High flow-rate synchronizer/scheduler apparatus and method for multiprocessors

Claim 1

A hardware synchronization/scheduling apparatus for performing synchronization/scheduling in a multiprocessor system by controlling, during program run-time, a process of monitoring and detecting which computational tasks are allowed for execution and allocating computational tasks to processors, the tasks being represented by instructions and data accessible to the processors via instruction and data storage hardware, said synchronization/scheduling apparatus comprising: means for monitoring and detecting which computational tasks are allowed for execution, said monitoring/detecting means containing a task map describing the precedence relations among the computational tasks of the program, and communication/distribution means connected to said monitoring/detecting means, and to said processors for distributing, to the processors, information on computational tasks detected by said monitoring/detecting means to be allowed for execution in a processor and for forwarding, to said monitoring/detecting means, information on termination of execution of computational tasks at the processors, said communication/distribution means comprising a network of nodes possessing both the capability of decomposing information on a pack of allocated computational tasks into messages of finer partial packs of allocated computational tasks to be sent toward the processors and possessing the capability of unifying packs of information on termination of computational tasks into a more comprehensive pack, to be sent to said means for monitoring/detecting of allowed computational tasks.

Claim 5

A method of performing synchronization/scheduling in a multiprocessor system by controlling, during run-time, a process of monitoring and detecting which computational tasks are allowed for execution and allocating computational tasks to processors, the tasks being represented by instructions and data accessible to the processors via instruction and data storage hardware, said method comprising the steps of: monitoring and detecting which computational tasks are allowed for execution in accordance with a task map describing the precedence relations among the computational tasks of the program; and distributing to the processors information on computational tasks detected in said monitoring and detecting step to be allowed for execution in a processor and forwarding information on termination of execution of computational tasks at the processors, said distributing step being performed in a network comprising nodes possessing both the capability of decomposing information on a pack of allocated computational tasks into messages of finer partial packs of allocated computational tasks to be sent toward the processors and possessing the capability of unifying packs of information on termination of computational tasks into a more comprehensive pack, provided in said monitoring and detecting step with respect to allowed computational tasks.

Claim 6

A method of performing synchronization/scheduling in a multiprocessor system by controlling, during run-time, a process of monitoring and detecting which computational tasks are allowed for execution and allocating computational tasks to processors, the tasks being represented by instructions and data accessible to the processors via instruction and data storage hardware, said method comprising the steps of: monitoring and detecting which computational tasks are allowed for execution in accordance with a task map describing the precedence relations among the computational tasks of the program; and distributing to the processors information on computational tasks detected in said monitoring and detecting step to be allowed for execution in a processor and forwarding information on termination of execution of computational tasks at the processors, said distributing step being performed in a network comprising nodes possessing both the capability of decomposing information on a pack of allocated computational tasks into messages of finer partial packs of allocated computational tasks to be sent toward the processors and possessing the capability of unifying packs of information on termination of computational tasks into a more comprehensive pack, provided in said monitoring and detecting step with respect to allowed computational tasks.