tvm_tirx.backend.cuda.launch.LaunchConfig

tvm_tirx.backend.cuda.launch.LaunchConfig#

class tvm_tirx.backend.cuda.launch.LaunchConfig(*, grid, block, cluster=None, preferred_cluster=None, dynamic_smem_bytes=None, stream=None, cooperative=None, programmatic_stream_serialization=None, cluster_scheduling_policy=None, priority=None, mem_sync_domain=None, mem_sync_domain_map=None, access_policy_window=None, preferred_shared_memory_carveout=None, nvlink_util_centric_scheduling=None, portable_cluster_size_mode=None, shared_memory_mode=None, programmatic_event=None, launch_completion_event=None)#

Bases: Config

CUDA LaunchConfig configuration. See the generated launch reference.

Parameters:

  • grid (Any)

  • block (Any)

  • cluster (Any)

  • preferred_cluster (Any)

  • dynamic_smem_bytes (Any)

  • stream (Any)

  • cooperative (Any)

  • programmatic_stream_serialization (Any)

  • cluster_scheduling_policy (Any)

  • priority (Any)

  • mem_sync_domain (Any)

  • mem_sync_domain_map (Any)

  • access_policy_window (Any)

  • preferred_shared_memory_carveout (Any)

  • nvlink_util_centric_scheduling (Any)

  • portable_cluster_size_mode (Any)

  • shared_memory_mode (Any)

  • programmatic_event (Any)

  • launch_completion_event (Any)

grid: Any#
block: Any#
cluster: Any = None#
preferred_cluster: Any = None#
dynamic_smem_bytes: Any = None#
stream: Any = None#
cooperative: Any = None#
programmatic_stream_serialization: Any = None#
cluster_scheduling_policy: Any = None#
priority: Any = None#
mem_sync_domain: Any = None#
mem_sync_domain_map: Any = None#
access_policy_window: Any = None#
preferred_shared_memory_carveout: Any = None#
portable_cluster_size_mode: Any = None#
shared_memory_mode: Any = None#
programmatic_event: Any = None#
launch_completion_event: Any = None#