Systems and methods may automatically generate code for deep learning networks. The systems methods may provide a code generation framework for generating target specific code. The code generation framework may include one or more predefined class hierarchies for constructing objects of the generated code. The objects of the class hierarchies may provide an interface to predefined libraries of deep learning functions optimized for use on a target platform. The systems and methods may perform one or more optimizations on the code being generated.

The present invention discloses an application data synchronization method and an apparatus. When a first operating system and a second operating system are installed in a terminal, and a first application and a second application that have a same function are installed on the first operating system and the second operating system respectively, the method includes: when the second application runs on the second operating system, performing the function by using second application data, and updating the second application data, where the second application data is updated according to first application data, and the first application data is updated when the first application runs on the first operating system to perform the function; where the first application data and the second application data are stored in the terminal. By using the solutions, sharing of data of a same application between different systems is more convenient and less time-consuming.

An instance of universally shared generic code is generated. A runtime parameter enables the size of a stack frame on which local data can be stored to be determined. Dynamically sized locals can be stored on a stack enabling precise garbage collection reporting. One frame of the stack is allocated for each code segment to simplify GC reporting. A reporting region in the frame memory region comprises a count of locals and a location at which the local is found in the stack.

Systems and methods improve performance and resource-efficiency of Just-in-Time (JIT) compilation in a hypervisor-based virtualized computing environment. A user attempts to launch an application that has been previously compiled by a JIT compiler into an intermediate, platform-independent format. A JIT accelerator selects a unique function signature that identifies the application and the user's target platform. If the signature cannot be found in a repository, indicating that the application has never been run on the target platform, the accelerator generates and stores the requested executable program in shared memory and saves the signature in the repository. The system then returns to the user a pointer to the stored platform-specific executable. If multiple users of the same platform request the same application, the system recognizes an affinity among those requests identified by their shared signature, and provides each user a pointer to the same previously stored, shared executable.

Provided are a transmission method, terminal and system for application software. In the method, a first terminal creates a wireless hotspot; a second terminal joins a wireless network constructed by the wireless hotspot when joining request information sent by the second terminal is received; and interaction of application software with the second terminal is performed based on the wireless network. By the technical solution, the application software may be shared among these terminals, and the users do not have to download the application software from the network so as to save the data consumption of the users. Moreover, downloaded application software may also be shared among these terminals through application store clients on the terminals, so as to improve enthusiasm of the user in searching for and downloading the application software and facilitate popularization of the application store clients as well as the application software thereof.

Preloading enhanced application startup is disclosed. For example, a first local socket associated with a first copy of an executable program loaded in a memory receives a first instruction to launch a second copy of the executable program. The executable program executes in one of two modes, a server mode and an active mode, and the first copy of the executable program executes in the server mode. The first copy of the executable program is cloned to launch the second copy of the executable program, which is launched in the active mode. A third copy of the executable program associated with a second local socket is launched in the server mode. The third copy of the executable program is determined to be actively running, after which the first copy of the executable program is terminated.

An apparatus and method are described for a hardware transactional memory (HTM) profiler. For example, one embodiment of an apparatus comprises a transactional debugger (TDB) recording module to record data related to the execution of transactional memory program code, including data related to the execution of branches and transactional events in the transactional memory program code; and a profiler to analyze portions of the recorded data using trace-based replay techniques to responsively generate profile data comprising transaction-level events and function-level conflict data usable to optimize the transactional memory program code.

A method implemented in a data center management node including obtaining, from memory, a physical stack describing a configuration of platform components across multiple operating platforms on a data center infrastructure, generating, by a processor, a graph describing correlations between the operating platforms and the data center infrastructure based on a platform library, wherein the platform library describes configurations of the platform components for each of the operating platforms separately, generating, by the processor, one or more logical stacks based on the graph, wherein the one or more logical stacks indicate deployable configurations of the operating platforms without depicting the platform components, and representing the logic stack to a user.

Systems and methods may automatically generate code for deep learning networks. The systems methods may provide a code generation framework for generating target specific code. The code generation framework may include one or more predefined class hierarchies for constructing objects of the generated code. The objects of the class hierarchies may provide an interface to predefined libraries of deep learning functions optimized for use on a target platform. The systems and methods may perform one or more optimizations on the code being generated.

Dynamic sharing of RAM in a software-defined communication system includes storing program code in a flash memory, categorizing parts of the code into groups of transmit categories according to when a part of the code needs to be copied into a section of a RAM and then executed during a first state of a TX state machine and according to how another part of the code can be later fit into the same section and then executed during a second state. Similarly, parts of the code are categorized into groups of receive categories according to when a part of the code needs to be copied into a section of RAM and then executed during a first state of a RX state machine and according to how another part of the code can be later fit into that section and then executed during a second state of the RX state machine, to reduce the amount of RAM without sacrificing speed performance.