereference on OOM in subgraph compilation and NCHW convolution create
#10887 opened Jul 30, 2026 by destro4evr-rgb Contributor Loading…
Add SDPA benchmark to YNNPACK delegate and optimize attention model performance scaling.
#10885 opened Jul 30, 2026 by copybara-service Bot Loading…
1. Improves single-core matrix dot product scheduling by tiling matrices according to L2 and L3 proxy cache sizes. Makes the smaller of the A and B matrices resident in L2 while streaming the other matrix in panels.
#10881 opened Jul 29, 2026 by copybara-service Bot Loading…
Introduced Packers/Matmul for QC2W weights and QP8 activations
#10877 opened Jul 29, 2026 by damdoo001-arm Loading…
12
Add f16_dwconv AVX512FP16 kernels
#10874 opened Jul 28, 2026 by rahulchaphalkar Loading…
qd8_qc2w / qs8_qc2w AVX2 and AVX256SKX use int16 accumulators.
#10863 opened Jul 28, 2026 by copybara-service Bot Loading…
If k is split, make the k loop outermost and serial. This should be a no-op since we are not yet splitting k.
#10860 opened Jul 27, 2026 by copybara-service Bot Loading…
Copybara import of the project:
#10859 opened Jul 27, 2026 by copybara-service Bot Loading…
Fix double-free and use-after-free of FP16 static values in XNNPACK runtime creation.
#10832 opened Jul 24, 2026 by copybara-service Bot Loading…
[XLA:CPU] Add vectorized sin and cos intrinsic kernels for codegen.
#10821 opened Jul 23, 2026 by copybara-service Bot Loading…
Add a schedule dumping function for debugging YNN scheduler decisions.
#10799 opened Jul 21, 2026 by copybara-service Bot Loading…
reallocate deconv igemm indirection buffer on output-size change
#10791 opened Jul 20, 2026 by aizu-m Contributor Loading…
ProTip! Mix and match filters to narrow down what you’re looking for.