A cache memory controllor according to the present invention sequentially checks into a controlling circuit to see whether or not data words accessed are memorized in a cache memory and has a combination of an address generating circuit and a strobe signal producing circuit for producing a plurality of strobe signals in the presence of a request signal for a burst transmission mode of operation, so that a microprocessor sequentially fetches the data words in the cache memory even if the cache memory does not cope with the nibble mode of operation.

Logic for decreasing the number of cache lines dedicated to user data. When pools for allocation are selected using a dynamic storage allocation procedure, the size of a data block is compared to the size of the allocated pool. If the comparison results meet a predetermined criterion, the logic aligns the data to the beginning of a cache line and places the header in a separate cache line that may be deallocated. And if the data will fit within one-half of a cache slot in the allocated pool, then the line or lines having the header data can be re-used as the header is deallocated. Otherwise, user data blocks are placed in cache lines that are spatially local.

Buffer lines and corresponding buffer tags and buffer valid/dirty registers in conjunction with buffer control circuit are provided to a DMA controller enabling the DMA controller to conditionally pre-fetch data from memory while data being read are transferred from the DMA controller to the I/O device during read operations and to conditionally write data back to memory while data being written are transferred from the I/O device to the DMA controller during write operations, thereby improving asynchronous read/write throughputs. Read requests, random as well as sequential, are satisfied with pre-fetched data if they are validly stored. Write requests, random as well as sequential, are deferred, batched and optimized. The improved throughput is achieved in a manner completely transparent to system software.

The invention describes a modification of FIFO hardware to allow improved use of FIFOs for burst reading from or writing to a processor direct memory access unit via either an expansion bus or an external memory interface using FIFO flag initiated bursts. The hardware and FIFO signal modifications make the FIFO-DMA interface immune to deadlock conditions and generation of spurious interrupt events in the process of initiating burst transfers. The FIFO function is modified to synchronize the frame transfer on the digital signal processor even if the digital signal processor lacks this functionality. By delaying the programmable flag assertions within the FIFO until after the current burst is complete the DSP-FIFO interface may be made immune to deadlock conditions and generation of spurious events.

A method and circuit utilizing first and second tag bits for granting access to a block of memory in a multiprocessor system having a shared memory are provided. When a memory request for exclusive access to a block of memory is granted, the starting address for that block of memory is placed in a register bank, thereby opening a semaphore. The starting address of a memory block of a subsequent memory access request is compared with the starting addresses corresponding to open semaphores within the register bank and access is denied to the requested block of memory if a match is found. The starting address associated with a request which is denied access is placed in a temporary buffer and the request is later granted access after the corresponding open semaphore becomes closed. A request which is granted memory access to a memory block which results in an open semaphore, has exclusive access to that block of memory until the semaphore is closed.