IMPORTANT: To view this page as Markdown, append `.md` to the URL (e.g. /get-started.md). For the complete documentation index, see llms.txt.
Skip to main content
For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Mojo struct

Struct_matmul_dynamic_block_scaled_mxfp6

struct Struct_matmul_dynamic_block_scaled_mxfp6[FP6_FORMAT: Int = Int(0), preshuffled_b: Bool = False]

Registers the matmul.dynamic.block.scaled.mxfp6 graph op.

Separate from the MXFP4/MXFP8 op rather than another lane_bytes value: both FP6 encodings put 24 bytes in a lane, so the byte count cannot choose between them.

Parameters​

  • ​FP6_FORMAT (Int): 0 selects E2M3, 1 selects E3M2, matching FP6Format.
  • ​preshuffled_b (Bool): When True, b and b_scales must already be in the plane-split / packed-scale layouts from Shuffler.preshuffle_b_planes / preshuffle_scale_4d (a one-time cost for the static weight, which the graph applies with preshuffle_block_scaled_b). Ignored (falls back to the row-major dispatch) when M is within the decode range -- see mxfp6_block_scaled_matmul_amd.

Implemented traits​

AnyType, Deinitable, Movable

Methods​

execute​

static def execute(c: T, a: T, b: T, a_scales: T, b_scales: T, context: DeviceContext)

Was this page helpful?