onednn
https://github.com/oneapi-src/onednn
C++
oneAPI Deep Neural Network Library (oneDNN)
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
C++ not yet supported2 Subscribers
Add a CodeTriage badge to onednn
Help out
- Issues
- x64 JIT binary-select post-op treats negative s8 conditions as false
- [x64][matmul] AMX brgemm kernel is 14-20% slower than avx512_core_bf16 for M=1 bf16 x s8 (weight decompression); small-shape fallback does not cover this case
- cpu: aarch64: Preserve precomputed input in BRGEMM LSTM
- AArch64: Add NHWC KleidiAI conv implementations
- common: memory_desc: bound metadata index in query_v2
- [Release] oneDNN provide the built binary package in release: https://github.com/uxlfoundation/oneDNN/releases
- cpu: x64: f16 depthwise 1xK convolution hangs creating brdgmm_dw:avx10_1_512 (jit_brdgmm_kernel_t::generate never returns)
- [CPU][AMX] `brg_conv_fwd:avx10_1_512_amx_fp16` returns wrong results for f16 forward convolution (nhwc, 16x16 kernel, stride 16, IC 6/8/16)
- [WIP] ci: aarch64: Fix Windows Build
- cpu: aarch64: add NEON bf16 gemm
- Docs
- C++ not yet supported