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
dispatch_sm100_batched_matmul
def dispatch_sm100_batched_matmul[c_type: DType, a_type: DType, b_type: DType, transpose_b: Bool, pdl_level: PDLLevel = PDLLevel.OVERLAP_AT_BEGINNING](c: TileTensor[c_type, Engine=c.Engine, address_space=c.address_space, linear_idx_type=c.linear_idx_type], a: TileTensor[a_type, Engine=a.Engine, address_space=a.address_space, linear_idx_type=a.linear_idx_type], b: TileTensor[b_type, Engine=b.Engine, address_space=b.address_space, linear_idx_type=b.linear_idx_type], ctx: DeviceContext)
Dispatch batched matmul to SM100 kernel.
First, try to dispatch to a batched matmul config from the tuning table. Then try to find a optimized config for the given shape. If not found, then dispatch to a default config.
Parameters:
- c_type (
DType): Element type of the output tensorc. - a_type (
DType): Element type of the LHS operanda. - b_type (
DType): Element type of the RHS operandb. - transpose_b (
Bool): Whetherbis stored transposed. - pdl_level (
PDLLevel): Programmatic dependent launch level for the dispatched kernel (defaults toPDLLevel.OVERLAP_AT_BEGINNING).
Args:
- c (
TileTensor[c_type, Engine=c.Engine, address_space=c.address_space, linear_idx_type=c.linear_idx_type]): Output batched matrix as a rank-3 mutableTileTensorof shape[batch, M, N]. - a (
TileTensor[a_type, Engine=a.Engine, address_space=a.address_space, linear_idx_type=a.linear_idx_type]): LHS input batched matrix as a rank-3 immutableTileTensorof shape[batch, M, K]. - b (
TileTensor[b_type, Engine=b.Engine, address_space=b.address_space, linear_idx_type=b.linear_idx_type]): RHS input batched matrix as a rank-3 immutableTileTensorof shape[batch, K, N], or[batch, N, K]whentranspose_bis set. - ctx (
DeviceContext): Device context used to enqueue the selected kernel.