Provided are an apparatus and method for a non-power-of-2 size cache in a first level memory device to cache data present in a second level memory device having a 2.sup.n cache size. A request is to a target address having n bits directed to the second level memory device. A determination is made whether a target index, comprising m bits of the n bits of the target address, is within an index set of the first level memory device. A determination is made of a modified target index in the index set of the first level memory device having at least one index bit that differs from a corresponding at least one index bit in the target index. The request is processed with respect to data in a cache line at the modified target index in the first level memory device.

Disclosed herein includes a system, a method, and a device for reading and writing sparse data in a neural network accelerator. A mask identifying byte positions within a data word having non-zero values in memory can be accessed. Each bit of the mask can have a first value or a second value, the first value indicating that a byte of the data word corresponds to a non-zero byte value, the second value indicating that the byte of the data word corresponds to a zero byte value. The data word can be modified to have non-zero byte values stored at an end of a first side of the data word in the memory, and any zero byte values stored in a remainder of the data word. The modified data word can be written to the memory via at least a first slice of a plurality of slices that is configured to access the first side of the data word in the memory.

A queuing requester for access to a memory system is provided. Transaction requests are received from two or more requestors for access to the memory system. Each transaction request includes an associated priority value. A request queue of the received transaction requests is formed in the queuing requester. Each transaction request includes an associated priority value. A highest priority value of all pending transaction requests within the request queue is determined. An elevated priority value is selected when the highest priority value is higher than the priority value of an oldest transaction request in the request queue; otherwise the priority value of the oldest transaction request is selected. The oldest transaction request in the request queue with the selected priority value is then provided to the memory system. An arbitration contest with other requesters for access to the memory system is performed using the selected priority value.

Disclosed herein includes a system, a method, and a device for reading and writing sparse data in a neural network accelerator. A plurality of slices can be established to access a memory having an access size of a data word. A first slice can be configured to access a first side of the data word in memory. Circuitry can access a mask identifying byte positions within the data word having non-zero values. The circuitry can modify the data word to have non-zero byte values stored starting at an end of the first side, and any zero byte values stored in a remainder of the data word. A determination can be made whether a number of non-zero byte values is less than or equal to a first access size of the first slice. The circuitry can write the modified data word to the memory via at least the first slice.

Provided are a method and apparatus for method and apparatus for encoding registers in a memory module. A mode register command is sent to the memory module over a bus, initialization of the memory module before the bus to the memory module is trained for bus operations, to program one of a plurality of mode registers in the memory module, wherein the mode register command indicates one of the mode registers and includes data for the indicated mode register.

A system and method for providing dynamic I/O virtualization is herein disclosed. According to one embodiment, a device capable of performing hypervisor-agnostic and device-agnostic I/O virtualization includes a host computer interface, memory, I/O devices (GPU, disk, NIC), and efficient communication mechanisms for virtual machines to communicate their intention to perform I/O operations on the device. According to one embodiment, the communication mechanism may use shared memory. According to some embodiments, the device may be implemented purely in hardware, in software, or using a combination of hardware and software. According to some embodiments, the device may share its memory with guest processes to perform optimizations including but not limited to a shared page cache and a shared heap.

Provided is a method for optimising memory writing in a device implementing a cryptography module and a client module calling functions implemented by the cryptography module. The device includes a random access memory including a first memory zone that is secured and dedicated to the cryptography module and a second memory zone dedicated to the client module. When the client module calls a series of functions implemented by the cryptography module including a first function and at least one second function, with each second function executed following the first function or from a further second function and providing a runtime result added to a runtime result of the preceding series function, each runtime result is added to a value contained in a buffer memory allocated in the first memory. The buffer memory value is copied to the second memory zone following the execution of the last function of the series.

A host includes a cache including a plurality of cache lines, a command descriptor list configured to store a command transmitted from one of the plurality of cache lines, a host controller including a direct memory access (DMA) device that accesses the command descriptor list, and a processor configured to determine a size of the command descriptor list based on a size of the one of the plurality of cache lines.

The present invention discloses a memory access method, a buffer scheduler, and a memory module, which can support multiple application scenarios without changing the memory module or a memory chip. The method includes: receiving an operation request message for memory access data, where the operation request message includes tag information of the memory access data, operation information of the memory access data, and a memory address of the memory access data; and performing, according to at least one of the tag information of the memory access data, a memory address of the memory access data, and the operation information of the memory access data, an operation on the tag of the memory access data and/or the memory access data stored in the memory module. The present invention is applicable to the computer field.

Methods of operating a memory include performing a memory access operation, obtaining an address corresponding to a subsequent memory access operation prior to stopping the memory access operation, stopping the memory access operation, sharing charge between access lines used for the memory access operation and access lines to be used for the subsequent memory access operation, and performing the subsequent memory access operation.