Coarrays in the Context of XcalableMP
103
Fig. 1 XMP compiler and an example of coarray program compilation
• Each initialization subroutine xmpf_init_foo is generated from user-defined
procedure foo by the coarray translator, which initializes all static coarrays
declared in foo.
3.2 Allocation and Registration
In order to be accessed using the underlying communication library, the allocated
coarray data must be registered to the library. The registration contains all actions
to allow the data to be accessed from the other nodes, including pin-down memory,
acquirement of the global address, and sharing information among all nodes.
3.2.1 Three Methods of Memory Management
The coarray translator and the runtime library implements three methods of memory
management.
• The runtime sharing (RS) method allocates and registers a large memory for
all static and dynamic coarrays at the initialization phase. The registered memory
is shared by all static and allocatable coarrays.
103
Fig. 1 XMP compiler and an example of coarray program compilation
• Each initialization subroutine xmpf_init_foo is generated from user-defined
procedure foo by the coarray translator, which initializes all static coarrays
declared in foo.
3.2 Allocation and Registration
In order to be accessed using the underlying communication library, the allocated
coarray data must be registered to the library. The registration contains all actions
to allow the data to be accessed from the other nodes, including pin-down memory,
acquirement of the global address, and sharing information among all nodes.
3.2.1 Three Methods of Memory Management
The coarray translator and the runtime library implements three methods of memory
management.
• The runtime sharing (RS) method allocates and registers a large memory for
all static and dynamic coarrays at the initialization phase. The registered memory
is shared by all static and allocatable coarrays.