For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo package

mha_depth512

Re-exports the depth-512 multi-head attention kernel package for NVIDIA SM100 GPUs.

Modules

• ​barriers: Barrier infrastructure for depth=256/512 pair-CTA SM100 attention kernels.
• ​config: Configuration for pair-CTA SM100 (Blackwell) MHA kernels (depth 256/512).
• ​correction_warp: Correction warp group logic for depth=256/512 pair-CTA SM100 attention.
• ​dispatch: Dispatch for depth=256/512 pair-CTA SM100 (Blackwell) MHA prefill.
• ​kernel: Kernel entry point for depth=256/512 pair-CTA SM100 (Blackwell) MHA prefill.
• ​load_warp: TMA load warp logic for depth=256/512 pair-CTA SM100 attention.
• ​mma_warp: MMA warp logic for depth=256/512 pair-CTA SM100 attention.
• ​smem: Shared memory layout for depth=512 pair-CTA SM100 attention kernels.
• ​softmax_warp: Softmax warp group logic for depth=256/512 pair-CTA SM100 attention.

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit