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

coop_group_size

def coop_group_size(ctx: DeviceContext, batch_size: Int, block_size: Int, n_vec: Int) -> Int

Chooses a resident AMD block group for each row.

Every block must be resident because the barrier spins until all blocks arrive. The launch therefore uses at most one block per compute unit. The timeout catches contention that prevents a group from becoming resident.

Args:

• ​ctx (DeviceContext): Device the launch targets.
• ​batch_size (Int): Number of rows in the launch.
• ​block_size (Int): Threads per block.
• ​n_vec (Int): Vectors in one row.

Returns:

Int: Blocks per row, or 1 when the row should not be split.