47results about How to "Optimize memory usage" patented technology

In one embodiment of the present invention a convolution engine configures a parallel processing pipeline to perform multi-convolution operations. More specifically, the convolution engine configures the parallel processing pipeline to independently generate and process individual image tiles. In operation, for each image tile, the pipeline calculates source locations included in an input image batch. Notably, the source locations reflect the contribution of the image tile to an output tile of an output matrix—the result of the multi-convolution operation. Subsequently, the pipeline copies data from the source locations to the image tile. Similarly, the pipeline copies data from a filter stack to a filter tile. The pipeline then performs matrix multiplication operations between the image tile and the filter tile to generate data included in the corresponding output tile. To optimize both on-chip memory usage and execution time, the pipeline creates each image tile in on-chip memory as-needed.

In one embodiment of the present invention, a convolution engine configures a parallel processing pipeline to perform multi-convolution operations. More specifically, the convolution engine configures the parallel processing pipeline to independently generate and process individual image tiles. In operation, for each image tile, the pipeline calculates source locations included in an input image batch based on one or more start addresses and one or more offsets. Subsequently, the pipeline copies data from the source locations to the image tile. The pipeline then performs matrix multiplication operations between the image tile and a filter tile to generate a contribution of the image tile to an output matrix. To optimize the amount of memory used, the pipeline creates each image tile in shared memory as needed. Further, to optimize the throughput of the matrix multiplication operations, the values of the offsets are precomputed by a convolution preprocessor.

The invention relates to a stacked display algorithm of a progress bar for a breathing lamp effect. The stacked display algorithm comprises the following steps of (1) carrying out background adaptive processing on the progress bar, wherein the progress bar is a circular progress bar, the background of the progress bar is round and the diameter of the circular progress bar is smaller than that of the background of the progress bar; (2) setting a control and drawing related parameters; (3) drawing the circular progress bar in a stacking manner in the background area of the progress bar; (4) displaying the breathing lamp effect; (5) starting a change thread of the breathing lamp; (6) drawing progress text displayed along with the progress; and (7) drawing description text at the center of the circular progress bar. Through the circular progress bar and surrounding of the breathing lamp effect, aesthetic, individual, intuitive and easy-to-understand progress bar showing is achieved, and the algorithm is high in efficiency, few in occupied resources, good in universality and suitable for an Android system.

An input data stream is received for processing at an integrated circuit (IC) through multiple channels. The input data stream is interleaved and the interleaved input data stream is stored in a memory buffer associated with the IC. An addressing scheme is defined for reading and writing data samples from and to the memory buffer within the IC. The addressing scheme includes determining a current increment by analyzing a pattern associated with data samples within the memory buffer and determining the memory address for each data sample within the memory buffer using the current increment. A data sample for a frame is read from a corresponding address within the memory buffer using the addressing scheme and a subsequent data sample from the interleaved input data stream is written into the corresponding address of the memory buffer. The current increment and the addressing scheme are stored in the memory buffer. The current increment is dynamically determined by analyzing data samples in the memory buffer after processing of each frame of data and is used in redefining the addressing scheme for subsequent data processing.

A FIR filter for use in an adaptive multi-channel filtering system, includes a first memory for storing data, and a second memory for storing filter coefficients. The second memory stores only non-zero valued coefficients or coefficients that are above a predetermined magnitude threshold such that the overall number of coefficients processed is significantly reduced.

The invention discloses a 2D laser data-based real-time positioning method of a robot. The 2D laser data-based real-time positioning method is mainly used for solving the problem of low real-time performance and excessively low positioning accuracy of an existing robot positioning algorithm. On the basis of 2D laser data, an automatic Monte-Carlo algorithm is used as a basis, the gesture of the robot is updated in real time by the laser data according to a speed of the robot, and unreasonable gesture is filtered according to estimated gesture confidence. The method is used for positioning therobot, relatively high real-time performance is achieved, and moreover, the positioning accuracy can be controlled with a range being 1 centimeter.

The invention provides a virtual machine memory management method and equipment. The method comprises the following steps: identifying a non-operating system memory of a virtual machine from total memory allocated to the virtual machine; wherein the total memory comprises the memory of the virtual machine and the management memory of the virtual machine monitor; wherein the memory of the virtual machine comprises the memory of an operating system of the virtual machine and the memory of a non-operating system of the virtual machine; taking a memory of a non-operating system of the virtual machine as an exchangeable memory; and storing the data in the exchangeable memory into a nonvolatile memory. The problem of obvious reduction of the performance of the computer in an excessive submissionscene of the virtual machine can be avoided, the running stability of the virtual machine is improved, and the user experience is improved.

Disclosed is a method for automatic snapshot of an inner memory. The method comprises the steps of initializing inner memory snapshot records and a list of the inner memory snapshot records, setting the total size of the allocated inner memory for a current inner memory snapshot record as zero, and storing the current inner memory snapshot record to the list of the inner memory snapshot records; by monitoring a function of allocating inner memories to an operation system from a Mono platform of a Unity engine, obtaining a notification for new inner memory allocation from the operation system; according to the received notification for new inner memory allocation from the operation system, calculating the total size of the inner memory currently allocated from the system; calculating the difference between the total size of the inner memory allocated in a current moment and the total size of the allocated inner memory in the previous inner memory snapshot record; automatically generating the current inner memory snapshot records according to the difference and/or the interval between the current moment and the moment of the previous memory snapshot record, and storing the current inner memory snapshot record to the list of the inner memory snapshot records. The method for the automatic snapshot of the inner memory has the advantages that developers of mobile phone applications can conveniently monitor the changing conditions of the inner memory in the running process of a game.

The invention provides a file merging method and device of a big data platform. The method comprises the following steps of monitoring the directory change of the big data platform, and judging whether the number of files under the changed directory changes or not; grouping files having similar characteristics under the changed directory in a case where the number of files under the changed directory changes; judging whether there are small files smaller than an integer multiple of a set number of data blocks in the same group of files; when the small files exist in the files of the same group, obtaining the small files of the same group, and merging the small files of the same group. Through the above scheme, small files can be reduced, memory usage of namenode can be optimized, and the large data platform can accommodate more files.

The invention relates to the technical field of computers, and provides an interface rendering processing method and device, a computer readable storage medium and electronic equipment. The method comprises the steps of obtaining an engineering file and a picture resource file of a target application; determining a scene mode corresponding to each interface of the target application according to the engineering file; performing picture merging processing on the picture resource file of the target application according to the scene mode corresponding to each interface to obtain a picture set corresponding to each interface; and storing the atlas corresponding to each interface, wherein the atlas is used for rendering each interface. According to the technical scheme, rendering batches can be effectively reduced, and the interface rendering efficiency can be improved.

The invention provides a real-time dynamic management method and system for a distributed memory file system applied to a computer cluster. The method is used for dynamically adjusting a memory usage amount of the distributed memory file system on the computer cluster. The method comprises the following steps of collecting a memory usage amount of each computing node on the computer cluster at a control time point in real time; predicting a memory usage amount of the memory file system at the next control time point of each computing node according to the collected memory usage amount; and performing judgment through the predicted memory usage amount, and sending out a control instruction after the judgment to control the distributed memory file system to send out recovery and allocation instructions. By adopting the dynamic control system and method, the memory use of the computer cluster can be optimized, so that the efficiency of data-intensive and compute-intensive computing on the computer cluster is improved.

On a computer system having a processor, a single OS and a first instance of a system driver installed and performing system services, method for sharing driver pages among Containers, including instantiating a plurality of Containers that virtualize the OS, wherein the first instance is loaded from an image, and instantiating a second instance of the system driver upon request from Container for system services by: allocating virtual memory pages for the second instance and loading, from the image, the second instance into a physical memory; acquiring virtual addresses of identical pages of the first instance compared to the second instance; mapping the virtual addresses of the identical pages of the second instance to physical pages to which virtual addresses of the corresponding pages of the first instance are mapped, and protecting the physical pages from modification; and releasing physical memory occupied by the identical pages of the second instance.

A circuit having at least one processor and a microprogrammed machine for processing the data which enters or leaves the processor in order to input or output the data into/from the circuit in compliance with a communication protocol.

The invention discloses a memory management method and device for neural network reasoning, and the method comprises the steps: dividing a memory space into a first type of region and a second type ofregion, enabling the first type of region to be only used for storing FM data with the life cycle of 1, and enabling the second type of region to be used for storing FM data with any life cycle; analyzing the neural network to which the memory space is to be allocated, and determining whether to start a first type of region and a second type of region according to the number of layers with multiple inputs in the neural network and the total number of layers of the neural network; and allocating a memory space to the FM data of each layer in the neural network from the first type of region and/or the second type of region. According to the method, a proper memory management strategy is adaptively selected according to the structure of the neural network, and memory use is optimized. According to the method, a greedy algorithm is used for searching for the optimal memory allocation scheme layer by layer, memory occupation of neural network reasoning can be reduced, and memory usage is minimized as much as possible.