14results about How to "Improve Parallel Computing Efficiency" patented technology

The invention discloses a large model cloud computing dynamic updating method based on elastic adaptive incremental learning, which comprises the following steps: S1, in a cloud computing platform, acquiring a current version of a large model and a new data stream to be processed; s2, local parameter updating is conducted on the large model according to the new data, and parameters or model layers needing to be updated are dynamically selected according to the characteristics of the data in the updating process; s3, introducing an adaptive learning strategy, and monitoring the change condition of data distribution in real time; s4, in a large model updating process, splitting a large model updating task to a plurality of computing nodes for parallel execution through a multi-node collaborative scheduling mechanism of the cloud; s5, after the large model is updated, the system verifies the performance of the large model; and S6, in the process of updating and deploying the large model, the system dynamically recycles computing resources which are not used any more. According to the method, idle computing resources can be fully utilized in a large-scale cloud computing environment, so that computing tasks are distributed in a balanced manner, and dynamic adjustment of task loads is ensured.

The invention relates to the technical field of GPU parallel processing, in particular to a multi-thread parallel computing efficiency optimization system, which comprises a GPU, a processor and a memory in which a computer program is stored, and when the computer program is executed by the processor, the following steps are realized: a plurality of instruction streams corresponding to a parallel computing module at a preset time point, traversing each instruction stream, acquiring first I instructions in the corresponding instruction stream to obtain a plurality of instructions, generating a plurality of requests to be arbitrated according to the instructions meeting preset conditions, traversing all the requests to be arbitrated, and adding the requests to be arbitrated into request queues of corresponding access objects according to the access objects corresponding to the requests to be arbitrated, according to the multi-thread parallel computing method, the request queues corresponding to the access objects are obtained, each request queue is arbitrated, and each arbitration result is sent to the corresponding access object to be processed, so that instructions of different access objects can be executed in parallel, and the multi-thread parallel computing efficiency is improved.

The invention discloses a multi-entity cluster model parallel simulation method based on a directed graph, belongs to the technical field of cluster simulation, and solves the problem of low entity cluster parallel simulation calculation efficiency in the prior art. The method comprises the following steps: decomposing an entity simulation model, and constructing a directed graph of the entity simulation model according to a data stream dependency relationship among basic operator units; packaging each basic operator unit into a kernel function; determining the number of thread bundles required by each kernel function; at each moment, determining a kernel function which can be subjected to parallel calculation at the current moment based on the directed graph of the entity simulation model; issuing a calculation instruction of the kernel function which can be subjected to parallel calculation at the current moment to the GPU at the same time; the GPU side generates threads used for processing data of the corresponding entities according to the number of the thread bundles needed by each kernel function; and when the branch calculation instruction of the kernel function is executed, the entity corresponding to each thread is re-determined through double reference remapping. And efficient entity cluster parallel simulation calculation is realized.

The embodiment of the invention discloses a wake rendering method and device, electronic equipment and a storage medium, a graphics processor obtains appearance sampling data of to-be-rendered patches, obtains survival time information and patch position information of each to-be-rendered patch in a current frame, and interpolates the appearance sampling data according to the survival time information to obtain a wake rendering result; according to the method, a central processing unit acquires appearance sampling data of to-be-rendered patches in a current frame to obtain target appearance parameters of the to-be-rendered patches, and then the to-be-rendered patches in the current frame are rendered according to the target appearance parameters and the patch position information to obtain target wakes, and the appearance sampling data are acquired by sampling appearance parameter change curves of the to-be-rendered patches by the central processing unit. The central processing unit only needs to carry out sampling, the calculation process of obtaining the target appearance parameters is sunk to the graphics processing unit, calculation resources of the central processing unit are released, resource occupation of the central processing unit is reduced, wake rendering efficiency is effectively improved, and the method can be applied to the technical fields of cloud technologies, games, intelligent traffic, auxiliary driving and the like.

The present application relates to the field of GPU parallel processing technology, and in particular to a multi-threaded parallel computing efficiency optimization system. The system includes: a GPU, a processor, and a memory storing a computer program. When the computer program is executed by the processor, the following steps are implemented: at a preset time point, multiple instruction streams corresponding to a parallel computing module are traversed, the first I instructions in the corresponding instruction stream are obtained, and a plurality of instructions are obtained; according to the instructions that meet the preset conditions, a plurality of requests to be arbitrated are generated; all requests to be arbitrated are traversed, and according to the access objects corresponding to the requests to be arbitrated, the requests to be arbitrated are added to the request queue of the corresponding access object to obtain a plurality of request queues of the corresponding access objects; each request queue is arbitrated respectively, and each arbitration result is sent to its corresponding access object for processing respectively, so that the instructions of different access objects can be executed in parallel, thereby improving the parallel computing efficiency of multi-threads.

The present application discloses a load balancing method, device, equipment and storage medium, the method comprising: separating matrix grids and crack grids, preserving the connection relationship between the crack grids and the matrix grids; generating the adjacency relationship between the matrix grids and the adjacency relationship between the crack grids; calculating the weight of each matrix grid according to the preserved connection relationship and the generated adjacency relationship; performing multi-level coarsening decomposition on the matrix grid according to the adjacency relationship between the matrix grids and the weight of each matrix grid; allocating corresponding crack grids to each matrix grid sub-region after decomposition according to the connection relationship between the crack grid and the matrix grid; renumbering the crack grids and matrix grids of each sub-region, and constructing the connection relationship between adjacent sub-regions. In this way, the computational overhead is low, and the influence of the connectivity of the two sets of grids on the generated sub-region grid system is balanced in the process of decomposing only the matrix grid, there is a minimum interface between the sub-regions, and the parallel computing efficiency is high.

This invention belongs to the field of medical temporal signal processing technology, specifically a method for predicting respiratory-related displacement based on attention-embedded dilated convolution. The method involves acquiring respiratory-related temporal data of the target object over continuous time; inputting this data into a trained respiratory-related displacement temporal prediction network model, which outputs the predicted respiratory displacement. The network model comprises multiple cascaded dilated convolutional residual blocks and a prediction layer. Each dilated convolutional residual block includes a causal dilated convolutional unit, a self-attention mechanism unit, and a residual connection unit. The output of the last dilated convolutional residual block is input into the prediction layer, which outputs the respiratory-related displacement of the target object at future time points. This invention dynamically weights key time steps during each layer's temporal feature extraction process, thereby improving the predictive ability for respiratory-related displacement and tumor motion.

The invention discloses an operator fusion model training method and apparatus, a device, a storage medium and a program product. The method comprises the steps of obtaining performance data of an operator fusion model in a historical training process; wherein the performance data comprises at least one of operator operation performance data and operator calculation structures corresponding to different operator combinations; preprocessing the performance data, and obtaining key feature data from the preprocessed performance data; carrying out updating training on the operator fusion model by utilizing the key feature data; in the updating training process, adjusting and optimizing hyper-parameters of the operator fusion model to obtain a trained operator fusion model; wherein the operator fusion model is used for predicting the performance of the operator fusion strategy. By adopting the embodiment of the invention, the problem of overhigh memory occupation during model training can be avoided, the normal operation of model training is ensured, and the model training efficiency is improved.

The invention belongs to the field of database maintenance, and discloses a method and a related system for rapid data statistics of industrial time series data, which realize standardization and structuring of statistical logic by analyzing a customer data statistical request, extracting a tag name and a statistical task. Semantic differences of different databases in the aspects of statistical functions, aggregation rules, time window definitions and the like are avoided, and statistical tasks can be executed in a unified logic model. According to the method, the statistical task is divided into a plurality of time periods according to the time region, so that the system can independently execute statistical operation in different time windows. The sectional calculation not only improves the parallel calculation efficiency, but also avoids the deviation caused by different time index precisions of the bottom layer of the database, and effectively ensures the consistency and comparability of statistical results.

The embodiment of the invention discloses an image deblocking filtering method and device, electronic equipment and a computer readable medium. A specific embodiment of the method comprises the following steps: receiving a video frame image after video compression processing; generating information of each unit to be filtered based on the video frame image; performing filtering processing on each piece of to-be-filtered unit information to obtain each piece of filtering unit information group; generating a filtered video frame image based on each piece of to-be-filtered unit information and each piece of filtering unit information group; and storing the filtered video frame image to a video memory. According to the embodiment, the parallel efficiency of image deblocking filtering is improved.

The invention discloses a GPU parallel computing data dual mapping method and system, belongs to the technical field of parallel computing, and solves the problem of low parallel computing efficiency in the prior art. The method comprises the steps that the number of thread bundles needed by a kernel function is determined, parallel computing threads are generated, and original IDs and index IDs of the threads are determined; if the current instruction to be executed is a branch calculation instruction, each current effective thread determines a group, a block and a position in the block to which an instance corresponding to the current index ID of the effective thread belongs according to a branch judgment condition; reference redirection is carried out on each thread bundle; each thread carries out reference redirection on the thread according to the group corresponding to the thread bundle to which the thread belongs to obtain a new index ID of the thread; if the current instruction to be executed is a branch calculation ending instruction, modifying the index ID value of each thread into an original ID value; otherwise, each effective thread determines the corresponding instance according to the index ID to calculate the current to-be-executed instruction. And efficient parallel computing is realized.

This application provides a parallel computing method, apparatus, electronic device, storage medium, and product. The method includes: receiving a computing request, the computing request including input data; sending the input data to a local accelerator according to the computing request, so that the local accelerator performs computing processing on the input data through a first module; receiving first boundary data sent by the local accelerator, obtaining second boundary data from a neighboring processor based on the first boundary data, and sending the second boundary data to the local accelerator, wherein the neighboring processor corresponds to a different computing node than the local accelerator; receiving a target computing result sent by the local accelerator, and writing the target computing result to a local hard disk through an input / output library. The method of this application, by performing parallel computing through accelerators with a larger number of cores, can execute multiple threads simultaneously, thereby improving parallel computing efficiency.

The application discloses a kind of real-time radar signal processing systems based on high-speed acquisition and isomeric parallel, belong to real-time radar signal processing technical field;Including high-speed data acquisition and analysis module, CPU and GPU isomeric parallel processing module, the high-speed data acquisition and analysis module includes Gigabit fiber optic network card and WinPcap analysis unit, the Gigabit fiber optic network card is connected using QSFP interface with radar digital transceiver board, the WinPcap analysis unit bypasses operating system protocol stack, directly with Gigabit fiber optic network card drive interaction through driver, the high-speed data acquisition and analysis module is connected with CPU and GPU isomeric parallel processing module communication;The CPU and GPU isomeric parallel processing module includes mutually communicating CPU module and GPU module, the real-time radar signal processing system based on high-speed acquisition and isomeric parallel of the application, through the synergistic cooperation of high-speed data acquisition and analysis front end and multilayer parallelization signal processing, realizes the real-time processing requirement from data acquisition to target detection.

The present application belongs to the technical field of semiconductor wafer detection, and particularly relates to a wafer surface topography defect detection system based on three-dimensional point cloud reconstruction, comprising: a division module that obtains three-dimensional point cloud data of a wafer to be measured and fits a reference plane, selects the original point closest to the center of the reference plane as an initial sampling point, and divides the data into N subsets and distributes them to parallel computing units; a calculation module that iteratively performs farthest point sampling, takes the distance of the new point as a step size, combines a preset constant to update the Z-axis weight of the anisotropic Euclidean distance, recalculates the distance in each unit and selects the locally farthest point, and obtains the globally farthest point; and an identification module that fits a reference polynomial surface based on the sampled point set, calculates the orthogonal distance deviation of the three-dimensional point cloud data of the entire wafer to be measured to the surface, determines an abnormal threshold value according to the deviation, and identifies topography defects. The present application enhances the stability of the system to outliers, and reduces the false detection and missed detection rates of topography defects.