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

mma

Provides SM100 (Blackwell) MMA operation structs for warp-specialized GEMM kernels using UMMA instructions.

Structs

• ​Major: Selects the major (contiguous) dimension for an MMA operand tile layout.
• ​MmaOpSM100_BlockScaled_SS: SM100 (Blackwell) warp-specialized MMA operation for block-scaled GEMM (MXFP4, MXFP8, NVFP4).
• ​MmaOpSM100_SS: SM100 (Blackwell) warp-specialized MMA operation for standard (non-block-scaled) GEMM.

Functions

• ​max_contiguous_tile_shape: Returns the maximum shape of a tile that's contiguous in memory for mma op. This is used to create TMA descriptor.
• ​smem_descriptor: Creates an MMASmemDescriptorPair for an SM100 MMA operand tile in shared memory.

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