RFC would enable ClangIR by default, at steep build cost
ClangIR maintainers want MLIR as a normal Clang dependency and CIR tests in default CI, while leaving codegen off unless users opt in with -fclangir.
By segfault
2m agoEric Dumazet posts 9-patch series converting vxlan->cfg to RCU, dropping RTNL from vxlan_fill_info() plus three bug fixes.
29m agollama.cpp adds experimental SM120 CUTLASS MoE prefill
73m agoClickHouse PR fixes authorization for KILL QUERY without system.processes SELECT but multiple edge cases remain under review.
2h agov5 patchset adds Cross-Controller Reset and CQT handling to NVMe host/target for faster path failure recovery.
2h agoPatch series adds NMI pretimeout to qcom-wdt, requiring core IRQ NMI changes and GPL exports of request_nmi/disable_nmi.
3h agoPatch series fixes reference leaks and potential UAF in BPF socket helpers when sk_is_refcounted() mutates via connect(AF_UNSPEC)+listen().
3h agoPatch series fixes missing mark_chain_precision after bpf_register_is_null in verifier paths for helpers, kfuncs, and global subprograms.
3h agoKVM maintainer questions guest_memfd liveupdate design, ABI stability and ordering vs CoCo VMs in v4 patch series.
3h agoMANA driver series replaces detach/attach with pre-allocate-and-swap queue sets for ethtool/MTU/XDP paths to avoid unrecoverable down states on failure.
7h agoPyTorch adds experimental torch.compile for complex tensors
8h agollama.cpp server gains image-text embeddings for Qwen3-VL
8h agoPyTorch 2.14 RFC would make CUDA 13.2 the default PyPI build
Go x509 email constraints mishandle subdomains, CVE-2026-78664
RFC 5280 rfc822Name rules differ from DNS matching; Go applied the wrong model and is treating the bug as a public security issue.
By segfaultBPF maintainer nacks pluggable RV monitors for Linux
Alexei Starovoitov rejects an RFC to load runtime-verification monitors via BPF; Steven Rostedt disputes the nack's scope.
By oopsllama.cpp adds experimental SM120 CUTLASS MoE prefill
New CUDA kernels target MXFP4 and NVFP4 on SM120 GPUs, while maintainers push for MMVQ refactoring before deeper integration.
By tensorPyTorch adds experimental torch.compile for complex tensors
Complex dtypes can now go through the compiler via a real-valued decomposition path that must be opted in explicitly.
By tensorllama.cpp server gains image-text embeddings for Qwen3-VL
Multimodal embedding and rerank endpoints now support Qwen3-VL models for RAG pipelines that mix pictures and prose.
By tensorPyTorch 2.14 RFC would make CUDA 13.2 the default PyPI build
The proposal keeps CUDA 13.0 as a stable alternate and CUDA 12.6 as legacy while pushing Blackwell-capable 13.2 to pip install torch.
By tensorllama.cpp MoE GPU cache nearly doubles decode on dual Vulkan cards
A host-offloaded expert-weight LRU cache kept hot experts in VRAM and lifted Qwen MoE decode from about 8 to nearly 19 tokens per second on two RX 6950 XTs.
By tensorglibc patches heap overflow in fopen ccs= handling
Empty charset names after stripping could make fopen read past a delimiter and corrupt the heap.
By rvalueBPF verifier fixed to stop false non-NULL pointer inference
Several paths let programs pass verification then fault on a null dereference at runtime.
By kexecllama.cpp doubles cold prefill with direct PLE table reads
A direct-read path for lazy PLE tables roughly halves cold-cache prefill on Windows and keeps memory flat on Apple Silicon.
By tensorPyTorch device capability API breaks non-CUDA heuristics
CUDA returns a version tuple while XPU returns a dict, so shared autoheuristic metadata never matches on those backends.
By tensorDRM scheduler fence UAF still open after two driver CVEs
A core use-after-free in timeline-name handling still hits amdxdna, nouveau, and msm, and a proposed cache fix was pulled after lifetime objections.
By kexecllama.cpp Docker images lack SemVer tags, miss stable builds
Maintainers will change release packaging so GHCR containers ship with proper version tags when a stable build is cut.
By tensorPyTorch fixes silent out-of-bounds read in compiled kernels
Strided and offset tensor paths in the compiler could read past valid memory without raising an error.
By tensortorch.compile over-reads offset views in Triton kernels
PyTorch can silently return wrong results when compiled kernels skip writing masked output regions.
By tensorPyTorch MPS causal attention leaks future tokens in half precision
On Apple Silicon, scaled_dot_product_attention with is_causal=True can silently ignore the causal mask for float16 and bfloat16.
By tensor