[CuTeDSL] Flash Attention v2 for SM120 (Blackwell GeForce)#3030

Open

blake-snc wants to merge 7 commits intoNVIDIA:mainfrom

blake-snc:sm120-flash-attention-v2

Commits on Feb 13, 2026

[CuTeDSL] Add Flash Attention v2 example for SM120 (Blackwell GeForce)

blake-snc
and
claude
committed

Commits on Feb 19, 2026

Add TMA variant with warp specialization for SM120 Flash Attention

blake-snc
and
claude
committed

Commits on Feb 20, 2026

Document SM120 FP8 MMA limitation and clean up kernel entry point

blake-snc
and
claude
committed

Commits on Feb 21, 2026

Add FP8 Flash Attention kernel and benchmark for SM120

blake-snc
and
claude
committed

Commits on Feb 23, 2026

Fix TMA causal masking: skip expensive per-element mask for non-diagonal tiles

blake-snc
and
claude
committed

Commits on Feb 24, 2026

Fix TMA non-causal OOB masking and tighten SMEM capacity check

blake-snc
and
claude
committed

Commits on Feb 26, 2026

Add missing license header and normalize format

blake-snc
and
claude
committed