Mojo module

grid_controls

Grid Dependent Control primitives for NVIDIA Hopper (SM90+) GPUs.

This module provides low-level primitives for managing grid dependencies on NVIDIA Hopper architecture and newer GPUs. It enables efficient orchestration of multi-grid workloads by allowing grids to launch dependent grids and synchronize with them.

The module includes functions that map directly to CUDA grid dependency control instructions, providing fine-grained control over grid execution order:

• launch_dependent_grids(): Triggers execution of grids that depend on the current grid
• wait_on_dependent_grids(): Blocks until all dependent grids complete execution

These primitives are essential for implementing complex GPU execution pipelines where multiple kernels need to execute in a specific order with minimal overhead. They eliminate the need for host-side synchronization when orchestrating dependent GPU work.

Functions

• ​launch_dependent_grids: Launches dependent grids that were previously configured to depend on the current grid.
• ​wait_on_dependent_grids: Waits for all dependent grids launched by this grid to complete execution.

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit