gpt-oss decode performance optimization#20392

Merged

HaiShaw merged 21 commits intosgl-project:mainfrom

HaiShaw:gpt-oss-aiter-decode

Mar 19, 2026

Commits on Mar 3, 2026

Support sliding window attention in aiter for decode stage (eager mode)
wunhuang
committed
Add unified_attention to support sliding window decode attention
wunhuang
committed
Accuracy pass with cuda-graph enable
wunhuang
committed

Commits on Mar 4, 2026

remove testing code
wunhuang
committed
some minor changes
wunhuang
committed

Commits on Mar 6, 2026

Fix the page size > 1 accuracy issue for unified_attention
wunhuang
committed

Commits on Mar 9, 2026

fix max_kv_len issue
wunhuang
committed
gemm opt
wunhuang
committed

Commits on Mar 10, 2026

fused some elementwise kernel for kv cache store
wunhuang
committed

Commits on Mar 12, 2026

Merge branch 'main' into gpt-oss-aiter-decode
kkHuang-amd
authored

Commits on Mar 13, 2026

Fused kv_indices convert swa_kv_indices elementwise kernels
root
committed

Commits on Mar 17, 2026

Merge branch 'main' into gpt-oss-aiter-decode
HaiShaw
authored

Commits on Mar 18, 2026

Add kernel description and comment to new functions introduced.
root
committed

Commits on Mar 19, 2026

Fix CI errors
root
committed