Runtime and Library Contexts#

Library#

A Library class is an interface that every library descriptor needs to implement. Each library should tell the Legate runtime how to initialize and configure the library, and this class provides a common way to reveal that information to the runtime. Each library should register to the runtime a library descriptor object that implements Library directly or via duck typing. (See legate.core.runtime.Runtime.register_library().)

`Library.get_name`()	Returns a name of the library
`Library.get_shared_library`()	Returns the path to the shared library
`Library.get_c_header`()	Returns a compiled C header string for the library
`Library.get_registration_callback`()	Returns the name of a C registration callback for the library
`Library.get_resource_configuration`()	Returns a ResourceConfig object that configures the library

Resource configuration#

A ResourceConfig object describes the maximum number of handles that a library uses.

`ResourceConfig.max_tasks`
`ResourceConfig.max_reduction_ops`
`ResourceConfig.max_mappers`

Context#

A Context object provides APIs for creating stores and issuing tasks and other kinds of operations. When a library registers itself to the Legate runtime, the runtime gives back a context object unique to the library.

`context.Context.create_store`(ty[, shape, ...])	Creates a fresh store.
`context.Context.create_task`(task_id[, ...])	Creates a task.
`context.Context.create_manual_task`(task_id)	Type safe version of `Context.create_task`.
`context.Context.create_auto_task`(task_id[, ...])	Type safe version of `Context.create_task`.
`context.Context.create_copy`([mapper_id])	Creates a copy operation.
`context.Context.create_fill`(lhs, value[, ...])	Creates a fill operation.
`context.Context.issue_execution_fence`([block])	Issues an execution fence.
`context.Context.tree_reduce`(task_id, store)	Performs a user-defined reduction by building a tree of reduction tasks.
`context.Context.get_tunable`(tunable_id, dtype)	Queries a tunable parameter to the mapper.
`context.Context.provenance`	Returns the current provenance string.
`context.Context.annotation`	Returns the current set of annotations.
`context.Context.set_provenance`(provenance)	Sets a new provenance string
`context.Context.reset_provenance`()	Clears the provenance string that is currently set
`context.Context.push_provenance`(provenance)	Pushes a provenance string to the stack
`context.Context.pop_provenance`()	Pops the provenance string on top the stack
`context.Context.track_provenance`(func[, nested])	Wraps a function with provenance tracking.

Legate Runtime#

`runtime.Runtime.num_cpus`	Returns the total number of CPUs in the system
`runtime.Runtime.num_omps`	Returns the total number of OpenMP processors in the system
`runtime.Runtime.num_gpus`	Returns the total number of GPUs in the system
`runtime.Runtime.register_library`(library)	Registers a library to the runtime.
`runtime.Runtime.create_future`(data, size)	Creates a future from a buffer holding a scalar value.

Annotation#

An Annotation is a context manager to set library specific annotations that are to be attached to operations issued within a scope. A typical usage of Annotation would look like this:

with Annotation(lib_context, { "key1" : "value1", "key2" : "value2", ... }:
  ...

Then each operation in the scope is annotated with the key-value pairs, which are later rendered in execution profiles.

context.Annotation.__init__(context, pairs)

Constructs a new annotation object