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 module
launch_config
NVIDIA launch configurations that carry CUDA launch attributes.
Launch attributes, such as programmatic dependent launch (PDL), are an NVIDIA
performance capability, so they live here rather than on the portable
GridBlockLaunchConfig and ClusterLaunchConfig. On an NVIDIA target,
DeviceContext[].ExecutionConfigType is one of the two configurations below.
Launching one on a target without launch attributes fails to compile.
from max.gpu.host import DeviceContext, Dim
from max.gpu.host.nvidia import ClusterAttributeLaunchConfig
from max.gpu.primitives.grid_controls import PDLLevel, pdl_launch_attributes
def kernel():
pass
with DeviceContext() as ctx:
ctx.enqueue_function[kernel](
ClusterAttributeLaunchConfig(
grid_dim=4,
block_dim=128,
cluster_dim=Dim(2, 1, 1),
shared_mem_bytes=64 * 1024,
attributes=pdl_launch_attributes(PDLLevel.ON),
)
)Structs
-
ClusterAttributeLaunchConfig: A launch configuration with thread block clusters, dynamic shared memory, and CUDA launch attributes. -
GridBlockAttributeLaunchConfig: A launch configuration with grid, block, dynamic shared memory, and CUDA launch attributes.
Traits
-
LaunchAttributeHolderExecutionConfig: AnExecutionConfigthat carries CUDA launch attributes.