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 function
tma_tile_o
def tma_tile_o[dtype: DType, //, swizzle_mode: TensorMapSwizzle, *, BM: Int, BK: Int, depth: Int](ctx: DeviceContext, ptr: Pointer[Scalar[dtype]], rows: Int, out res: TMATensorTile[dtype, Int(3), IndexList(Int(1), BM, BK, __list_literal__=NoneType(None)), _default_desc_shape[Int(3), dtype, IndexList(Int(1), BM, BK, __list_literal__=NoneType(None)), swizzle_mode]()])
Creates the MLA decode output TMA descriptor.
The row axis is its own descriptor dim of extent BM, reached through a separate outer dim with the same row stride. This makes the stored row count a per-copy coordinate, built by store_row_coords. Box row r lands on tensor row row + r, and box rows at or past rows_to_store fall outside the extent and are dropped. The descriptor base sits BM rows before ptr, but masked rows are never written, so memory before ptr is never touched.
Parameters:
- dtype (
DType): Element type of the output tensor (inferred). - swizzle_mode (
TensorMapSwizzle): TMA swizzle mode applied to the descriptor. - BM (
Int): Row extent of the box. - BK (
Int): Tile width in columns for each TMA copy. - depth (
Int): Column count of the full output tensor.
Args:
- ctx (
DeviceContext): Device context used to create the TMA descriptor. - ptr (
Pointer[Scalar[dtype]]): Base pointer of the output tensor in device memory. - rows (
Int): Number of rows in the full output tensor.
Returns: