Sperry Corporation

Unisys Corporation

Sperry Rand Corporation

John R Trost

James H Scheuneman

Daniel K Zenk

A priority scheme is utilized wherein various Instruction Processors possess predetermined levels of prioritization for gaining storage access. Basically, access locking devices used in this environment perform the function of allowing a single port requester to gain unlimited access to the storage unit port while preventing or locking out access by all other requesters seeking access to that same port. In operation, a plurality of pending prioritized port requests are 'snapped' into a plurality of input latches. This snapping step is often referred to as a 'snapshot'. It is only during this 'snapshot' period that pending port requests are allowed to enter the priority latches. In past systems, when a port request was made, accompanied by an access lock signal, all port requests presently residing in the priority latches waiting to be serviced were cleared out and no new requests were accepted into the latches until the next 'snapshot' was performed. By causing these lower priority pending requests to be cleared out and not reinstated until a successive 'snapshot', many such pending requests timed-out. This invention reduces the number of pending requests timing out.

Access lock apparatus for use with a high performance storage unit of a

The preferred embodiment shown involves forming the memory system of B memory banks, where B is preferably a prime number, but may be a nonbinary number, i.e., B=2.sup.X, where X is a positive integer, and where the requested address=(Q+R)B. The address translation system for each requestor seeking access to the memory system includes a ROM and an adder. The ROM is comprised of two ROMs, Q ROMa and Q ROMb. ROMb stores in successive memory locations a first portion Qb of the memory system address and Q ROMa stores in successive memory locatins a second portion Qa of the memory system address. An adder sums the data, Qa+Qb, stored in the addressed memory locations of Q ROMa and Q ROMb while Q ROMa stores in successive memory locations a Bank R portion that specifies the one of the B banks in which the sum Qa+Qb addresses the selected memory address in the selected memory bank of the memory system.

System for addressing a multibank memory system

In an n-wide (n nominally equals 4) snapshot priority network apparatus the access of n+1 requestors to a memory unit is prioritized. Two requestors--the lowest priority one of normal system requestors called instruction processors plus a maintenance exerciser type requestor--share a single memory port which is nominally the lowest priority one of n such prioritized ports. Requests from both requestors are both honored upon a single priority snap, the instruction processor request nominally proceeding before the maintenance processor request. Although the n-wide priority network remains generally faster than any (n+1)-wide priority network, the maintenance exerciser type requestor is expediently serviced and cannot be locked out of access to memory by the competing higher priority requests of the instruction processor.

Modified snapshot priority enabling two requestors to share a single

A system for accelerating the granting of prioritized memory requests to a multi port memory system of a data processing system is disclosed. The priority requestor accelerator system detects the fact that one remaining requestor is in the priority memory system. The priority system logic is cleared out before the end of the normal requestor cycle. This allows the acceptance of a new set of requestors to be presented to the priority circuits at that time rather than waiting until presentation of the final request. Thus, the accelerator detects that the requestors from a previous requesting snap are on their last cycle. This allows a preclearance of the lower ranks as the priority circuit finishes its last cycle. The new requests are then loaded and the priority inputs are snapped shut beginning a new set of cycles. The overall operation happens as if the priority circuit is just moving from one requestor to another that is already in residence after the snap. Without this acceleration, a full memory cycle would have been lost with each reloading of the priority register.

Priority requestor accelerator

A method of and apparatus for optimizing the data transfer rate from a dynamic storage subsystem to an asynchronously clocked requestor using a variable rate oscillator and a first-in-first-out or FIFO buffer. On writing data into the dynamic storage subsystem, data is transferred from the requestor to the FIFO buffer and the variable rate oscillator clocks the data from the FIFO buffer into the dynamic storage subsystem at some relatively slow rate. If the FIFO buffer begins to fill, the frequency of the variable rate oscillator increases causing data to be clocked from the FIFO buffer at a higher rate. On reading data from the dynamic storage subsystem, the variable rate oscillator begins at a maximum frequency clocking data into the FIFO buffer from where it is transferred to the requestor. If the FIFO buffer beings to fill up, the frequency of the variable rate oscillator is decreased. Therefore, for both reading and writing, data is always transferred via the FIFO buffer and the frequency of the variable rate oscillator controlling the dynamic storage subsystem is determined by the amount of data stored in the FIFO buffer.

Dynamic storage synchronizer using variable oscillator and FIFO buffer

An improvement to a semiconductor memory subsystem containing single bit error correction/double bit error detection (SBC/DBD) which provides correction of double bit errors through the utilization of a modest amount of additional circuitry. The present invention accomplishes this result through the technique of sequentially complementing each double bit pair within the semiconductor memory subsystem data word determined to contain a multiple error and rechecking the modified data word with the existing SBC/DBD circuitry, one double bit pair at a time, until it is determined by the SBC/DBD circuitry that such double bit pair complementing has corrected the double bit error.

Double bit error correction using double bit complementing

A method of and an apparatus for obtaining double bit error correction capabilities in a large scale integrated (LSI) semiconductor memory system using only single bit error correction, double bit error detection (SEC, DED) logic are disclosed. The method is based upon the statistical assumption that in a large scale integrated semiconductor memory, substantially all errors in the data bits that make up a data word are initially a single bit error and that increasing multiple, i.e., double, triple, etc., bit errors occur in a direct increasing ratio of the use or selection of the data word. In the present invention, all data words are priorly tested to be error free. Subsequent detection of single bit errors results in the correction of the single bit error and the storage of the single bit error correcting syndrome bits in a syndrome bit memory. Subsequent detection of double bit errors, in the previously single bit error detected and corrected data words, results in the correction, by single bit error correcting syndrome bits, of the previously detected single bit error. This single bit error corrected data word is then again single bit error corrected, i.e., two successive single bit error corrections, to provide a twice corrected double bit error data word.

Growing community of inventors

Anoka, MN, United States of America

John R Trost