7871results about "Processor architectures/configuration" patented technology

The invention relates to a method and a device for detecting machine vision dynamic defects of a precise structural part. The method comprises the following steps: scanning a sample precise structural member to obtain an original image sequence; inputting the original image sequence into a GPU for parallel processing to obtain a preprocessed image sequence and a target feature point set; performing image matching to obtain a camera pose transformation matrix; dynamically correcting the preprocessed image sequence to obtain high-precision three-dimensional point cloud data; inputting the high-precision three-dimensional point cloud data into an improved U-Net deep learning network containing a geometric perception module for training and transfer learning to obtain a curved surface adaptive defect detection model; and based on the curved surface adaptive defect detection model, carrying out full-matrix sampling analysis and geometric model compensation on target three-dimensional point cloud data scanned by the precision structural member to be detected in real time to obtain a target defect detection result. According to the method, the accuracy and robustness of defect identification of the precision structural part are improved.

The embodiment of the invention provides a three-dimensional model adaptive generation method based on video and point cloud data, scene data is collected through a multi-view camera array and a depth scanning array, and key features are extracted and screened by using a hierarchical feature extraction network and an attention mechanism. And organizing local data blocks by adopting a hierarchical index tree based on a union-check set, and realizing self-adaptive parameter optimization by combining integrity and accuracy scores. Meanwhile, efficient reconstruction and material mapping of a local three-dimensional grid are realized through a parallel operation channel and a dynamic calculation resource scheduling mechanism. According to the method, the precision and efficiency problems of a traditional three-dimensional reconstruction scheme in processing a large-scale scene are effectively solved, and the model generation quality is improved.

The present invention relates to the field of AR performance optimization, and in particular to an integrated AR device performance optimization method and device based on a multi-core processor. The method includes the following steps: obtaining real-time AR display scene images and real-time user interaction data; performing scene object visual recognition on the real-time AR display scene images, and encapsulating image rendering data to generate multiple AR scene rendering tasks; performing visual rendering computing power calculations on multiple AR scene rendering tasks one by one, and identifying rendering requirements of different scenes to generate rendering requirement data for each task; identifying multi-core processor parameters in the system, performing image rendering computing power perception to generate image rendering computing power for each processor; performing computing power resource pre-allocation matching on the image rendering computing power of each processor according to the rendering requirement data of each task, and constructing a dynamic computing power resource allocation strategy. The present invention realizes dynamic AR device computing power resource allocation and improves the operating performance of AR devices.

The invention relates to the technical field of chip mounting processing, and discloses an optical alignment system and method for chip mounting processing, and the method comprises the steps: collecting a target chip image and a target substrate image in a mounting process through an upper visual field camera disposed above a suction nozzle assembly and a lower visual field camera disposed above a substrate carrying platform; chip edge accurate coordinates in the target chip image are extracted, and reference mark accurate coordinates in the target substrate image are extracted; solving a chip mounting process compensation parameter matrix of an upper and lower camera-suction nozzle-substrate four-coordinate system in the chip mounting process based on the chip edge accurate coordinate and the reference mark accurate coordinate; calculating an X-Y-theta three-dimensional alignment compensation vector in the chip mounting process according to the mounting process compensation parameter matrix; on the basis of the X-Y-theta three-dimensional alignment compensation vector, the optimal mounting track parameter is calculated, an execution instruction sequence is generated, the problem of error accumulation caused by traditional independent calibration is effectively solved, and the alignment precision is remarkably improved.

The invention discloses a method for automatically drawing an OpenGL (Open Graphics Library) program by using Vulkan. The method comprises the following steps of: creating a context used by the Vulkan, initializing each module, processing an OpenGL instruction related to texture and data buffering, and managing storage of texture and data buffering resources in a video memory; a shader program used by the OpenGL is preprocessed into a format acceptable to Vulkan, and an OpenGL shader program instruction is created and destroyed; processing an OpenGL (Open Graphics Library) instruction related to frame buffering to generate structural body information required by Vulkan dynamic rendering; an OpenGL instruction of the sampler is also created; processing an OpenGL (Open Graphics Library) instruction for creating a vertex input format and managing a vertex data buffer area, and maintaining vertex input information, a vertex buffer area and an index buffer area required by Vulkan; and finally, drawing or calculating, distributing and calling Vulkan on the basis of all the instructions.

The invention discloses a digital twin three-dimensional scene modeling method based on a webGPU, and relates to the technical field of three-dimensional scene modeling and graphic computing, multi-source sensing data are asynchronously sampled from sparse point cloud, a video texture sequence, a scene semantic tag graph and a structure boundary tuple, and a six-dimensional structure unit group is generated through a normalization operator; constructing a structural unit atlas with nodes representing component entities and edges representing constraint relations, and introducing a tension balance mechanism and multi-scale constraints to generate a modeling path prior model; constructing a graph calculation and graph rendering dual-channel assembly line in the WebGPU, and executing parallel texture mapping and boundary fitting operation; a dynamic sensing module is used for capturing scene disturbance and driving atlas response, and incremental reconstruction of the model is achieved; and finally, mapping the model to a Web terminal, and supporting microscopic semantic query and multi-layer data linkage. According to the method, the response speed, semantic consistency and structural adaptability of three-dimensional modeling in a complex environment are improved.

Apparatuses, systems, and techniques to train neural networks. In at least one embodiment, a first neural network is trained to match accuracy of a second neural network independently of outputs of the second neural network based on, for example, a third neural network that generates weights for the first neural network.

The invention provides a large-scale satellite three-dimensional image data processing method, medium and system, and belongs to the technical field of satellite three-dimensional image data processing.The method comprises the steps that Gaussian filtering and wavelet noise reduction preprocessing is conducted on an image, and then cooperative computing of a GPU and a CPU is achieved based on a computing task distribution function. Through multi-scale feature extraction and matrix optimization, a stable feature matrix and a variable feature matrix are obtained. A quadtree spatial index is adopted to carry out data partitioning, and SIFT feature extraction and K-means clustering are combined to optimize feature representation. And establishing a stereoscopic image registration model, determining an optimal splicing sequence through a minimum spanning tree algorithm, realizing image splicing by adopting a progressive texture fusion method, and finally generating a high-precision three-dimensional earth surface model. According to the invention, the unification of processing efficiency and precision is realized, and the technical problem that the processing efficiency of large-scale stereoscopic image data is difficult to improve on the premise of ensuring the processing precision in the prior art is solved.

Disclosed in the present invention are a GPU feature mining and analysis method and system based on computer microarchitecture independence. The method comprises: building a target instrumentation library, analyzing a calling sequence of a target application program, and extracting a plurality of levels of performance features, wherein the performance features are high-level features of a program and are independent of a microarchitecture of a processor; analyzing a correlation between the performance features and execution performance of the target application program; and outputting an analysis result according to the obtained correlation, wherein the analysis result reflects a performance bottleneck of the target application program and a corresponding optimization suggestion. The present invention can help developers to adjust and optimize application programs, so as to fully use computing capabilities of GPUs.

The invention discloses a point cloud data processing method and device, a storage medium and electronic equipment. The method comprises the following steps: acquiring initial point cloud data to be sent; dividing the initial point cloud data into a group of point cloud blocks; performing graphical representation on the basis of the point cloud data in each point cloud block in the group of point cloud blocks to generate a group of graph structure data, one graph structure data in the group of graph structure data corresponding to one point cloud block in the group of point cloud blocks, and the graph structure data being used for representing features of the corresponding point cloud block; and encoding the group of graph structure data to obtain a group of data packets, and sending the group of data packets to a target terminal so as to perform decoding reconstruction at the target terminal to obtain target point cloud data. The technical problems that the processing efficiency of the point cloud data is low and the effect cannot be quickly achieved in the related technology are solved.

A network environment comprises a plurality of host machines that are communicatively coupled to each other via a network fabric comprising a plurality of switches that in turn include a plurality of ports. Each host machine comprises one or more GPUs that execute customer workloads. Described herein are different approaches that provide for addressing the problem of handling network overlay encapsulation without causing adverse impact to the performance of workloads executed on the GPU clusters.

Techniques relating to processor communication structures are disclosed. In some embodiments, a processor includes a plurality of client circuits and a structure circuit including at least a first instance and a second instance of a tile. The tile may include a client input configured to interface with client circuitry, a tile input configured to interface with one or more other tile instances, and communication resources assignable to the client input and the tile input. The communication resources may include a plurality of internal links, a client output configured to interface with client circuitry, and a tile output configured to interface with one or more other tile instances. The control circuitry may assign communication resources for a given tile instance to the client input and at least a portion of the tile input for a next cycle based on priority information in a given cycle. The control circuit may update priority information based on the assignment result over a plurality of cycles.

The invention discloses a Direct3D memory model compatible method based on adaptive placeholder resources, which comprises the following steps: establishing resource metadata, accessing a scene library, checking an instruction template and double sandboxes when a DXVK is started, describing related resources by the DXVK through the metadata after a D3D application is started, completing resource mapping and metadata dynamic updating, and allocating the resources to the corresponding sandboxes; the DXVK compiles an application shader code, identifies a resource access instruction, matches an access scene and a check template by combining a pipeline stage and binding slot query metadata, instantiates the check instruction and adds the check instruction to the front of a target instruction; according to the unbound resources, adaptive generation of corresponding occupied resources is carried out, the authority and the state are configured, the unbound resources are bound to a Vulkan descriptor set to replace VKNULLHANDLE, and adaptation of the shader interface and the pipeline state parameters is completed to create PSO (Particle Swarm Optimization); and intercepting an application rendering instruction analysis parameter to construct a command buffer area, submitting the command buffer area to a Vulkan queue to trigger a GPU (Graphic Processing Unit) to execute rendering, and finally completing complete conversion and adaptation from the D3D rendering logic to the Vulkan.

The invention provides a multi-task dynamic resource sharing method and system for a universal graphics processor, and belongs to the technical field of computing graphics process.The method comprises the steps that a plurality of computing tasks are distributed to processing subunits in a cooperative processing unit respectively; obtaining the load state of the computing resource in each processing subunit, and determining the available computing resource capacity of each processing subunit according to the load state; according to the available computing resource capacity, marking the processing subunit of which the current execution thread beam instruction queue length exceeds the own available computing resource capacity as a source processing subunit, and marking the processing subunit with idle computing resources as a target processing subunit; and migrating part or all of the to-be-executed thread beam instructions in the to-be-executed thread beam instruction queue of the source processing subunit to the idle computing resources of the target processing subunit for execution. According to the method and the device, the cross-processing subunit dynamic migration is carried out on the thread beam instruction based on real-time load monitoring, so that the throughput rate and the computing resource utilization rate are improved.

An apparatus to facilitate lossless and lossy automatic hardware compression in graphics-to-graphics network links is disclosed. The apparatus includes compressor / decompressor circuitry (CDC) integrated with physical layer (PHY) intellectual property (IP) hardware circuitry for a graphics processor unit (GPU)-to-GPU communication link communicably coupling a first GPU to one or more other GPUs, the CDC to: receive a data message from the first GPU, wherein the data message is in an uncompressed format; determine that a compression process is to be applied to the data message; apply the compression process to the data message to generate a compressed data message; and cause a GPU link IP hardware circuitry that comprises the PHY IP hardware circuitry to transmit the compressed data message over the GPU-to-GPU communication link.

The invention belongs to the field of computer graphics, and discloses a fast parallel multi-scale smooth symbol distance curved surface reconstruction method and system, and the method comprises the steps: carrying out the local fitting of point cloud data through employing a moving least square method, and calculating the Gaussian curvature of a point cloud; constructing a curvature adaptive octree based on Gaussian curvature, wherein the octree is used for dynamically adjusting local resolution and capturing geometric details to improve curved surface reconstruction precision; global fitting is carried out through an implicit function, and a zero contour surface is obtained through an octree contour surface extraction algorithm; and multiple threads are introduced in the process of iteratively solving the linear system to carry out parallel computing. The adaptive octree based on Gaussian curvature can effectively capture complex geometric details, and the algorithm realizes high-precision curved surface reconstruction under different scales. In addition, the local and overall calculation efficiency is remarkably improved through parallel calculation, and the high efficiency of the algorithm in large-scale data processing is ensured. The method provided by the invention provides an efficient and accurate solution for complex curved surface reconstruction.

The invention relates to a resource-limited large model heterogeneous training method, computer equipment and a storage medium, and the method comprises the steps: determining a first activation amount according to a dependency relationship between calculated amounts corresponding to the activation amounts and / or the calculated amount and the storage amount of each activation amount in the activation amounts generated when an image processor carries out model forward propagation, the first activation amount is asynchronously transmitted to the central processing unit; loading a first activation amount to the image processor from the central processing unit in a model back propagation process; obtaining the gradient of the model parameter calculated by the image processor based on the first activation amount, and asynchronously transmitting the gradient to the central processing unit; and determining an optimizer obtained based on gradient updating in the central processing unit, obtaining model parameters obtained by adjustment of the updated optimizer, and asynchronously transmitting the model parameters to the image processor. By adopting the method, the problems that a large model with limited video memory resources is difficult to train and the computing efficiency of a video card is low can be solved.

An apparatus providing tensor core matrix multiplication and accumulation (MMA) with hardware-based statistics collection and outlier suppression is disclosed. The apparatus includes processor circuitry comprising at least one processor core comprising matrix multiplication circuitry to: execute a matrix multiplication operation on first input data from a first set of registers and on second input data from a second set of registers; collect, as part of executing the matrix multiplication operation via statistics collection hardware circuitry of the matrix multiplication circuitry, output statistics data corresponding to the matrix multiplication operation; and output the output statistics data along with a result of the matrix multiplication operation; and output statistics storage to store the output statistics data.

The invention is suitable for the field of GPU management, and provides an intelligent dynamic management method for GPU computing power and a cloud platform, and the method comprises the following steps: collecting task data and GPU state data, and generating a task queue and a resource allocation strategy in combination with a scheduling plug-in; based on the task queue and the GPU real-time load, dynamically adjusting a resource allocation proportion through reinforcement learning to analyze a task dependency relationship and generate a migration plan; optimizing a communication path and adjusting asynchronous transmission delay according to the task dependency relationship and the GPU communication topology, and outputting a synchronous state mark; and monitoring abnormity in combination with the synchronization state and the GPU hardware state, executing thermal migration according to the migration plan, performing video memory recovery, and updating the resource idle list. According to the invention, through algorithm innovation and hardware collaborative optimization, intelligent, dynamic and efficient resource scheduling in the multi-GPU system is realized.

Provided in the present disclosure are a data processor, a data processing method, an electronic device and a non-transitory computer-readable storage medium. The data processor comprises a tensor operation unit and N compute units, wherein the tensor operation unit is configured to execute tensor computation on input data, so as to obtain tensor computation results; and the N compute units are configured to execute at least one of a vector operation of the tensor computation results and the generation of input data, wherein first data transmission channels are provided between the tensor operation unit and at least some compute units among the N compute units, and the first data transmission channels are used for directly providing the tensor computation results to the compute units, and directly providing the input data from the compute units to the tensor operation unit. By means of the data processor, access to a global memory can be reduced, the waste of resources is reduced, the data transmission delay is shortened, the overall efficiency of operators is significantly improved, the strong computing power of a tensor operation unit itself is effectively used, and the computation efficiency of the data processor is improved.

The invention provides a game scene optimization method and system based on dynamic rendering, and belongs to the technical field of game development.The game scene optimization method comprises the steps that firstly, a real-time rendering state data set of a current game scene, rendering resource occupation information containing scene elements and picture frame rendering time consumption parameters are obtained; then, a pre-trained rendering demand analysis model is called for analysis, a rendering demand feature set of the scene elements is obtained, the rendering demand feature set comprises element visual importance features and rendering resource sensitivity features, and a scene element dynamic optimization strategy is generated based on the rendering demand feature set; the scene element dynamic optimization strategy comprises an element detail level adjustment rule and a rendering resource allocation priority parameter, executing a rendering parameter adjustment operation on a target scene element according to the scene element dynamic optimization strategy, generating adjusted scene rendering configuration data, and finally inputting the scene rendering configuration data into a game rendering pipeline for real-time rendering processing. And outputting the optimized game scene picture frame sequence. Therefore, the game picture quality and the operation performance can be improved.

The invention provides an intelligent computing center model development method and device oriented to the Plevatory computing power, and relates to the technical field of computing power infrastructure, the method comprises the following steps: S1, receiving first mirror image data of a to-be-developed model; s2, a first node is obtained, the first node is a node occupying graphics processing unit (GPU) resources of the intelligent computing center, and the first node comprises a debugging service component; s3, adjusting the first mirror image data based on the debugging service component; and S4, after the adjustment of the first mirror image data is completed, closing the first node, and releasing GPU resources occupied by the first node. According to the method, GPU resources can be prevented from being continuously occupied, the load of the intelligent computing center is reduced, the intelligent computing center can provide computing power service for more users, and the general application of computing power is realized.

Techniques are discussed herein for determining low-cost and / or optimal driving trajectories for autonomous vehicles between various states in a driving environment. A lattice subgraph of lowest-cost trajectories associated with road segments in an environment may be precomputed off-vehicle, and provided to vehicles in the environment along with map data, lane graphs, etc. An autonomous vehicle may use the lattice subgraphs to determine costs associated with discrete entry points of the road segments along a driving route, and may perform additional computations using motion primitives and / or cost plots to determine costs associated with controlling the vehicle between various states to the discrete road segment entry points. The lattice subgraphs may be grouped and cost computations for the states within the various lattice subgraphs can be performed by the vehicle in parallel, to provide reduced computational complexity and faster computations of low-cost driving trajectories.

The invention discloses a plastic effect simulation display method and system based on a 3D simulation technology, and relates to the technical field of medical information, and the method comprises the steps: collecting the facial data of a patient, and constructing a multi-level facial structure initial model; constructing a hybrid deformation model; individual characteristic parameters of a patient are extracted, and an individual biomechanical parameter set is generated through the biomechanical parameter prediction network; applying the personalized biomechanical parameter set to the mixed deformation model; operation input parameters of plastic surgery are received, and the deformation result of each layer of structure is calculated and displayed in real time based on a GPU parallel computing framework; constructing a tissue healing and recovery kinetic model, and generating appearance prediction data at different postoperative time points in combination with the deformation result; and performing three-dimensional visualization processing on the appearance prediction data. According to the method, CT / MRI, 3D structured light scanning and expression dynamic video sequences are creatively integrated, the limitation of traditional single-mode data modeling is broken through, and the soft tissue boundary recognition precision is remarkably improved.

The invention discloses an image processing data analysis platform based on a memristor storage and calculation integrated architecture, and relates to the technical field of data identification, and the platform comprises an XC7Z020 FPGA control module which is connected with an OV5640 camera image acquisition module, a memristor array calculation module and an Internet of Things module. According to the image processing data analysis platform based on the memristor storage and calculation integrated architecture, through deep fusion of the memristor storage and calculation integrated architecture and a dynamic optimization technology, the speed, energy efficiency and reliability of image processing are remarkably improved. The data handling bottleneck of the traditional von Neumann architecture is eliminated, the processing speed is improved compared with that of the traditional GPU scheme, and meanwhile, the system energy consumption is reduced. Weight distribution and fragmentation processing strategies are adjusted through real-time feature feedback, so that the recognition accuracy in a complex industrial scene is improved, the throughput is broken through, and the real-time processing requirement of a high-resolution image is met.

The invention provides a reasoning method and device, equipment and a medium. The method comprises the steps that when a reasoning request is received, model parameters of a target model deployed on each GPU are obtained, the reasoning request comprises information of a reasoning task, and the information of the reasoning task comprises input data; then, according to the structure information of the target model, segmenting the reasoning task into a plurality of groups of subtasks corresponding to the GPU according to a combination mode of a calculation task and a communication task; segmenting the input data into input sub-data corresponding to each group of sub-tasks; afterwards, generating a CUDA flow corresponding to the GPU according to a sequence dependency relationship between the calculation task and the communication task in each group of subtasks, the CUDA flow being used for realizing parallel execution of multiple groups of subtasks; then, a CUDA kernel on the GPU is scheduled according to the CUDA flow, so that the GPU executes multiple groups of subtasks in parallel according to model parameters and input subdata, and a reasoning result corresponding to the reasoning request is obtained. According to the invention, the reasoning time delay can be reduced.

The embodiment of the invention provides an information mapping method for intelligent perception information in a virtual space, perception data is acquired by deploying a multi-sensor array, data features are extracted based on a multi-scale feature pyramid and an attention mechanism, and a space mapping relation is established by using self-supervised learning and geometric constraints. Local data blocks are organized by adopting a hierarchical index tree based on a union search set, an adaptive calculation flow graph is constructed to realize efficient mapping, and data transmission is optimized in combination with predictive caching and a differential updating strategy. And meanwhile, parallel processing of incremental mapping is realized through a pipeline data channel, and the data synchronization efficiency is ensured by a dynamic scheduling mechanism based on load prediction. The method effectively solves the problem of performance bottleneck when a traditional mapping scheme processes large-scale multi-source sensing data.

A hierarchical network enables access for a stacked memory system including or more memory dies that each include multiple memory tiles. The processor die includes multiple processing tiles that are stacked with the one or more memory die. The memory tiles that are vertically aligned with a processing tile are directly coupled to the processing tile and comprise the local memory block for the processing tile. The hierarchical network provides access paths for each processing tile to access the processing tile's local memory block, the local memory block coupled to a different processing tile within the same processing die, memory tiles in a different die stack, and memory tiles in a different device. The ratio of memory bandwidth (byte) to floating-point operation (B:F) may improve 50× for accessing the local memory block compared with conventional memory. Additionally, the energy consumed to transfer each bit may be reduced by 10×.