37 results about "Task sharing" patented technology

The present invention provides a message-oriented middleware solution for securely transmitting messages and files across public networks unencumbered by intervening network barriers implemented as security measures. It also provides a dynamic, dedicated, application level VPN solution that is facilitated by the message-oriented middleware. Standard encryption algorithms are used to minimize the threat of eavesdropping and an Open-Pull Protocol (OPP) that allows target nodes to pull and verify the credentials of requestors prior to the passing of any data. Messaging can be segregated into multiple and distinct missions that all share the same nodes. The security network's architecture is built to resist and automatically recover from poor, slow, and degrading communications channels. Peers are identifiable by hardware appliance, software agent, and personally identifiable sessions. The security network provides a dynamic, private transport for sensitive data over existing non-secure networks without the overhead and limited security associated with traditional VPN solutions.

An arrangement for converting speech into text comprises a mobile device (202) and a server entity (208) configured to perform the conversion and additional optional processes in co-operation. The user of the mobile device (202) may locally edit the speech signal prior to or between the execution of the actual speech recognition tasks, by replacing an inarticulate portion of the speech signal with a new version being recording of the portion. Task sharing details can be negotiated dynamically based on a number of parameters.

The invention discloses a robot weak light environment grabbing detection method based on a multi-task sharing network, and belongs to the technical field of computer vision and intelligent robots. According to the method, strict matching of shot images in a weak light environment and a normal light environment is ensured through image acquisition, a corresponding data set d is constructed, then Darknet-53 is adopted as a backbone network to construct a weak light environment capture detection model, multi-scale features with strong feature expression ability are extracted, and, through a parallel cascade capture detection module and an image enhancement module, capturing detection and weak light image enhancement tasks are realized respectively; training samples are randomly selected from the data set d, a weak light environment capture detection model is used for prediction, and when the change of a loss value within the iteration number iter is smaller than a threshold value t, the capture detection model G is converged, that is, training of the model G is completed; and an image Ilow shot in the weak light environment is input into the trained model G to obtain a predicted capture frame parameter and an enhanced image, and a capture detection task in the weak light environment is completed.

The invention discloses a method for realizing data sharing among tasks. The problems that deadlock is easy to be generated by multi tasks when visiting the sharing data area and the fault is not easy to be localized are solved. The invention includes that the task without the authority to visit the sharing data area sends an ACTION message to the task with visit authority; after the task with visit authority receives the message, the task runs corresponding program and operates the sharing data area according to the content of the message; after the running is over, the task with visit authority sends a responses message to the sender of the ACTION message; the responses message comprises the result of running; the sharing data area of the invention need not to be protected and the sharing data area data can be safely visited among tasks. The lock and other protection mechanisms are not considered and the code is simplified. The deadlock can not be generated without lock ,so the problem of deadlock is completely avoided.

In a multitask execution environment, a debugging device performs debugging setting for rewriting part of original recording content in a memory area shared by at least two tasks, and debugging cancellation for restoring rewritten recording content back to original recording content. The debugging device stores a memory area used by each task, and address information specifying each debugging target task and a respective address. When task switching occurs, if a next task is not a debugging target, recording content at a physical address specified by address information other than that of the next task and within the physical address space range used by the next task is put into a post-debugging cancellation state. If the next task is a debugging target task, in addition to the above processing, recording content at the physical address specified by the address information of the next task is put into a post-debugging setting state.

A semiconductor device may include a task controller configured to generate a target share for a plurality of task generators according to respective target states and respective measured states of the plurality of task generators, a task scheduler configured to schedule the plurality of tasks according to an allocated share, the plurality of tasks being provided from the plurality of task generators, and a share controller configured to determine the allocated share according to the target share and a measured share of the plurality of task generators.

The invention discloses a fine-grained GPU resource management method for mixed loads. A capacity-based streaming multiprocessor abstract model CapSM is proposed and serves as a basic unit of resource management; when the mixed loads (including an online task and an offline task) share GPU resources, the use of the GPU resources by different types of tasks is managed through fine granularity, and task resource allocation and resource online adjustment are supported; and while the GPU resources are shared, the service quality of the online task is ensured. The resources allocated to the tasks are determined according to the types of the tasks, a resource request and a current GPU resource state of a system; the use of the GPU resources by the offline task can be met under the condition of sufficient resources; and when the GPU resources are insufficient, the resource use of the offline task is dynamically adjusted to preferentially meet the resource demand of the online task, so that when the mixed loads run at the same time, the performance of the online task can be ensured and the GPU resources can be fully utilized.

The invention discloses a parallel implementation method of a big data Karatsuba algorithm. According to the method, the correlation problems of storage and calculation of partial products are solved through ingenious formula transformation technique, pointer operation and storage manner on the basis of 64-bit unsigned long integer operation, the algorithm is parallelized by adopting a section task sharing policy through OpenMP multi-thread programming, so that the first-layer parallelization of 8 threads in a recursive program is started to solve 8 partial products, each section is responsible for the calculation task of one partial product, and after all the partial products are solved, serial merging is carried out, so that the Karatsuba algorithm is parallelized and the efficiency of the algorithm is improved.

The present invention is applicable to the technical field of computer information processing, and provides a method for improving task switching efficiency and a terminal device. The method includes: controlling the CPU's All coprocessors are in a disabled state; when the task runs to a coprocessor instruction, the coprocessor corresponding to the coprocessor instruction is in a disabled state, an undefined instruction exception is triggered, and the corresponding undefined instruction exception processing flow is executed, After determining that the undefined instruction belongs to the instruction set range of the coprocessor, the protection and recovery of the context content information of the coprocessor is performed, so that the context content information of the coprocessor can be automatically protected on demand according to task requirements and recovery, which not only improves the efficiency of task switching, but also enables multiple tasks to share the use of coprocessors.

The present application provides a task processing method, a service platform, and an electronic device. When the method obtains an instruction to start a first task, it determines the currently running second task that can share resources with the first task, and based on The remaining resources of the target resource used by the second task start the first task. Wherein, the resource remaining amount of the target resource used by the second task is not lower than the resource amount required when the first task is running. It can be seen that the present application provides a multi-task resource sharing solution. When the application solution is applied to the resource scheduling of small tasks, more than one small task can be started and run based on the same GPU, so that This enables multiple small tasks to share the same GPU resource, effectively overcomes the problem of low utilization of GPU resources in the prior art due to small tasks monopolizing the GPU, and reduces waste of resources.

The present invention relates to a processing unit task scheduling method and device, the method comprising: when the scheduling opportunity arrives, judging whether there is a forecast assertion for the idle processing unit involved in the scheduling opportunity; when predicting the assertion, acquiring the forecast assertion For the associated target task pair, the secondary task in the target task pair is scheduled to be executed on the idle processing unit; the main task is scheduled to be executed on the first processing unit after the prediction assertion is created. From the perspective of task demand estimation and hardware resource management, the present invention expands the scope of task and resource fusion by constructing target task pairs, and achieves good shared execution of tasks on the basis of task demand and hardware resources. Reach a balance, and finally improve the efficiency of processing unit task scheduling.

The invention discloses a method and system for multi-level caching of a main thread shared by tasks in a thread pool, and equipment. The method for multi-level caching of the main thread shared by the tasks in the thread poolcomprises the following steps: setting a multi-level cache query method; defining a thread pool, and setting a task execution method in the thread pool; starting a main thread, and caching the data to cache middleware and local according to a multi-level cache query method; and creating a sub-task in a main thread, submitting the sub-task to a defined thread pool, executing the sub-task, and obtaining data from the local. The system for multi-level caching of the main thread shared by the tasks in the thread pooll comprises: a definition setting module which is used for setting a multi-level cache query method and is used for defining a thread pool and setting a task execution method in the thread pool; a command execution module which is used for starting a main thread and caching the data to the cache middleware and the local according to a multi-level cache query method and is used for creating a sub-task in the main thread, submitting the sub-task to the defined thread pool, executing the sub-task and acquiring data from the local.

The invention discloses a machine-oriented universal coding method, including: compression coding stage: for different machine intelligence analysis tasks, adopt independent task mode and / or multi-task sharing mode to compress corresponding video or image data to obtain corresponding feature information ; Put the feature information into the code stream corresponding to the video or image data, and at the same time combine the compression mode to give the label information required in the decoding, so as to obtain the feature code stream; Decoding stage: read the header information of the feature code stream, according to the The annotation information determines the compression mode used in the compression stage, so that the feature information is read in a corresponding way, and then the feature information is used as the input of the machine intelligence analysis task to obtain the analysis result. This method can realize the encoding of video / image feature information required for each task, thereby improving the efficiency of intelligent task analysis, reducing transmission pressure, supporting edge analysis and computing, and other possible future needs.

The invention relates to a reconstruction method of tomographic imaging. According to an embodiment of the invention, the reconstruction method of tomographic imaging includes the following steps: projection data are obtained; a graphics processing unit uses an open computing language to filter the projection data; and the graphics processing unit uses the open computing language to perform back projection on the filtered projection data, and performs normalization processing on a volume image reconstructed through the back projection. By coordinating task sharing between a CPU and the GPU, the reconstruction method of tomographic imaging provided by the embodiment of the invention ingeniously utilizes structural features of the open computing language and reduces extra time occupied by data transmission, and thus the speed of reconstruction can be substantially improved. In addition, the reconstruction method of tomographic imaging also improves quality of imaging through processing such as weighting and normalization.

The invention discloses a real-time task scheduling method for sensing temperature-scheduling length of a heterogeneous multi-core processor. The real-time task scheduling method comprises the following steps that a default weight factor alpha of a system is read; tasks stored in a task queue is allocated to an optimal processor; optimal frequency is selected for each task in processor task queues; an execution sequence of the tasks in each processor task queue is determined; the tasks exceeding temperature constraints are segmented; whether the current scheduling length of the system is smaller than or equal to the given stop time D shared by all the tasks is judged, if yes, task execution is performed, and scheduling is completed, if not, scheduling failure is returned, the tasks cannot be executed, and the scheduling is completed. The scheduling length of the system can be minimized under the condition that the system meets peak temperature constraints and time constraints by using the real-time task scheduling method.