142 results about "Workgroup" patented technology

The invention relates to the field of heterogeneous parallel computing, in particular to an attention operator head dimension block computing method applied to a sea light DCU, which comprises the following steps of: expanding a sequence length block computing range in charge of each thread block in an attention operator and a matrix computing range processed by a single thread working group; a blocking parameter splitqk and a blocking parameter splitv are selected, wherein the blocking parameter splitqk and the blocking parameter splitv are selected; selecting a proper block cutting parameter splitqk for the head dimensions of the query tensor Q and the key tensor K for cutting, accumulating block calculation results, and calculating S block calculation results through a softmax function to obtain P block calculation results; selecting a proper block cutting parameter splitv for the head dimensions of the value tensor V and the output tensor O for cutting, wherein a block calculation result corresponds to a corresponding block of the output tensor O; and finally, segmenting the tensor O according to the splitv, and writing each partitioning result back to the global memory. The method is suitable for a heterogeneous parallel computing system composed of a CPU and a DCU, computing resources of a DCU computing unit can be saved, performance loss caused by resource overflow is reduced, and computing efficiency is improved.

The invention provides a method and device for an intelligent computing center cloud platform to adjust model training parameters according to computing power operation states, and relates to the technical field of intelligent computing centers, intelligent computing centers and computing power infrastructures. The first working component is a component deployed in a node in the intelligent computing center cloud platform, the target link comprises a data preparation link, a model training link, a model evaluation link and / or a model test link in the first working component, and the target link performs data processing or model training based on preset hyper-parameters; s2, calculating target correction parameters corresponding to the operation related parameters in a preset period based on a preset regulation and control model; and S3, adjusting the preset hyper-parameter based on the target correction parameter to obtain a target hyper-parameter. The computing power utilization rate of the intelligent computing center cloud platform can be greatly improved, and then the model training efficiency is greatly improved.

The invention discloses a traffic information sharing method and device for vehicle and road cloud integration, and a medium. The method comprises the following steps: constructing a vehicle social network graph; analyzing a natural language subscription request of a user, and generating a structured decision element; a data and knowledge dual-driven collaborative decision framework is adopted to carry out multi-attribute decision, and optimal working group configuration is generated; the sensing demand parameters are encrypted and issued to all nodes in the optimal working group, sensing fusion of multi-node space-time association is achieved, and multi-source sensing data are obtained; performing standardized packaging and credibility evaluation on the acquired multi-source sensing data, and filtering untrusted data to obtain trusted data; cleaning the credible data, and performing hierarchical desensitization on sensitive information in the credible data to obtain traffic information; and pushing the traffic information to the subscribing user, and constructing an intelligent closed-loop optimization system based on user feedback to optimize the traffic information sharing service. According to the invention, the dynamic cooperative sensing capability in traffic information service can be improved.

The invention discloses a control method and device of a heat dissipation device, a computer readable storage medium, an electronic device and a computer program product. The method comprises the steps that a current working scene matched with current running data of a display card to be cooled is determined in a preset scene library; corresponding temperature weights are set for all the working assemblies according to the current working scene, and the higher the load of the working assembly is, the higher the temperature weight of the working assembly is; determining the comprehensive temperature of the display card according to the temperature of each working component and the corresponding temperature weight; and controlling the running state of the heat dissipation device of the display card according to the comprehensive temperature to reduce the temperature of the display card. The comprehensive temperature is obtained by weighting the temperatures of all the working assemblies, so that subsequent control over the operation state of the heat dissipation device of the display card can better meet the heat dissipation requirement of the working assembly with the higher load in the current working scene, and the technical problem that the heat dissipation effect of the display card is poor in the related technology is solved; and the heat dissipation effect of the display card is improved.

A system for privacy-preserving communication and compliance enforcement is disclosed. The invention introduces Virtual Identities (VIs) and Virtual Numbers (VNs) cryptographically bound to Compliance Jurisdiction Tokens (CJTs) generated in secure enclaves. Each CJT carries session identifiers, expiry, consent artifacts, and jurisdictional codes covering GDPR (EU), CCPA (US), HIPAA, PSD2, PCI-DSS, FATF AML, the Digital Personal Data Protection Act (India), and the Personal Information Protection Law (China). Forwarding of calls, payments, or data packets is blocked unless VI–CJT validation and ledger anchoring succeed. Unlike tokenization, the system enforces real-time consent revocation, immutable regulator-verifiable audit, cross-border adequacy validation, and dual cryptographic signatures (RSA / ECC with post-quantum upgrade). Industrial use spans telecom, finance, healthcare, e-commerce, SaaS, and IoT, ensuring privacy-by-design and data protection obligations are enforced at protocol level.

Quantum-based dispatch of workgroups is described. An example of an apparatus includes a computer memory to store data for processing, including data for an application; and one or more processors including a graphical processing unit (GPU), the GPU including multiple chiplets, each of the multiple chiplets including compute containers and a cache, each compute container including a plurality of processing resources, and a dispatcher for dispatching workgroups to the processing resources of the GPU, wherein dispatching workgroups includes dispatching workgroups for the application according to a selected workgroup quantum, the selected workgroup quantum having a certain size and shape.

The invention provides a synchronization signal processing method, electronic equipment and a computer readable storage medium, and the method comprises the steps: executing a first synchronization instruction by a meta thread, generating an arrival and request response signal carrying a working group slot position identifier and a barrier object identifier according to first synchronization information, and sending the arrival and request response signal to a synchronization engine; the execution engine executes the meta-thread function instruction and generates an arrival signal according to the second synchronization information; aggregating the arrival signals with the same slot position identifier and barrier object identifier to obtain a final arrival signal; the synchronization engine determines a target barrier object and updates a count value, and constructs a response signal and sends the response signal to the target meta-thread when the count value meets a preset condition. According to the technical scheme, the synchronization signal processing overhead can be reduced, and the parallel execution efficiency is improved.

A processing system (100) includes a scheduling mechanism for generating data for fine-grained reordering of work groups of kernels to generate data blocks, such as for communication across devices, to enable producer computations to overlap with communication that is fully reduced across the network. The scheduling mechanism enables the first parallel processor (702) to schedule and execute a set of work groups of producer operations to generate data for transmission in a desired traffic pattern to the second parallel processor (704). Meanwhile, the second parallel processor schedules and executes a different set of work groups of the producer operations to generate data for transmission to a third parallel processor (706) or back to the first parallel processor in a desired traffic pattern.

The invention provides a resource allocation method, a resource allocation device, electronic equipment and a medium, and the method comprises the steps: determining whether the number of available memory logic sub-blocks in a shared memory meets the demand of a work group task or not according to the demand number of shared memory resources; under the condition that the number of available memory logic sub-blocks in the shared memory meets the requirements of the working group tasks, the corresponding memory logic sub-blocks are allocated to the working group tasks, and for each thread bundle task in the working group tasks, the corresponding thread bundle task is allocated to the corresponding thread bundle task based on the required number of universal register resources of the corresponding thread bundle task. Determining whether the number of available register logic sub-blocks in one execution unit in the plurality of execution units meets the requirement of a corresponding thread bundle task; and under the condition that the number of the available register logic sub-blocks in one execution unit in the plurality of execution units meets the requirement of the corresponding thread beam task, allocating the corresponding register logic sub-blocks in the execution unit to the corresponding thread beam task. The problem of storage space fragmentation is solved.

A processor includes a plurality of processing elements. The processor is configured to execute a work graph including a plurality of nodes representing kernels executable by one or more processing elements of the plurality of processing elements. A first processing element of the one or more of the processing elements associated with a first node of the plurality of nodes is configured to assign each logical division of a sparse input matrix to a bin of a plurality of bins. Responsive to a dispatch condition associated with a bin of the plurality of bins, the processor is configured to dispatch a workgroup to at least a second processing element associated with at least a second node of the plurality of nodes corresponding to the bin. The workgroup includes a plurality of work items based on one or more logical divisions of the sparse input matrix assigned to the bin.

The invention provides a station-level production task generation method for a ship manufacturing workshop. The method is applied to the field of general control systems, and comprises the following steps: extracting process characteristic parameters in structured model data; obtaining real-time data of a workshop, and calculating a processing load value corresponding to each process characteristic parameter according to the real-time data of the workshop; setting a capacity adaptive threshold value, and matching a corresponding station type and a working group based on a comparison result of the processing load value and the capacity adaptive threshold value; according to the process adaptation range of the station type and the operation capability of the working group, dividing production batches and distributing station resources, and generating an initial station-level production task; and acquiring execution data of the initial station-level production task in real time, calculating a task execution deviation rate based on the execution data, dynamically adjusting a productivity adaptive threshold value or resource allocation of production batches according to the task execution deviation rate, and outputting the optimized station-level production task. Therefore, the production task generation efficiency is improved.

The utility model discloses a machine room, and relates to the technical field of machine rooms, the machine room comprises an outer shell, an inner shell, a working assembly, a cooling assembly and a movable assembly, the outer shell is provided with a ventilation opening; the inner shell is arranged in the outer shell, and the inner shell is made of a high-thermal-conductivity material; the working assembly is arranged in the inner shell; the cooling assembly is arranged in the inner shell and used for cooling air in the inner shell. The movable assembly is movably connected with the outer shell and can move relative to the outer shell to open or close the ventilation opening. According to the technical scheme, the problems that an existing machine room is high in cost, large in size and difficult to maintain and manage due to the fact that complex or expensive equipment is input for cooling can be solved.

The invention provides a computing chip, a reduction operation method, a related device and a medium, and the computing chip comprises a plurality of computing units which are used for executing a computing task containing a thread bundle and a working group, and each computing unit comprises a loading storage unit and a local shared memory; the data caching unit is used for caching the work group level reduction results from the plurality of computing units and executing a reduction operation across work groups; the global memory is used for storing a grid-level reduction result after the reduction operation of the data cache unit; wherein a first-level hardware reduction module is integrated in the loading storage unit, and is used for executing a thread bundle and reduction operation in a working group, and submitting an obtained working group reduction result to a local shared memory; and a second-stage hardware reduction module is integrated in the data cache unit, is coupled with the first-stage hardware reduction module and is used for executing reduction operation across working groups. According to the method, the memory access delay and the calling overhead are reduced, and the execution efficiency of the grid reduction operation is improved.

A method for compressing work item coordinate data for work items in a work group and sending the data across an interface between a computation requesting unit and a computation sequencing unit. A work item valid mask is created in dependence on the number and positions of work items in the work group, the work item valid mask indicating valid work items in the work group. A first swizzle mask indicates which bits of a swizzle index for each work item in the work group correspond to the value of a first coordinate for that work item. A second swizzle mask indicates which bits of the swizzle index for each work item in the work group correspond to the value of a second coordinate for that work item.

The embodiment of the invention provides a scheduling method of a working group, a graphics processor and electronic equipment, the scheduling method of the working group is applied to an arbitration module of the graphics processor, the graphics processor further comprises at least two execution modules, and the scheduling method comprises the following steps: aiming at each execution module, executing the execution modules; determining a first score of the execution module based on the resource request of the working group and the residual resource information of the execution module; under the condition that the first score of each execution module is smaller than a first score threshold value, a target execution module is determined from the at least two execution modules based on the first waiting duration of each execution module, and the first waiting duration of the execution module is the duration needed by the prediction execution module to release the target resource information; the target resource information is determined based on the resource request of the working group and the residual resource information of the execution module; and scheduling the working group to the target execution module.

The application relates to an unmanned aerial vehicle (UAV) intelligent fault detection system and method for inspection, which comprises an UAV fault monitoring server, a communication network, ground fault detection terminals and random fault detection terminals. Each ground fault detection terminal forms a detection work group with at least one random fault detection terminal. The random fault detection terminals are connected with the ground fault detection terminals in the same detection work group through the communication network. The ground fault detection terminals are connected with the UAV fault monitoring server through the communication network. The detection method comprises three steps of system configuration, dynamic detection and ground detection. The application can effectively meet the needs of fault detection and troubleshooting of UAVs with different structures, and can effectively realize the cooperation of synchronous detection in the UAV operation process and ground static detection, so that the flexibility and convenience of UAV detection operation are greatly improved.

A system for low earth orbit (LEO) satellite communication. In one example, the system includes a control computer and a workgroup including a plurality of communication devices. Each communication device includes an electronic processor configured to establish a wireless network connection with the control computer, determine connectivity data associated with a signal between the communication device and a first LEO satellite, and transmit the connectivity data to the control computer via the wireless network connection. The control computer is configured to receive the connectivity data from each communication device, and select a first one of the plurality of communication devices as a primary communication device based on each of the connectivity data. The primary communication device is configured to receive media from a different one of the plurality of communication devices via the wireless network connection and transmit the media to the first LEO satellite.

The present invention describes how a workgroup expert system can be established to mimic a real world task-expert and possess the equal four expert-Human-Intelligent (expert-HI) Problem-Solving (PS) competencies, including: 1) real-time concurrent workgroup-AI PS-processing, 2) real-time semantic workgroup-AI PS-transactions, 3) real-time task-domain workgroup-AI PS-collaborations and 4) real-time fine-grained adaptive workgroup-AI PS-services, based on multi-node workgroup architectures with derived workgroup-software methods and developed workgroup-system disciplines. Therefore, according to the Turing Test, the workgroup expert-task system should be deemed "Strong-AI-PS competent" for solving any task that is handled by one task expert with the help of functional processors, while all the current nodes-service-infrastructures with four node-AI-PS competencies can only mimic a group of real world functional processors with pre-developed logic-modelled processor-Human-Intelligent (processor-HI) PS-competencies for solving a pre-defined / specific multi-function-modelled task-oriented problem and should be deemed "Weak-AI-PS competent".

The invention relates to the technical field of intersection of computer software and medical information management, and discloses a work data management method and system for clinical laboratory personnel, and the method comprises the steps: obtaining the work authority of personnel; matching a working component from a database according to the working permission; the working components are put into the operation interface according to the first sequence, and the higher the use frequency is, the higher the position is; obtaining the latest adjustment time of the work permission; calculating permission duration according to the current time and the adjustment time; calculating the operation duration of the worker on the working component from the adjustment time to the current time; calculating a management duration according to the permission duration and the operation duration; sorting the management durations of the personnel with the same work authority from small to large to obtain a second sequence; and obtaining a temporary position in the second sequence according to the management duration of the personnel, and amplifying the display area of the working component corresponding to the temporary position in the first sequence in the operation interface corresponding to the personnel.

The invention discloses a scheduling method, a graphics processor, a scheduling device, a chip and equipment, and belongs to the field of chips. The scheduling method is executed by the GPU, the GPU at least comprises Y scheduling units, and each scheduling unit comprises a core or a processing unit. One core comprises at least one processing unit; the method comprises the following steps: recording an xth scheduling unit which is finally scheduled during nth scheduling; wherein the scheduling granularity of the nth scheduling is a working group or a working item, and one working group comprises a plurality of working items; in the (n + 1) th scheduling, scheduling is started from the (x + 1) th scheduling unit, so that polling scheduling in the Y scheduling units is realized; wherein the scheduling granularity of the (n + 1) th scheduling is a working group or a working item; n is a positive integer, and x is a positive integer smaller than or equal to Y. When multiple times of scheduling are executed, different scheduling units are cyclically executed for the first scheduling granularity, and therefore load balancing of the multiple scheduling units can be achieved for multiple times of scheduling.

This specification provides a Rootfs quota management system, method, computer device, and storage medium, relating to the field of computer technology. The Rootfs quota management system includes a master control component, worker components, and instruction execution components corresponding to the worker components. The worker components can connect to the container runtime of their respective worker nodes. The master control component is used to obtain the quota request of a target task, which corresponds to a target container, and the target container mounts a corresponding root file system (Rootfs). Based on the quota request, it queries quota management data, determines the quota instruction, and sends the quota instruction to the target worker component. The worker component, when successfully connected to the container runtime, receives the quota instruction sent by the master control component. If the target container corresponding to the container identifier carried by the quota instruction exists and is running, it calls the instruction execution component to execute the quota at the writable layer of the target container's Rootfs. This improves the flexibility and compatibility of quota management.

Policy-based genomic digital data sharing facilitates a variety of sharing scenarios, including public access, tenant-to-tenant sharing, workgroup sharing, and access by external service providers. Genomic digital data can be published to the platform and controlled by access tokens that are generated based on access policies. The policies can support conditions that are evaluated at execution time and effectively place control of access to information in hands of the owning tenant. Sharing conditions can be easily specified to support various use cases, relieving administrators from excessive access control configuration.

The processing system (100) includes a scheduling mechanism for generating data for fine-grained reordering of workgroups of kernels to generate data blocks, such as for communication across devices to enable overlap of AllReduce communication and producer computations over a network. This scheduling mechanism enables a first parallel processor (702) to schedule and execute a set of workgroups of producer operations to generate data for transmission to a second parallel processor (704) in a desired traffic pattern. Concurrently, the second parallel processor schedules and executes a different set of workgroups of producer operations to generate data for transmission in a desired traffic pattern to a third parallel processor (706) or back to the first parallel processor.

The invention discloses a data processing method and device for a matrix multiplication kernel function, a medium, equipment and a product, and the method comprises the steps: respectively configuring a corresponding ready state identifier for each output block in an output result of a main cycle stage; wherein each ready state identifier is triggered to be updated by the sub-working group executing the relevant calculation task of the corresponding output block; after the nth ready state identifier represents that the main loop calculation task of the nth output block is completed, post-processing operation of the tail sound stage is executed on the nth output block, so that execution time overlapping of the main loop stage and the tail sound stage is achieved; wherein 1 < = n < = N; n is the total number of the output blocks. According to the method, the idle waiting time of the general core can be effectively shortened, and the utilization rate and the calculation efficiency of hardware resources are remarkably improved.

The invention belongs to the technical field of cloud edge collaboration, and particularly relates to a computing resource optimal configuration method oriented to cloud edge collaboration. Comprising the steps that K0, the operation sequence of data matrix operation examples is optimized through memory addresses and demand quantity sorting of the data matrix operation examples; k1, configuring a data memory and a shared memory in the computing service unit; k2, establishing a parallel data buffer area in the shared memory for pre-reading; k3, optimizing and matching the scheduling function to ensure that the computing service unit can realize better data multiplexing; and K4, optimizing a calculation work group load distribution strategy to improve the utilization rate of calculation resources. According to the computing resource optimization configuration method, the computing efficiency of the cloud edge cooperation task can be improved, the resource consumption of data matrix processing is reduced, and the cloud edge cooperation processing capability is improved.

The embodiments of the present application provide an image annotation method, device, electronic device and storage medium, which relate to the field of image processing technology. The method comprises: first determining the workgroups to which multiple target accounts logged in in parallel belong respectively, then determining the annotation object corresponding to the workgroup to which the target account belongs based on the first correspondence between the workgroup and the annotation object, and generating an annotation task for the target account, then determining the designated annotation object for the target account, and based on the designated annotation object, retrieving the image to be annotated of the target account, thereby determining the image to be pushed of each target account, and then pushing the image to be pushed of each target account to the target terminal device, so that the target terminal device displays the image to be pushed, and finally obtaining the annotation information of the image to be pushed of the target account. By applying the embodiments of the present application, the accuracy of image annotation is improved while ensuring the efficiency of image annotation.

Programmable processing unit (65) executes graphics processing programs for execution threads corresponding to work items. A storage region (work group local storage) (31) can be allocated for temporary use by a respective group of threads corresponding to a group of work items being executed by the processing unit while the threads are being executed. A clear operation circuit (Clear Daemon) (33) can write (e.g. incrementally) a clear value to all entries of a storage region (which can include the region). A command can be sent from the processor to the clear circuit to write a clear value. Clear operation control parameters may comprise both an indication of the storage region to be written to, and an indication of the clear value to be written to that region. Access to a storage may be stalled (paused) until clear values have been written in that region. Arbitration circuits may decide between storage access requests relating to regions from execution threads or other requesters. Clear operation may be triggered before de-allocating a region for a work group, or in response to a thread executing an instruction. Enforcing a barrier may ensure all threads reach the barrier before triggering a clear operation.

Embodiments herein dynamically calculate a scale / offset on a per tile (or per block) basis rather than on a per tensor or channel basis. This enables the scale to be determined in place in the compute unit (e.g., a workgroup)—e.g., without having to perform a second pass or retrieve data from main memory. The scale for the tile can be determined by the compute unit using different techniques. In one embodiment, the scale is determine from the data in the tile itself. In another embodiment, a historical scale could be used.