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

mxfp4_grouped_matmul_amd

Grouped MXFP4 matmul kernels for AMD CDNA4 GPUs.

Provides MoE expert-dispatched grouped matmul in two variants: the native path (mxfp4_grouped_matmul_amd) with on-the-fly B layout handling, and the pre-shuffled-B path (mxfp4_grouped_matmul_amd_preb / PreShuffledBGroupedGEMM) where weights are pre-arranged into a layout that enables coalesced shared-memory reads and direct MFMA consumption.

Structs

• ​PreShuffledBGroupedGEMM: Grouped GEMM for MXFP4 on AMD CDNA4 with pre-shuffled weights.

Functions

• ​mxfp4_grouped_matmul_amd: Launch native MXFP4 grouped matmul on AMD CDNA4.
• ​mxfp4_grouped_matmul_amd_kernel: MXFP4 grouped matmul kernel with expert dispatch via block_idx.z.
• ​mxfp4_grouped_matmul_amd_preb: Launches grouped MXFP4 matmul on AMD CDNA4 with pre-shuffled weights.

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