11 results about "Kernel level" patented technology

Malware attacks seek to identify vulnerabilities that can be exploited by enumerating currently-executing processes in the operating system of a target device for injection of a malicious payload. By detecting process enumeration events occurring at the kernel level, known or suspected malware enumeration activity can be identified and mitigated.

The application discloses a Direct3D 12 to rendering metadata mapping method based on kernel level nested analysis, establishes a structure body and a global index of D3D metadata in a VKD3D compilation stage, creates D3D metadata UBO when starting, intercepts resource binding, dynamic state configuration, multi-view setting and drawing calling and the like operations of VKD3D when executing, captures related parameters and generates complete D3D metadata in combination with kernel analysis, groups metadata according to a mapping dictionary between system values dependent on D3D12 colorizers and metadata demand types, writes the grouped metadata into the UBO of the corresponding metadata group according to a single view or multi-view scene, combines the metadata into a drawing group, submits the command buffer to the GPU to complete rendering after batch recording of the drawing command, and realizes compatible running of the D3D12 application in the case that the GPU does not support the ShaderDrawParameters feature.

The invention discloses a Direct3D 12 depth migration semantic implementation method based on shader compensation, and the method comprises the steps: carrying out the hierarchical detection of GPU depth migration characteristics through a kernel after a system is started when a GPU does not support VKEXTdepthbiascontrol, constructing an adaptive strategy, and storing the adaptive strategy into a kernel mode cache; after the VKD3D and the application are started, the VKD3D creates a first historical buffer area and a three-level PSO (Particle Swarm Optimization) buffer area; when the application creates a shader and creates a D3D12 PSO, the VKD3D extracts parameters to compile a Vulkan shader module, analyzes a depth offset parameter query adaptation strategy, and generates a shader variant of a corresponding type; and constructing a PSO fixed template, generating a cache key, querying a PSO dynamic variant in a three-level PSO cache, binding the PSO dynamic variant when an application sets a pipeline state, and submitting a Vulkan command buffer area to a GPU (Graphic Processing Unit) to finish rendering after a drawing instruction is submitted.

The invention discloses a Linux system kernel-level external device starting control method and system. The method comprises the steps that kernel starting parameters are modified according to device information of all devices needing to be closed and transmitted to a Linux kernel; the method comprises the following steps: when a Linux kernel is started, reading and analyzing equipment information of equipment needing to be closed in kernel starting parameters, judging whether to-be-created equipment is the equipment needing to be closed or not according to the analyzed equipment information before starting and creating the equipment, selectively creating equipment nodes according to a judgment result, and if the to-be-created equipment is the equipment needing to be closed, selectively creating the equipment nodes according to the judgment result; if yes, equipment node creation of the corresponding equipment is stopped to realize equipment shutdown, and otherwise, equipment node creation of the corresponding equipment is carried out. The method can realize shutdown of specific equipment when the system is started, and ensures stable operation of the system.

An attention region-based DCNN accelerator, by periodically updating the attention region, has high recognition accuracy and low computational complexity; multiple computing cores are designed on the hardware, and sub-pictures are dynamically allocated to cope with different computing scenarios; fine-grained precision quantization is performed at the convolution kernel level, and a convolution kernel / output channel rearrangement method is proposed to support the calculation of convolution kernels with different precisions; real-time precision quantization is performed on the convolution window set, which can perform zero-rate detection on the output feature map in real time, and adaptive PEs and PE arrays are designed; the OS data flow system with the "two-dimensional data sliding" function can maximize the retention of data in the PE, thereby reusing Weight and Activation data in time and space, reducing data transmission bandwidth and additional power consumption caused by repeated data transmission, and supporting online convolution window set real-time zero-rate detection.

The invention discloses a GPU (Graphics Processing Unit) kernel characterization method. The method comprises the following steps: step 1, instruction standardization preprocessing: preprocessing an original GPU assembly instruction; 2, performing fine adjustment in a language model by utilizing the GPU assembly code corpus subjected to standardization preprocessing in the step 1, and retraining a special Tokenizer; step 3, construction of instruction semantic information; the method specifically comprises the steps of target instruction wrapping and context modeling; 4, performing fine tuning on the language model by using the shielding language modeling task, and generating a hidden state for the input target instruction by using the fine-tuned language model; and 5, averaging the embedded vectors of all the target instructions in each warp to obtain a warp embedded vector, and carrying out global fusion on all the warp embedded vectors to obtain a complete semantic vector representation of the GPU kernel. The GPU kernel characterization method supports multi-level modeling from an instruction level to a kernel level, and has good generalization ability and efficient reasoning performance.

Embodiments of the present application provide a method and device for determining switching frequency, a storage medium and an electronic device, wherein the method for determining switching frequency comprises: creating a device file and a load statistics entry, wherein the device file is used to indicate a communication interface between a kernel state and a user state, and the load statistics entry is information to be obtained in the case of performing a thread switching event, and the thread switching event is used to indicate an event of switching a thread between the kernel state and the user state; in the case of determining that there is a thread switching event, obtaining thread information corresponding to the thread switching event according to the device file, wherein the thread information is information corresponding to the load statistics entry; and determining a first switching frequency corresponding to the thread switching event according to the thread information. Through the embodiments of the present application, the problem that there is no method for monitoring a thread switching event in real time at the kernel level of an operating system in the related art can be solved.

The invention relates to a process control method and device, electronic equipment, a storage medium and a program product. The method comprises the steps that an eBPF program associated with a target real-time process is obtained, and the eBPF program is loaded into a kernel of a system; monitoring the target real-time process in a kernel through the eBPF program to obtain a running state of the system; and in response to the situation that the running state does not meet a preset state condition, outputting prompt information for indicating to regulate and control the target real-time process. Through the method, the target real-time process is accurately monitored on the kernel level under the condition that the kernel is kept safe and stable so as to obtain the running state of the system, and abnormal running conditions of the system can be found in time and guided to be regulated and controlled.

The invention relates to the technical field of computer application, in particular to a distributed data state acquisition system, which comprises the following hierarchical modules: an edge layer, a transmission layer, a storage layer, a calculation layer and a control layer. A distributed data state acquisition system is characterized in that an eBPF program is directly mounted in a kernel space, events such as system calling, network flow and file operation are captured, the overhead of switching between a user mode and a kernel mode of a traditional Agent is bypassed, when the node load is low, an eBPF probe of a non-key index is automatically suspended, and only basic monitoring is reserved; the edge node caches the original data, only uploads an aggregation result after calculating a mean value and a quantile, and reduces 80% of transmission data volume; a filtering rule is loaded through a WASM module, invalid data processing is reduced by 70%, direct collection is carried out on a kernel layer based on an eBPF, and CPU occupation is smaller than 0.5%.

The invention discloses a control method and device for training a model and electronic equipment, and the method comprises the steps: packaging calculation and communication tasks, which can be executed in parallel, in a single fusion kernel according to an obtained model calculation graph in a model training initialization stage; when the astronomical large model is iteratively trained through the training data set, according to performance state information collected in real time, current operation configuration parameters are adjusted to obtain target configuration parameters, and a target execution strategy of a fused kernel is generated; and controlling the fusion kernel, executing the task based on the target execution strategy, and returning to the step of collecting the performance state information until a preset iteration condition is met. Therefore, before model training, computing and communication tasks which can be executed in parallel are fused at a kernel level, so that communication waiting overhead is reduced. In the model training process, operation configuration parameters are dynamically adjusted through performance state information collected in real time, efficient fusion and overlapping execution of communication and calculation are achieved, and the model training efficiency is improved.

The application discloses a Direct3D 12 depth offset semantic implementation method based on shader compensation, and when GPU does not support VK_EXT_depth_bias_control, the system starts after the kernel level detects GPU depth offset characteristics, and builds an adaptive strategy and stores it in the kernel state buffer; after VKD3D and application start, the first history buffer and three-level PSO cache are created by VKD3D; when the application creates a shader and creates a D3D12 PSO, the VKD3D extracts parameters, compiles a Vulkan shader module, analyzes depth offset parameters, queries an adaptive strategy, and generates a corresponding type of shader variant; a PSO fixed template is built, a cache key is generated to query a PSO dynamic variant in the three-level PSO cache, a PSO dynamic variant is bound when a pipeline state is set in the application, and after a drawing instruction is submitted, a Vulkan command buffer is submitted to the GPU to complete rendering.