Sync with Microsoft ONNX Runtime - 16052026#1091

Open

ai-fw-intg wants to merge 12 commits into

ovep-developfrom

sync_msft_16052026

Commits on May 14, 2026

fix(security): add SafeInt overflow protection in Expand and constant folding output size limit (#28055)
tianleiwu
authored
Improve error reporting for pre-allocated outputs with a wrong shape (#28481)
yuslepukhin
authored
[MLAS] Add an NHWC implementation of convolution to avoid transposes (#26834)

orlmon01
and
Copilot
authored
enable dynamic max_k_step in FA for nvidia (#28511)
guschmue
authored
Validate per-column weight_scale/weight_zero_point shape in CPU QAttention; harden integer arithmetic in QAttention and AttentionBase (#28480)

authored
Fill LSTM CUDA operator opset gap: extend coverage from opset 14 to opset 22 (#27737)

Copilot
and
tianleiwu
authored

Commits on May 15, 2026