89 results about "Video memory" patented technology

This application discloses a method and apparatus for managing video memory blocks, belonging to the field of data processing technology. The method includes: acquiring the occupancy status and access activity curves of video memory blocks within multiple consecutive address segments; for each consecutive address segment, performing frequency domain transformation and encoding based on the occupancy status and access activity curves of the video memory blocks within the segment to generate a video memory convolution spectrum characterizing the fragmentation degree and temporal access activity of that segment of video memory; determining candidate migration regions based on the video memory convolution spectra of each consecutive address segment; determining the differential tensors of candidate video memory blocks within the candidate migration regions; performing a comprehensive evaluation based on the differential tensors of all candidate video memory blocks under preset bandwidth and quality of service constraints to generate a video memory migration plan; and calling the underlying interface to execute the batch migration of candidate video memory blocks according to the video memory migration plan.

The invention provides a super-division method based on a graph driving program, a super-division device based on the graph driving program, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: determining a target super-resolution parameter under the condition of determining to start image super-resolution processing based on a submission time interval of a final frame in computer equipment; based on the hardware operation information of the computer equipment, determining a target execution device used for executing image super-division processing in the computer equipment and a corresponding target super-division algorithm; the hardware operation information represents the rendering time consumption of the computer equipment, the occupation condition of a video memory and the use condition of a hardware acceleration unit; and performing image super-division processing on the submitted final frame based on the target super-division parameter, the target execution device and the target super-division algorithm to obtain a super-division frame.

The present application relates to the technical field of image processing, and more particularly to a video decoding and Unreal Engine rendering method based on GPU full-link zero-copy, which comprises the following steps: step one, hardware decoding of an input video code stream on a GPU, and construction of a motion vector description buffer for storing compressed domain motion vector information in macroblock order; step two, reading of the motion vector description buffer by a calculation shader on the GPU to generate a dynamic special effect control buffer corresponding to each macroblock; and step three, in the post-processing material of the Unreal Engine rendering module, taking a shared video texture resource as an input texture and outputting a video picture. The present application realizes unified management of pixel data and motion data without relying on a host processor and without excessively occupying video memory bandwidth, significantly reduces delay, and improves special effect stability and direction consistency.

This invention discloses a method for displaying Linux container applications on the HarmonyOS platform based on virtual compositing nodes. Using HarmonyOS as the host and Linux as the container, the host-side Vulkan resource management hub pre-checks GPU characteristics, allocates shared Vulkan resources, creates rendering channels, and initializes global timeline semaphores to build a unified resource pool. After the container starts, it requests resources from the host to import shared video memory objects, and the container compositor registers virtual nodes. During the rendering phase, the application uses shared resources to complete parallel rendering and sends notifications to the container compositor. The container compositor, based on the global display state, uses Vulkan instructions to perform window blending, transformation, and other processing, encapsulates global compositing units, and submits them to the HarmonyOS global compositing scheduler. The scheduler reads the global compositing units and, when the timeline semaphore meets the on-screen conditions, completes the global compositing of the container application window and the host's native window, driving the display service to output the screen.

The present application belongs to the technical field of AI. Disclosed are a method for determining a output result of a model and a related apparatus. The method comprises: acquiring first input information of a network model; if the first input information hits a first intermediate operation result in a key-value cache, performing dimension-increasing processing on the first intermediate operation result by means of the network model to obtain a first key-value pair; and, on the basis of the first key-value pair, determining a first output result by means of the network model, the first output result being an output result corresponding to the first input information. The data volume of the first intermediate operation result is less than the data volume of the first key-value pair, that is, the key-value cache in the present application has a smaller data volume and occupies less storage space. Using intermediate operation results with a smaller data volume can increase the data loading speed of key-value caches and reduce the occupation of video memory by data loading, thereby accelerating reasoning and enhancing the overall operation performance of devices.

The application discloses a game picture acquisition method and device, a medium and a program product, and relates to the technical field of computers. The method comprises the following steps: according to the mode characteristics of a target instruction for controlling the sharing permission of a texture resource, an instruction sequence in a graphics rendering library is analyzed by using an assembly engine to identify the target instruction; the target instruction is modified, and a shared texture object is created in the video memory of an image processor after the modification; and a game rendering texture is copied to the shared texture object, so that an external acquisition process can obtain a real-time game picture by accessing the shared texture object. The target instruction for controlling the sharing permission of the texture resource in the graphics rendering library can be automatically located and modified, manual address updating and version updating are avoided, and the stability and reliability of game picture acquisition are improved.

This invention discloses a method and system for optimizing game resources in a virtual game. The invention relates to the technical field of virtual games. It loads the current load of the virtual game based on a target resource demand sequence and constructs the resource distribution of the virtual game. Then, it performs topological clustering based on the content to be loaded in the virtual game, generating a resource topology map containing multiple game resource clusters, thus improving the accuracy of the resource topology map. The node dependencies in the resource topology map are marked, and flexible scheduling is performed based on the resource content allocated to each resource cluster, thereby outputting a multi-level flexible scheduling strategy. The high-speed video memory area and low-speed memory area of ​​the virtual game are optimized along this flexible scheduling strategy. Resource optimization events of the virtual game are determined based on the smooth constraints of the virtual game. Furthermore, resource anomaly signals of the virtual game are used to determine optimization configuration measures for the high-speed video memory area and low-speed memory area in terms of resource dimensions, improving the accuracy of the optimization configuration measures.

This invention discloses a heterogeneous acceleration system and method for analog modulation recognition of large-scale multi-channel signals, comprising: a CPU host processing end and a GPU device computing end; the CPU host processing end is used for asynchronous pipelined processing of data reading, task scheduling, and result distribution, realizing parallel execution of data reading, GPU computing, and result distribution on the time axis; the GPU device computing end completes the entire process of signal preprocessing and deep learning model inference within the GPU memory after the original signal is transmitted to the video memory via the PCIe bus, and does not exchange data with the CPU host processing end until the recognition result is generated. This invention eliminates data transport and I / O blocking through a three-stage asynchronous pipeline architecture and a closed-loop processing link in the entire video memory, and combines parallel preprocessing operators, inference acceleration engines, multi-stream scheduling mechanisms, and dynamic video memory pools to achieve high throughput and low latency recognition of large-scale multi-channel analog modulation signals.

The application relates to the technical fields of multimedia image processing, GPU computing power scheduling and soft power-on resource optimization, and discloses a resource occupation prediction method in a video special effect processing process, which is executed by a terminal built-in processor and comprises the following steps: collecting a video special effect frame time interval, a GPU rendering pipeline thread occupation ratio and a video memory fragment distribution ratio; performing dimensionless normalization processing on the video special effect frame time interval to obtain a special effect frame time normalization factor; counting a special effect superposition level; and calculating a special effect time coupling factor, a pipeline blocking loss factor and a total resource occupation prediction value and outputting the total resource occupation prediction value. The application solves the problem of large prediction deviation in the prior art, realizes accurate prediction through a core innovation point, guarantees reasonable allocation of video special effect rendering resources, and improves rendering stability.

The present disclosure provides an extended video memory-based model training method, a model inference method, an apparatus, a device, a medium and a product, relating to the technical field of storage devices. The extended video memory-based model training method comprises: sending a first DSM instruction to an SSD controller, and performing multiple rounds of iterative training by a GPU. Each round of iterative training comprises multiple processing stages. The processing stages comprise: the SSD controller loading data units before calculation corresponding to hierarchical identifiers to the GPU, and the GPU calculating calculation units based on the data units. The first DSM instruction marks the priority of the calculation units. The GPU unloads the calculation units to the SSD. The SSD controller unloads the calculation units of high priority to SLC and unloads the calculation units of low priority to QLC. Through the technical solution of the present disclosure, the access delay of high-sensitive data is reduced without increasing the hardware cost, and the invalid erasing and writing of SLC is reduced to prolong the service life of the SSD.

This application discloses a data processing method, apparatus, computer device, and storage medium, belonging to the field of computer technology. The method includes: acquiring three-dimensional model data, which is used to render to a screen; rasterizing the three-dimensional model data to obtain multiple fragments, adding each fragment to a linked list of its respective pixel position, each fragment including a pixel position and depth; for any pixel position on the screen, merging the multiple fragments in the linked list according to their depth order to obtain display information for the pixel position; and displaying the image on the screen based on the display information for each pixel position. This application uses a dynamic linked list to maintain multiple fragments corresponding to each pixel position, enabling correct merging without relying on hardware mechanisms, reducing GPU computing resources, saving video memory, and improving rendering performance.

The application discloses an image generation method and related devices, equipment and storage media, wherein the image generation method comprises: obtaining an image generation request; obtaining the expected coordinates of an instance object based on the image generation request; normalizing based on the expected coordinates of the instance object, and determining the index token of the instance object in the index token corresponding to each of a plurality of preset normalized coordinates; constructing a prompt text instruction based on the index token of the instance object; and obtaining a target generated image for responding to the image generation request based on the model generated image output by the generative model in response to the prompt text instruction. The above scheme can reduce the calculation and video memory overhead of image generation, and improve the accuracy of the spatial layout of the instance object in the generated image, especially in a multi-instance object scene.

The application discloses an AI vision-based video VLM model inference acceleration method, comprising the following steps: S1, synchronously collecting a video stream and hardware state parameters; S2, dynamically sparsifying sampling to generate a sparse frame sequence; S3, allocating a continuous video memory physical address pool; S4, using an improved VideoMAE model and constructing a video memory direct-reading forward propagation layer to obtain a base address pointer to skip a standard interface, performing streaming calculation to reconstruct features in an SM, and outputting a compressed Token sequence; S5, using remaining space to store key-value caches to perform large model calculation to output semantic features; S6, using an Orca algorithm to insert micro-batch calculation during SM idle periods to cover up delayed output inference texts; and S7, releasing the caches and dynamically adjusting a frame rate and a compression ratio to perform a cycle. The application eliminates redundant copying and significantly improves inference throughput.

Belonging to the field of computer vision technology, this invention provides a method and apparatus for teacher-assisted atomic motion recognition based on an improved end-to-end network. [Solution] The method determines time-dimensional and spatial-dimensional video features from the target teacher's current lesson video based on an improved target end-to-end network, generates multi-window feature groups, merges the multi-window feature groups using a frame selection network, and recognizes the target teacher's atomic behavior based on the fusion result. This method embeds spatially adaptive units and time-series adaptive units within the end-to-end network, performs feature extraction on each, and reduces video memory consumption by training only the parameters of these units during network training, thereby effectively improving the accuracy and efficiency of atomic behavior recognition, reducing memory requirements, and lowering deployment difficulty.

The present disclosure provides a model inference method and device and electronic equipment, relates to the technical field of computers, in particular to the technical fields of artificial intelligence, large models, front ends, data processing, memory and video memory resource management, and the like. The specific implementation scheme is as follows: in response to an inference request, model data of an inference model is obtained, and the model data is loaded into a memory buffer; wherein the model data includes model structure metadata and weight data; according to the model structure metadata that has been loaded into the memory buffer, video memory resources required by the inference model are created, and the weight data loaded into the memory buffer is written into the video memory resources; the weight data in the video memory resources is read by a graphics processing unit, and the calculation logic of the inference model is executed to obtain an inference result.

This invention discloses a method and system for optimizing the dynamic loading efficiency of ultra-large scenes in 3D engines, belonging to the field of computer game design technology. It aims to solve problems such as high computational overhead, resource lag, and fragmented video memory in existing scene loading methods. The method includes: constructing a dynamic sparse octree index to achieve ordered spatial data organization; predicting future positions based on viewpoint motion vectors and constructing pre-loaded view frustums; establishing a multi-threaded asynchronous I / O scheduling mechanism to read and convert data in parallel according to priority; implementing video memory pooling management and reference counting strategies to achieve resource reclamation and reuse by page blocks; and using a time-weighted linear interpolation algorithm to achieve smooth switching between multiple levels of detail. The system includes corresponding index construction, state prediction, streaming scheduling, pooling management, and dynamic update modules. Through the above solutions, this invention significantly improves the real-time performance of data loading and hardware utilization, effectively eliminating screen stuttering, white models, and visual jump phenomena.

The application discloses a three-dimensional engine rendering control method based on adaptive deep learning. By collecting scene geometric complexity, texture complexity, processor utilization, video memory occupation and historical rendering performance data, a rendering state feature vector is constructed, and an initial rendering parameter is generated by inputting the prediction model; on this basis, an exploration search mechanism is introduced to perform multidimensional trial on the parameter, and candidate parameters are screened in combination with the trial rendering result; further based on a directional search strategy, the parameter optimization direction is determined and iterative search is performed to obtain refined rendering parameters; the optimized parameters are applied to the three-dimensional engine to complete rendering processing, and performance and image quality data are synchronously acquired; meanwhile, a rendering cycle data feedback mechanism is constructed to realize dynamic updating of the parameters and the state. The method fuses the prediction and search strategies, realizes collaborative control of rendering efficiency and image quality, and has adaptive adjustment and cyclic optimization capabilities.

This invention provides an automated testing method and system for segment LCD screens. The method includes: establishing a data path between a test platform and the device under test (DUT), wherein the data path establishment strategy is adaptively selected based on the architecture of the DUT; pre-establishing a video memory mapping configuration file and test cases, and loading them onto the test platform; establishing a frame synchronization mechanism when the test platform receives video memory data from the DUT; detecting whether the DUT has physical faults or electrical connection anomalies, and generating physical test results; translating the segment codes of the test cases into bit arrays based on the video memory mapping configuration file, and comparing them with the actual arrays obtained by parsing the actually received video memory data; and generating a corresponding diagnostic report based on the comparison results and the physical test results.

The application discloses a low-delay scheduling method and system for real-time interactive AI services and a medium, the method comprising: determining an initial priority coefficient according to each tenant level and task type; dynamically analyzing the priority coefficient of each task request according to the waiting time; predicting and dynamically allocating the video memory space capacity of a first cache area; obtaining the real-time residual amount of the video memory capacity of the predicted and allocated cache area, dynamically analyzing the upper limit value of the number of task requests processed in parallel, and based on the number of currently waiting task requests and the preset regulation mode, analyzing the number of currently optimal task requests processed in parallel to balance resource utilization and delay waiting time. The application effectively solves the delay problem of existing AI services when processing high real-time and high-interactive tasks, improves the service response speed and efficiency of real-time interactive tasks, and meets the strict performance index standards of various high-demand AI services.

The present disclosure provides a distributed model training task execution method and device, equipment, medium and product, relates to the field of artificial intelligence, in particular to the field of deep learning. The specific implementation scheme is: obtaining a plurality of shard parameters required by a target network layer indicated in a target computing task; in the process of calculating the plurality of shard parameters in the target network layer, adjusting the number threshold of the shard parameters that have completed calculation and can be retained in the video memory and the network layer depth of the shard parameters pre-fetching according to the available capacity of the video memory and the current calculation stage of the target network layer; determining a pre-fetch network layer corresponding to the network layer depth and located after the target network layer based on a preset network layer execution sequence, and pre-fetching at least part of the shard parameters required by the pre-fetch network layer; in the case that the number of the shard parameters that have completed calculation in the video memory does not reach the number threshold, retaining the plurality of shard parameters that have completed calculation in the target network layer to the video memory for use by the pre-fetch network layer.

This invention discloses a rendering method, system, and storage medium for skeletal animation. The method includes: acquiring animation control parameters for each model instance, the animation control parameters including an animation identifier, the current playback frame, and a world transformation matrix, but excluding the skeletal transformation matrix; writing the animation control parameters of each model instance into an instance data buffer and sending a single-draw instruction to the GPU, the instance data buffer being located in GPU video memory and configured to be writable by the CPU and readable by the GPU; the GPU responding to the single-draw instruction by sampling the skeletal transformation matrix from a pre-stored animation matrix texture based on the animation control parameters in the instance data buffer, and performing parallel skinning calculations on each model instance based on the sampled skeletal transformation matrix. This method frees the CPU from skeletal matrix calculations, avoiding CPU rendering thread overload and preventing severe frame rate fluctuations and frame drops.

The application relates to a GPU progressive optimization method for neural operator backbone network reasoning and belongs to the electronic information technical field.The method comprises the following steps: step S1, initializing constraints and tolerances;step S2, obtaining baseline execution time and reference results through baseline measurement;step S3, obtaining an estimated split value based on a working set model and expanding;step S4, performing KSweep evaluation and consistency checking, and preferably obtaining; a process of scanning and measuring each candidate set is called KSweep; step S5, fixing, adding fusion reasoning optimization, and performing consistency verification, if the consistency verification is passed, outputting and saving the optimal configuration, and directly loading and executing in the running stage; otherwise, backtracking or adjusting and recalibrating, and returning to step S3. The application shortens the reuse distance between adjacent layers, reduces the round-trip overhead of intermediate activation between the video memory and the cache, and thus improves the reasoning efficiency.

This invention discloses a Gaussian radiance field rendering optimization method based on proxy depth and hierarchical indexing, comprising: constructing a Gaussian radiance field scene and proxy mesh using existing images; constructing a hierarchical spatial index of the Gaussian radiance field; rendering the proxy mesh onto a depth buffer using GPU hardware rasterization units to obtain the proxy mesh depth; applying a pixel-by-pixel offset oriented behind the viewing direction to the proxy mesh depth; calculating the offset depth based on the proxy mesh depth; performing Gaussian cluster culling based on cell partitioning based on the offset depth; and sorting and rendering the Gaussian primitives in the visible Gaussian clusters. This invention significantly improves the average rendering frame rate in multi-scale scenes and effectively reduces the video memory overhead of rendering large-scale Gaussian radiance field scenes, while maintaining high-quality novel view compositing effects, providing strong support for real-time rendering applications of large-scale Gaussian radiance fields in cross-platform scenarios.

This invention relates to a method, unit, and apparatus for anomaly simulation in a large-scale signal processing algorithm business chain, comprising the following steps: Step 1: Based on the testing requirements of the signal processing business chain, identify the signal processing algorithm software that needs to be embedded with the anomaly simulation unit and the business chain links that need to run the anomaly simulation unit independently; Step 2: Configure the anomaly simulation unit, enable the corresponding sub-units, and initialize the behavior mode, utilization rate, and quantity parameters of each sub-unit. This invention supports the simulation of anomalies in various hardware resources such as CPU, GPU, memory, video memory, and network, and can quantitatively assess the propagation impact of anomalies in the business chain.

The application discloses a kind of based on localization platform core video decoding display system, including sequentially connected CPU, core graphics card and display, CPU includes several cores, core graphics card includes decoding unit, image processing unit and display unit, and decoding unit is equipped with API interface.This based on localization platform core video decoding display system uses localization processor platform to decode and display video using core graphics card, reduces the load of CPU, and compared with independent core graphics card, using core graphics card for decoding has the advantages of small size, low overall power consumption, cheap;CPU sends the video data supported by core graphics card to the video memory of core graphics card, and keeps the video data not supported by core graphics card in the memory of CPU, solves the problem that directly sending screen data to core graphics card for decoding in prior art will cause format not supported, resulting in black screen, and serious will cause core graphics card blocking exception.

The present application relates to a convolutional neural network-based image color fusion method, device and storage medium, applied to the technical field of image processing, comprising: realizing color adaptation of foreground and background under the double constraints of foreground mask and skin segmentation mask through a color fusion model, optimizing the color naturalness of the skin area, effectively solving the problems of rigid fusion and strange skin color in the prior art, and improving the stability of the method; The high-resolution processing strategy of "low-resolution inference + detail preservation color migration" not only avoids the problems of large video memory consumption and long time consumption of direct high-resolution inference, but also solves the problems of loss of global color information in block inference and introduction of edge abnormalities in traditional high-frequency information recovery; The detail preservation color migration model can accurately extract the color features of the low-resolution fusion result and completely retain the texture details of the high-resolution original image, realizing the dual goals of color fusion and detail preservation; Automatic processing significantly reduces time and labor costs.