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 function

enqueue_causal_conv1d_update_gpu

def enqueue_causal_conv1d_update_gpu[dtype: DType, //, separate_state: Bool = False](ctx: DeviceContext, batch: Int, dim: Int, seqlen: Int, width: Int, state_len: Int, x: ImmTileTensor[dtype, Engine=x.Engine, linear_idx_type=x.linear_idx_type], conv_state_in: ImmTileTensor[dtype, Engine=conv_state_in.Engine, linear_idx_type=conv_state_in.linear_idx_type], conv_state: MutTileTensor[dtype, Engine=conv_state.Engine, linear_idx_type=conv_state.linear_idx_type], weight: ImmTileTensor[dtype, Engine=weight.Engine, linear_idx_type=weight.linear_idx_type], output: MutTileTensor[dtype, Engine=output.Engine, linear_idx_type=output.linear_idx_type], bias: ImmTileTensor[dtype, Engine=bias.Engine, linear_idx_type=bias.linear_idx_type], silu_activation: Bool)

Enqueues the functional update, retaining copy-plus-update by default.

The separate-state arm avoids materializing the previous state in the output allocation. Both the registered operation and its benchmark use this launcher so the measured operation matches production dispatch. The default preserves the original in-place kernel ABI and copy-plus-update operation for existing callers and benchmark comparisons. The copy-plus-update arm requires dense state storage. Independent strided state layouts are supported only by the separate-state arm. State views must be disjoint or exactly alias with identical layouts.

Was this page helpful?