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 module

matmul_kernels

Implements Hopper SM90 matrix multiplication kernels using TMA and WGMMA.

Provides the HopperMatmulSM90Kernel struct and its shared-memory companion HopperMatmulSM90Kernel_SMem, together with producer-consumer pipeline routines that overlap asynchronous TMA loads with warp-group matrix multiply-accumulate (WGMMA) tensor-core operations. The module supplies standard, split-K, and grouped (MoE) kernel entry points targeting NVIDIA H100 GPUs.

Structs

• ​HopperMatmulSM90Kernel: Hopper SM90 GEMM for NVIDIA H100 GPUs.
• ​HopperMatmulSM90Kernel_SMem: Shared memory layout for Hopper SM90 matrix multiplication kernel.

Functions

• ​find_K_alignment_upto_16B: Find alignment among 1B, 2B, 4B, 16B based on the row's bytes.

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