Electrical computers and digital processing systems: memory – Storage accessing and control
Reexamination Certificate
2001-05-15
2004-08-24
McLean-Mayo, Kimberly (Department: 2186)
Electrical computers and digital processing systems: memory
Storage accessing and control
C711S167000, C711S168000, C711S118000, C711S220000, C710S035000, C710S066000, C712S034000
Reexamination Certificate
active
06782445
ABSTRACT:
FIELD OF INVENTION
The invention relates to computer architectures involving a main processor and a coprocessor, and in particular to use of memory resources by the coprocessor in such architectures.
DESCRIPTION OF PRIOR ART
Microprocessor-based computer systems are typically based around a general purpose microprocessor as CPU. Such microprocessors are well adapted to handle a wide range of computational tasks, but they are inevitably not optimised for all tasks. Where tasks are computationally intense (such as media processing) then the CPU will frequently not be able to perform acceptably.
One of the standard approaches to this problem is to use coprocessors specifically adapted to handle individual computationally difficult tasks. Such coprocessors can be built using ASICs (Application Specific Integrated Circuits). These are built for specific computational tasks, and can thus be optimised for such tasks. They are however inflexible in use (as they are designed for a specific task alone) and are typically slow to produce. Improved solutions can be found by construction of flexible hardware which can be programmed with a configuration particularly suited to a given computational task, such as FPGAs (Field Programmable Gate Arrays). Further flexibility is achieved if such structures are not only configurable, but reconfigurable. An example of such a reconfigurable structure is the CHESS array, discussed in International Patent Application No. GB98/00262, International Patent Application No. GB98/00248, U.S. patent application Ser. No. 09/209,542, filed on Dec. 11, 1998, and its European equivalent European Patent Application No. 98309600.9.
Although use of such coprocessors can considerably improve the efficiency of such computation, conventional architectural arrangements can inhibit the effectiveness of coprocessors. It is desirable to achieve an arrangement in which computations can be still more effectively devolved to coprocessors, particularly where these computations involve processing of large quantities of data.
SUMMARY OF INVENTION
Accordingly, there is provided a computer system, comprising: a first processor; a second processor for use as a coprocessor to the first processor; a memory; at least one data buffer for buffering data to be written to or read from the memory in data bursts in accordance with burst instructions; a burst controller for executing the burst instructions; and a burst instructions element for providing burst instructions in a sequence for execution by the burst controller; whereby burst instructions are provided by the first processor to the burst instructions element, and data is read from and written to the memory by the second processor through the at least one data buffer in accordance with burst instructions executed by the burst controller.
This arrangement is particularly advantageous where the coprocessor is to work on large blocks of data, particularly where the memory addresses of such blocks vary regularly. This arrangement allows for such blocks to be moved effectively in and out of the main memory with minimal involvement of the main processor (which is the system component least well adapted to use them).
A particularly efficient structure can be achieved if the coprocessor is controlled in a similar way to the data buffers. This can be done with a coprocessor instructions element for providing coprocessor instructions to control execution of the second processor in a sequence (with said coprocessor instructions originally provided by the first processor). Advantageously a coprocessor controller receives the coprocessor instructions from the coprocessor instructions element and controls execution of the second processor accordingly. This coprocessor controller may control communication between the coprocessor and the at least one data buffers: for example, where a bus exists between the coprocessor controller and the data buffers, the coprocessor controller may control access of separate data streams in and out of the second processor to the bus.
Particular benefit can be gained if there is a synchronisation mechanism for synchronising execution of the coprocessor and of burst instructions with availability of data on which the coprocessor and the burst instructions are to execute. This is particularly well accomplished if the coprocessor executes on the basis of coprocessor instructions. An effective approach is for the synchronisation mechanism to be adapted both to block execution of coprocessor instructions requiring execution of the second processor on data which has not yet been loaded to the data buffers, and to block execution of burst instructions for storage of data from the data buffers to the memory where such data has not been provided to the data buffers by the second processor. A particularly effective way to implement the synchronisation mechanism is to use counters which can be incremented or decremented through appropriate burst and coprocessor instructions, and which block particular instructions if they cannot be decremented further.
In a further aspect the invention provides a method of operating a computer system, comprising: providing code for execution by a first processor; extraction from the code of a task to be carried out by a second processor acting as coprocessor to the first processor; determining from the code and the task burst instructions to allow data to be read from and written to a main memory in data bursts for access by the second processor by means of at least one data buffer; and execution of the task on the coprocessor together with execution of burst instructions by a burst controller controlling transfer of data between the at least one data buffer and the main memory.
Advantageously, following extraction of the task from the code, coprocessor instructions for execution by a coprocessor controller are determined to control execution of the task by the second processor.
It is further advantageous if in execution of the task, synchronisation between execution of coprocessor instructions and execution of burst instructions is achieved by a synchronisation mechanism. This synchronisation mechanism may usefully comprise blocking of first instructions until second instructions whose completion is necessary for correct execution of the first instructions have completed. This mechanism may employ counters which can be incremented or decremented through appropriate burst or coprocessor instructions.
REFERENCES:
patent: 4099236 (1978-07-01), Goodman et al.
patent: 4589067 (1986-05-01), Porter et al.
patent: 5708830 (1998-01-01), Stein
patent: 5784582 (1998-07-01), Hughes
patent: 5884050 (1999-03-01), Wheeler et al.
patent: 6321310 (2001-11-01), McCarthy et al.
patent: 6336154 (2002-01-01), McCarthy et al.
patent: 6442671 (2002-08-01), Faucon et al.
patent: 0 021 365 (1991-01-01), None
patent: 0 581 698 (1994-02-01), None
patent: 0 853 283 (1998-07-01), None
Messias et al., “Team an array processor with a &mgr;P for super performance in data comm,”Electronic Design 8, XP-000211543, pp. 115-120 (Apr. 12, 1980).
McCarthy Dominic Paul
Olgiati Andrea
Hewlett--Packard Development Company, L.P.
McLean-Mayo Kimberly
LandOfFree
Memory and instructions in computer architecture containing... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Memory and instructions in computer architecture containing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Memory and instructions in computer architecture containing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3309101