149 results about "Dataflow" patented technology

The invention discloses a variation test method and device based on an intelligent automation script. The method comprises the following steps: extracting a statement execution sequence and a branch dependency relationship in an original source code by using a context-aware syntax tree, and generating a control flow diagram and a data flow diagram; performing code semantic analysis on the multi-dimensional code features by using a large language model, identifying potential defect types and variation rules, predicting the test efficiency of variants in combination with a deep Q learning model, and generating to-be-processed variants; performing grammar check and equivalence analysis on the to-be-processed variants to obtain effective variants, and testing the effective variants; variation test indexes are calculated, code weak points are positioned, and test blind areas are identified according to survival variants, so that a visual analysis report is generated; and iteratively generating a new test case according to the current survival variant and the visual analysis report, and testing the current survival variant by using the new test case. According to the invention, the variation test efficiency can be improved.

The invention discloses a code reconstruction method, system and device based on function recognition and a medium, and the method specifically comprises the steps: scanning a project file topological structure, and constructing a knowledge graph based on a code import relation and a technology stack feature; performing context enhancement by analyzing the code function description and combining with the knowledge graph to generate a code modification instruction set; based on the code modification instruction set, in combination with data stream sensitivity marking and performance portrait analysis, evaluating the target code to obtain an evaluation result; based on an evaluation result, matching a historical optimal practice mode through a graph neural network, and generating an optimization scheme which retains an original design style; by comparing abstract syntax tree differences between original codes and verified optimization schemes, only functional logic nodes are replaced, and legal code style features are reserved. According to the method, the efficiency and quality of code reconstruction and optimization are improved, and a more convenient, efficient and intelligent code processing tool is provided for software developers.

The invention relates to an automatic code auditing method and device, computer equipment and a storage medium. The automatic code auditing method comprises the steps of obtaining a grammar structure, a control flow and a data flow of a to-be-audited code; constructing a context graph of the to-be-audited code according to the grammatical structure, the control flow and the data flow of the to-be-audited code; obtaining a multi-modal collaborative vulnerability detection method, wherein the multi-modal collaborative vulnerability detection method comprises a static analysis method based on rule matching, a symbolic execution method based on a code path, a large model reasoning method based on semantic understanding and weights of the methods; and identifying one or more code vulnerabilities, the vulnerability type of each code vulnerability and the confidence coefficient according to the context graph of the to-be-audited code and the multi-modal collaborative vulnerability detection method. According to the method, the audit codes of various vulnerability types can be processed while the code audit efficiency can be improved.

The invention provides a syntactic structure tree-based code change influence intelligent analysis method and system, and the method comprises the steps: extracting a control flow, a data flow and a call flow according to an AST generated through the analysis of an original code and a changed code, and constructing a static dependency graph SDG containing a multi-dimensional dependency relationship; the method comprises the steps that a base layer is used for comparing AST differences before and after code change and generating grammatical structure change feature vectors; the service layer matches a corresponding service label for the changed AST node by using the service domain template; the change labeling layer constructs a cross-version dependency chain for the change AST nodes and generates historical feature vectors; and calculating a risk value of changing the AST node by utilizing a feature calculation result and combining with a multi-dimensional influence diffusion algorithm, and automatically generating a decision suggestion. According to the method, multiple dimensions of a control flow, a data flow and a call flow are deepened, risk prediction is optimized by utilizing natural language processing and historical change data, and the influence brought by code change is assisted to be better understood and managed.

The invention relates to the field of artificial intelligence, in particular to an AI accelerator construction method and a related device. The method comprises the following steps: importing machine learning models of different AI frames through a multi-frame adaptation interface, and converting the machine learning models into linear algebraic dialect representation by adopting an MMIR dialect conversion chain; through target-independent optimization processing, core calculation semantics in linear algebraic dialects are reserved, and linear algebraic dialect representation is reconstructed into optimization intermediate representation with an efficient data flow structure; identifying target hardware characteristics of the target hardware; adaptively executing hardware optimization processing matched with target hardware characteristics on the optimization intermediate representation; and converting the target intermediate representation into a hardware operation unit executable by the target hardware, loading the hardware operation unit into the target hardware to obtain an AI accelerator, thereby running the hardware operation unit on the target hardware through the AI accelerator, scheduling corresponding computing resources in the target hardware, and completing hardware acceleration of the machine learning model.

The invention provides a single instruction stream multi-fiber programming model and a construction method thereof. The method comprises the following steps: constructing a task, thread and fiber three-level concurrent unit; different code logic units are configured for each task, and one task is mapped to a plurality of threads; mapping the threads and processor cores one by one, and mapping one thread to a plurality of threads; the threads within the thread are caused to share all registers on the processor core. According to the single-instruction-stream multi-fiber-thread programming model and the construction method thereof, the programming model is constructed based on the fiber threads, concurrent execution of asynchronous data streams is achieved, the resource utilization rate is effectively increased, and redundant calculation is eliminated.

The invention provides an automatic test generation platform and method for an operating system kernel derived function. The method comprises the following steps: (1) intelligent symbol extraction and analysis; (2) generating an LLM-driven test code; (3) performing automatic compiling and virtualization testing; and (4) intelligent error diagnosis and repair. According to the method, the test coverage degree is greatly improved, through the multi-level symbol context extraction technology, single function information is extracted, the dependency relationship of related functions in the same element is analyzed, a function call graph and data flow analysis are constructed, rich context information is provided for LLM, and the problem that traditional manual test case writing is incomplete in coverage is solved.

A test question recommendation method based on large language model adaptive multi-level evaluation constructs a multi-level architecture including semantic consistency verification, fine-grained fact alignment and dynamic cognitive evaluation, and comprises the following steps: constructing an initialized data stream and executing dynamic random sampling; calculating discrete semantic entropy by using NLI bidirectional implication clustering so as to quantify and eliminate illusion content with high uncertainty; under the RAG framework, the test questions are deconstructed into atomic statements, and a fact deviation is corrected by calculating a retrieval relevance vector and a logical implication consistency score; analyzing and screening low-quality texts based on multidimensional language features of syntax and logic; vector fusion is carried out on the generated intention and the cognitive portrait of the student, and a dynamic evaluation index system adaptive to a specific teaching scene is constructed in real time by utilizing context learning. The problems of accuracy and adaptability of automatic question setting are effectively solved, and intelligent closed-loop control from test question generation to cognitive alignment is realized.

The invention discloses a multi-language code generation method based on self-supervised pre-training, which comprises the following steps: acquiring and cleaning multi-language code data to form a training corpus; the method comprises the following steps: representing code data as an abstract syntax tree, extracting a control flow diagram and a data flow diagram of the code data, and obtaining unified semantic representation through combination of a diagram encoder and a sequence encoder; designing a self-supervised pre-training task, and pre-training the semantic representation based on the training corpus; constructing a multi-language pre-training model based on the structure-improved recurrent neural tensor network and the multi-language embedding matrix; when a user inputs a natural language, generating a target language code by using the multi-language pre-training model; and target language code correction is carried out through conventional function testing and grammar checking. According to the method, multi-channel recursive combination and a hierarchical recursive expansion mechanism are combined with self-supervised pre-training, so that accurate generation and performability improvement of cross-language codes are realized.

The invention discloses a hand-eye cooperative robot control system and method based on dynamic operator arrangement, an upper computer planning layer operates a master control computer to periodically trigger task scheduling, a joint controller of a lower computer execution layer receives an instruction through a redundant bus to perform servo control, hand-eye camera data triggers a visual assembly line through an interrupt event, and a visual assembly line is controlled through a control interface. According to the method, pressure is calculated through visual processing, motion planning and joint control in a load sharing mode, a double-buffering mechanism is adopted to enable current frame visual processing and previous frame motion control to be executed in an overlapping mode, real-time scheduling and parallel computing of tasks are completed, hot data are cached in a memory database, cold data are archived to an HDFS and migrated through an LRU strategy, and the real-time scheduling and parallel computing of the tasks are completed. Data type conversion is automatically derived and performed based on a feature type registry, and conditional branch execution is triggered according to real-time sensor data. According to the control system, cross-hardware plug and play, algorithm flow configurability and data flow real-time sharing are achieved, and the requirement for improving the efficiency of complex operation tasks is met.

A computer-implemented system analyzes program and data flows in a software system comprising code written in multiple programming languages using a generative large language model (LLM) directed by programming-language-specific prompts. The LLM identifies functional components within the code, generating labeled graph nodes that include a node type, a node name, and dependency information. A graph construction computer system processes the labeled graph nodes to generate a directed graph, where nodes represent functional components and directed edges represent dependencies. The system stores the graph in a database and provides a web-based interface for visualization, allowing users to explore, query, and analyze program and data flows across the software system. The system enables automated, language-agnostic dependency mapping, facilitating software analysis, debugging, and modernization.

A method, apparatus, and system are disclosed. The method includes constructing a static single assignment (SSA) form and a static single use (SSU) form for a program; setting a single-use disqualifying property locally for each SSU version, and propagating the single-use disqualifying property both forward and backward on an SSU graph uniquely formed from the constructed SSU form so it becomes a global property; transferring results from the SSU form to the SSA form to set a single-use property locally for each SSA version based on an occurrence of any use being associated with a disqualifying SSU version; performing data flow analysis on an SSA graph uniquely formed from the constructed SSA form so the single-use property becomes a global property to identify one or more definitions as dynamic single-use for variables in the program; and generating computer-readable instructions for executing the program based on the one or more definitions identified as dynamic single use for the variables in the program, wherein the one or more definitions identified as dynamic single-use has a defined value used exactly one time during execution of the program.

The invention provides an instruction-level parallel scheduling method and device in a deep learning compiler, and the method comprises the steps: decomposing a calculation graph of a deep learning task, decomposing a task represented by each calculation graph node on the calculation graph into a hardware calculation instruction according to a current task decomposition scheme, and obtaining an instruction set of the calculation graph; according to the data flow dependency relationship between the nodes on the calculation graph, obtaining a dependency relationship graph between the instructions in the instruction set; dividing the hardware calculation instructions in the instruction set into each sub-core in the processor; according to the dependency graph, determining a directed acyclic graph containing the dependency among the hardware calculation instructions on the sub-cores in the processor, and according to an emission model of the processor and the reciprocal throughput rate of the hardware calculation instructions, determining the execution sequence of the hardware calculation instructions divided to the sub-cores on the sub-cores, and the processor executes the instruction scheduling to obtain an execution result of the deep learning task.

The invention relates to the technical field of software security, and particularly discloses a real-time vulnerability detection and repair system of an AI code generation model. The system comprises a central coordinator, a real-time monitoring module, a deep analysis module, an intelligent repair module and a feedback learning module. The real-time monitoring module rapidly scans and triggers the generated codes; the deep analysis module performs interprocess stain analysis on the trigger code to generate a diagnosis report containing a precise data flow path; the intelligent repair module generates a repair patch under the function equivalence constraint according to the diagnosis report; and the feedback learning module collects the processing feedback of the developer on the repair patch, and drives the continuous optimization of each module of the system. According to the method, through layered progressive analysis and repair generation, the repair accuracy and the function safety are ensured; through a feedback learning mechanism, the self-adaptive evolutionary ability of the system is realized, the problems of unavailability of repair suggestions and static stiffness of the system in the prior art are solved, and real-time protection is provided for AI auxiliary programming.

A computer-implemented system analyzes program and data flows in a software system comprising code written in multiple programming languages using a generative large language model (LLM) directed by programming-language-specific prompts. The LLM identifies functional components within the code, generating labeled graph nodes that include a node type, a node name, and dependency information. A graph construction computer system processes the labeled graph nodes to generate a directed graph, where nodes represent functional components and directed edges represent dependencies. The system stores the graph in a database and provides a web-based interface for visualization, allowing users to explore, query, and analyze program and data flows across the software system. The system enables automated, language-agnostic dependency mapping, facilitating software analysis, debugging, and modernization.

Type definitions of user-defined types in application program code for which definitions are absent (“unknown types”) are inferred. A static analyzer implements two passes of a fixed-point type inference algorithm. Each pass encompasses a plurality of traversals of the application's control flow to build inferred definitions of unknown types until the inferred definitions are maximally built. To build an inferred definition, based on inferring a variable is an unknown type, the static analyzer infers member variables / functions of the unknown type based on contextual information associated with the variable. Type information of unknown types is propagated along control flow paths. After the first pass terminates, unknown types can be assigned known types based on matching of inferred definitions. Inferred definitions of remaining unknown types are incorporated into the application program code. A second pass of type inferencing and data flow analysis are then performed with the inferred definitions incorporated therein.

A system and method is provided for designing and optimizing hardware accelerators for neural networks. During a pre-design phase, rules are extracted from compilation patterns that describe conversion between neural network operators, coarse-grained operators, and fine-grained dataflow. A fast mapper for converting neural network models to coarse-grained operator descriptions and a dataflow mapper are generated. A coarse-grained design phase employs an architecture optimizer to generate plural provisional hardware accelerator designs The coarse-grained operator descriptions are simulated using a coarse-grained simulator to obtain performance metrics of each provisional accelerator design. A fine-grained design phase employs a dataflow mapper and fine-grained simulator to finalize provisional hardware accelerator designs. A hardware accelerator is generated from a finalized hardware accelerator design and a corresponding software toolchain is created including a compiler and software development kit (SDK) for programming, debugging, and deploying the hardware accelerator design.

The invention relates to the technical field of medical information parallel computing, in particular to a multi-thread parallel computing method and system for Brucella spondylitis diagnostic parameters, and the method comprises the following steps: a diagnostic task dependency graph construction step: constructing a directed acyclic graph based on the computing dependency relationship of the diagnostic parameters and determining the task priority; a self-adaptive thread scheduling step: adopting a work stealing strategy and a lock-free concurrent queue to realize dynamic load balancing; a vectorization parallel computing step: performing batch computing on the diagnosis parameters by using a single-instruction multi-data-stream instruction; and a feedback driving optimization step: dynamically adjusting a scheduling strategy according to performance indexes to form closed-loop optimization, and deeply coupling the four core modules to realize synergistic interaction.

The invention discloses a multi-disciplinary simulation task intelligent planning method based on a knowledge graph, and relates to the technical field of intelligent scheduling, and the method comprises the steps: constructing a preliminary knowledge graph comprising multi-disciplinary task description, resource limitation and execution dependence, analyzing historical task data through a pre-trained deep learning model, and obtaining a multi-disciplinary simulation task planning model; dynamically adjusting the preliminary knowledge graph through adaptive learning, and outputting an adaptive knowledge graph; carrying out resource redistribution optimization according to the updated task scheduling strategy, continuing to execute the task, starting monitoring, and collecting task progress and resource data as a new feedback data stream; the resource reallocation and the new feedback data stream are fed back to the adaptive knowledge graph, and multidisciplinary task description, resource limitation and execution dependency are updated through reasoning to generate a final optimization task execution sequence; the synchronous suppression of waiting time and resource contention is realized, the average utilization rate of resources and the time sequence stability are improved, and the adaptive adjustment capability of operation disturbance is formed.

A method implements static dataflow analysis for build pipelines. The method includes receiving a workflow file that includes an operation. The method further includes applying an extraction model to the workflow file to generate an extracted statement for the operation. The method further includes applying a statement model to the extracted statement to identify an unresolved parameter of the extracted statement. The method further includes applying the statement model to the unresolved parameter to generate a resolved parameter using a set of extracted statements including the extracted statement. The method further includes presenting an output statement including the extracted statement with the resolved parameter.

The application relates to an automatic code auditing method and device, computer equipment and a storage medium, comprising the following steps: obtaining the syntax structure, control flow and data flow of a code to be audited; constructing a context graph of the code to be audited according to the syntax structure, control flow and data flow of the code to be audited; obtaining a multi-modal collaborative vulnerability detection method, wherein the multi-modal collaborative vulnerability detection method comprises a static analysis method based on rule matching, a symbolic execution method based on a code path and a large model reasoning method based on semantic understanding, and the weights of the methods; and identifying one or more code vulnerabilities, the vulnerability types of the code vulnerabilities and the confidence according to the context graph of the code to be audited and the multi-modal collaborative vulnerability detection method. The method can improve the code auditing efficiency and process auditing codes of various vulnerability types.

The invention provides an evidence-driven large language model MISRA C rule review method and system for aviation safety. According to the method, formalized feature extraction is carried out from four aspects of grammar structure features, semantic structure features, logic analysis features and preprocessing behavior features for target codes according to various rules of MISRA C forcing class rules, and diagnosis information is diagnosed in combination with an industrial-grade compiler; generating a structured evidence set comprising abstract syntax tree node statistical features, code context structure features, control flow graph features, function internal data flow analysis features, symbol and type table features and macro definition analysis features; and based on the structured evidence set and the large language model, executing evidence tracing, rule matching and logical reasoning according to a preset thinking chain process, and outputting a structured judgment result containing illegal rule numbers, evidence description and code positions. According to the method, the false alarm rate can be remarkably reduced while the high rule coverage rate and the detection accuracy are kept.

The invention relates to a compiler design and code optimization technology, in particular to a compiler infrastructure based on multi-order intermediate representation and an implementation method of the compiler infrastructure. A traditional compiler adopts single intermediate representation, so that the coupling degree of the optimization process is high; the method is closely related to high-level / low-level languages, so that the compiler is poor in universality. In order to solve the problems, the invention provides the compiler infrastructure based on the multi-level intermediate representation and the implementation method of the compiler infrastructure, through modular design and hierarchical processing, the relevance of high-level / low-level languages is stripped, and the multi-level intermediate representation is designed, so that the problem that the compiler infrastructure cannot be used in the process of converting the high-level languages into the low-level languages is solved; the processing complexity of the control flow and the data flow is high, the universality is poor, and the subsequent maintenance difficulty is high. The method is suitable for real-time compiling, heterogeneous hardware adaptation and multi-language mixed programming and static optimization scenes, and the development and maintenance cost of the compiler can be greatly reduced.

Some embodiments are directed to systems and methods that dynamically generate user interfaces according to a user's task, context, or current environment. In one aspect, a computer system includes one or more processors and memory. The computer system obtains a stream of sensor data from one or more sensors. The computer system generates a context attribute based on the stream of sensor data, the context attribute characterizing a condition of a physical environment or a user, The computer system, based on the context attribute, determines a layout configuration of a user interface to be presented to the user. The computer system dynamically renders the user interface for the user based on the layout configuration and displays the user interface on the display of the user interface.

Various embodiments relate generally to data science and data analysis, computer software and systems, and computing architectures and data models configured to facilitate management and performance of enterprise functions, and, more specifically, to an enterprise computing and data processing platform configured to implement applications and computerized tools configured to facilitate implementation of an enterprise data model to generate and manage component data objects representing enterprise tasks and processes, as well as portions thereof, any of which may be used to create and modify an enterprise data flow or a process flow, for example, using an enterprise computerized tool. In some examples, a method includes activating one or more applications as computerized tools to access user account and enterprise function data, identifying permissions, and filtering access to logical components as a function of, for example, persona application data generated using a persona builder application.

The invention discloses a data flow diagram task scheduling method and device oriented to a data flow processor, a scheduler is used for scheduling the execution process of a plurality of DFU programs, each DFU program can be designed to occupy five possible versions of 1PE, 2PE, 4PE, 8PE and 16PE at most under the permissible conditions during parallel design, and the data flow diagram task scheduling method and device oriented to the data flow processor are provided. Different PE versions of each DFU program are stored in a DDR before execution, the different PE versions of each DFU program can be assumed to be Tetris in different shapes in the two-dimensional panel, the DFU programs are sequentially issued from the top end of the two-dimensional panel, and the execution process of the DFU programs is regarded to be issued from the top end of the two-dimensional panel to the bottom end of the two-dimensional panel, the number of PEs of the DFU programs corresponds to the transverse length of the Tetris, the execution time of the DFU programs corresponds to the longitudinal length of the Tetris, and when the CPU issues the multiple DFU programs from the top end in sequence in the mode that hardware is not specified, the scheduler modifies the length and width of the Tetris according to the current resource condition.

The invention discloses a fluid pulsation accelerator architecture for reinforcement learning of a large language model, and belongs to the technical field of artificial intelligence hardware acceleration. The system aims at solving the problems that when a current universal processor executes an RLHF working load, an instruction driving normal form is not matched with data flow calculation, fixed parallel granularity cannot adapt to a dynamic load, and the efficiency is low due to the fact that a framework does not sense data statistical characteristics. The core of the system is a liquid systolic array calculation fabric capable of being dynamically reconstructed, execution is triggered through data flow, and kernel boundaries and instruction overhead are eliminated. According to the system, a global lookup table subsystem is fused, and nonlinear calculation is optimized through merging query and parallel lookup by using data distribution prior; and a dynamic scheduling unit is configured, and elastic parallelism is realized by adopting a resource allocation algorithm supporting work stealing. According to the method, the throughput and the energy efficiency of RLHF training and a large language model reasoning stage can be remarkably improved, and a new design normal form is provided for a next-generation AI special computing architecture.

The application belongs to the technical field of computer architecture design, and discloses a unified accelerator design method and accelerator for diversified graph computing applications, which comprises the following steps: by comparing and analyzing the field-specific execution model of classical graph computing applications, observing the common data flow abstraction of different graph computing applications, establishing a graph computing execution model based on a sparse matrix, and abstracting the common part of the graph computing algorithm; based on the sparse matrix execution model, by analyzing the common hardware features of classical graph computing applications such as graph traversal, graph mining and graph learning, a field general chip architecture for graph computing is constructed, and general and efficient support for complex real-world scenarios is realized; by proposing a graph computing instruction set for a "scalar-vector-graph" heterogeneous core, the programmability of the graph computing chip is improved.

The invention relates to a code error feedback method combining static analysis and dynamic analysis, and mainly solves the problems of insufficient error feedback and low efficiency in an Online Judge system: reconstructing codes of correct and error versions, including alignment and renaming of variables, and rearranging statements on the premise of not influencing semantics; static and dynamic multi-level alignment is carried out, in the static alignment, an editing script with the minimum cost is calculated through a heuristic editing distance algorithm, an alignment block and a difference list are obtained, in the dynamic alignment, compiling execution is carried out on source codes of two versions, test cases which do not pass operation are dynamically printed through gdb, and execution track information of the test cases is dynamically printed through gdb; error positioning is achieved, data flow backtracking analysis and scoring are carried out on error positions, and error feedback is generated. The method has the characteristics of high execution efficiency, high accuracy and easier understanding of generated feedback, automatic feedback of error codes can be realized, and the learning efficiency of students in programming tasks is improved.

The invention discloses an online distributed parameter optimization method and system for intelligent modeling of a complex system, and relates to the technical field of electric digital data processing. The method comprises the following steps that: an end side adopts an adaptive recursive ridge regression algorithm to recursively update the weight of an output layer of a feedforward neural network in real time so as to quickly track the dynamic state of a data stream; when the edge side monitors that the performance continuously declines, node pruning based on the optimal data set is triggered, and node growth is selected and randomly configured; when concept drift is detected, the edge side uploads data and a model state, the cloud side carries out anti-forgetting deep optimization on the feature extraction module, and optimization parameters are downloaded to cooperate with reconstruction and deployment. The method solves the problems of insufficient dynamic working condition adaptability, low computing resource efficiency and easy knowledge forgetting in the prior art, realizes optimal balance of model precision, adaptive speed and resource consumption, and is especially suitable for online modeling of complex dynamic systems such as intelligent control of high-speed trains.