Skip to content

Pull requests: NVIDIA/cudnn-frontend

New pull request New
52 Open 365 Closed
52 Open 365 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

Further reduce FROST LA CPU overhead + fully support all L2norm, beta, gate fusion
#616 opened Aug 16, 2026 by jhjpark Collaborator Loading…
3 tasks done
18
frost: fix oversized-SMEM query crashing all frost GEMM on CUDA<13.4 cuda-python
#615 opened Aug 16, 2026 by YangXu1990uiuc Collaborator Loading…
11
frost(sdpa): order every execute-path tensor prep on the launch stream (Rule 5) cat-bugfix mod-cutedsl CuTeDSL kernels, generated kernels, examples, or related integration work. mod-frost orig-nv-eng Reported or requested by NVIDIA engineering.
#614 opened Aug 16, 2026 by vedaanta Collaborator Loading…
3 tasks done
4
First-class cudnn.Handle (create_handle returns an object owning {backend handle, device, stream})
#612 opened Aug 16, 2026 by YangXu1990uiuc Collaborator Loading…
8
FE execute-path host overhead: idempotent set_stream + skip the discarded per-call context
#611 opened Aug 15, 2026 by YangXu1990uiuc Collaborator Loading…
5
Prototype: fused dense SwiGLU-MLP autograd op + cuDNN framework-integration perf guide
#609 opened Aug 15, 2026 by YangXu1990uiuc Collaborator Loading…
11
frost(sdpa): SM120 THD zero-host-read execute — device-built metadata, declared-S_q envelope grid, CUDA-graph capturable (issue #552) cat-enhancements mod-cutedsl CuTeDSL kernels, generated kernels, examples, or related integration work. mod-frost orig-nv-eng Reported or requested by NVIDIA engineering.
#608 opened Aug 15, 2026 by vedaanta Collaborator Loading…
13
NSA compression: suppress aliased O1 store for static short queries
#599 opened Aug 14, 2026 by mgoldfarb-nvidia • Draft
2 of 3 tasks
1
Add cudnn.fla: a cuDNN drop-in for flash-linear-attention (GDN + KDA)
#596 opened Aug 14, 2026 by YangXu1990uiuc Collaborator Loading…
14
Add more features to sm120 frost fp8 sdpa_fwd kernel cat-feature Requests for new functionality, APIs, examples, or behavior improvements. mod-frost orig-nv-eng Reported or requested by NVIDIA engineering.
#595 opened Aug 14, 2026 by Aneureka Member • Draft
3 tasks done
@Aneureka
5
Add SM100 FP8 SDPA support for d192/d128 cat-feature Requests for new functionality, APIs, examples, or behavior improvements. mod-cutedsl CuTeDSL kernels, generated kernels, examples, or related integration work. mod-frost orig-nv-eng Reported or requested by NVIDIA engineering.
#594 opened Aug 14, 2026 by adshen Collaborator Loading…
3 tasks done
20
Avoid zero-initializing JAX grouped GEMM outputs
#592 opened Aug 14, 2026 by mgoldfarb-nvidia • Draft
2 of 3 tasks
1
Drive the grouped GEMM host overhead to the dispatch floor (wrapper 84->~22us; execute() trusts by default)
#591 opened Aug 13, 2026 by YangXu1990uiuc Collaborator Loading…
3 of 4 tasks
8
Memoize the CuTeDSL hot-path dtype/stride conversions (~20% off per-call host)
#589 opened Aug 13, 2026 by YangXu1990uiuc Collaborator Loading…
2 of 3 tasks
1
sdpa fp8: THD/varlen execute + d<=128 envelope (Rubin) for per-tensor FP8 cat-feature Requests for new functionality, APIs, examples, or behavior improvements. mod-cutedsl CuTeDSL kernels, generated kernels, examples, or related integration work. mod-frost orig-nv-eng Reported or requested by NVIDIA engineering.
#587 opened Aug 13, 2026 by vedaanta Collaborator Loading…
9
sdpa fp8 sm100: fused LDTM row-max on cc10.3 (per-tensor FP8) cat-enhancements mod-cutedsl CuTeDSL kernels, generated kernels, examples, or related integration work. mod-frost orig-nv-eng Reported or requested by NVIDIA engineering.
#586 opened Aug 13, 2026 by vedaanta Collaborator Loading…
4
Add LPT_L2 scheduler cat-feature Requests for new functionality, APIs, examples, or behavior improvements. mod-frost orig-nv-eng Reported or requested by NVIDIA engineering.
#585 opened Aug 13, 2026 by yanzhuo607 Collaborator Loading…
2 of 3 tasks
4
Improve large tensor fuzzer plan fallback and failure output
#583 opened Aug 13, 2026 by msalasooNV Collaborator Loading…
2 of 3 tasks
@Anerudhan @chmayer-gpu
3
Emit the FROST gemm parameter tables, and a demo that launches from them
#582 opened Aug 13, 2026 by YangXu1990uiuc Collaborator • Draft
2 of 3 tasks
8
frost(sdpa): dense-padded cu_seq_len support (kernel CU read mode) cat-feature Requests for new functionality, APIs, examples, or behavior improvements. mod-cutedsl CuTeDSL kernels, generated kernels, examples, or related integration work. mod-frost orig-nv-eng Reported or requested by NVIDIA engineering.
#578 opened Aug 13, 2026 by vedaanta Collaborator Loading…
3 tasks done
1
3
frost(sdpa): drop the redundant zero-fill of the never-read THD sinks dummy (SM100) cat-cleanup mod-cutedsl CuTeDSL kernels, generated kernels, examples, or related integration work. mod-frost orig-nv-eng Reported or requested by NVIDIA engineering.
#573 opened Aug 13, 2026 by vedaanta Collaborator Loading…
3
dsa(indexer_backward): validate output & plan signature before the score-grad precompute (default SM100/SM90)
#572 opened Aug 13, 2026 by zkyue Contributor Loading…
1
4
Add SM100 DSA sparse attention forward kernels
#569 opened Aug 13, 2026 by jiayus-nvidia Contributor Loading…
2 of 3 tasks
5
Fix bug in dglu xfilter math
#568 opened Aug 13, 2026 by vasunvidia Loading…
3 tasks
1
DSA: support per-token causal offsets in indexer forward
#561 opened Aug 12, 2026 by kunlunl Contributor • Draft
1 of 3 tasks
1
ProTip! no:milestone will show everything without a milestone.