Intel Corporation

Advanced Micro Devices Corporation

Aiqudo, Inc

Michael Lai

Alice Kwong

Julie Wang

Dz-Ching Ju

Mark Robinson

Mei Ye

Dinesh Suresh

James W O'Neill

Embodiments described herein are generally directed towards systems and methods relating to a crowd-sourced digital assistant system and related methods. In particular, embodiments facilitate techniques to crowdsource the training of a language model of the crowd-sourced digital assistant system. The digital assistant device can generate new action datasets based on manual inputs detected by the digital assistant device. The manual inputs can be recorded as a set of instructions, which can be interpreted by another digital assistant device to reproduce the detected manual inputs based on a command received by the other digital assistant device. The digital assistant server can receive action datasets, maintain action datasets, and distribute action datasets to one or more digital assistant devices. In various embodiments, the digital assistant device or server can also determine whether received action datasets are related. Each digital assistant device in the described system can participate in the training of a language model that establishes relationships between action datasets and/or command templates based on user feedback received via the digital assistant device.

Sharing commands and command groups across digital assistant operations

A system and method for optimization of code with non-adjacent loops. A compiler builds a node tree, which is not a control flow graph, that represents parent-child relationships of nodes of a computer program. Each node represents a control flow statement or a straight-line block of statements of the computer program. If a non-adjacent loop pair of nodes satisfy predetermined conditions, the compiler may perform legal code transformations on the computer program and corresponding node transformations on the node tree. These transformations may make adjacent this pair of loop nodes. The compiler may be configured to perform legal code transformations, such as head and tail duplication, code motion, and if-merging, in order to make adjacent these two loop nodes. Then loop fusion may be performed on this loop pair in order to increase instruction level parallelism (ILP) within an optimized version of the original source code.

Proactive loop fusion of non-adjacent loops with intervening control flow instructions

More effective compiler optimizations provide improved cache utilization. The compiler optimizations include a structure layout optimization that leaves the physical layout of the structure fields intact and instead changes the access order to these fields. The compiler optimizations apply to arrays using array remappings analogous to structure layout optimizations in which the array is considered as a structure and the physical layout of the array remains intact and instead the access order to the array changes.

Structure layout optimizations

A method of optimizing program code performance is disclosed. Applications written in the Java language can execute slowly. To speed up the execution time of a Java application certain slower executing methods within the application can be written in native code. When the slower executing methods are written in native code some of the advantages, such as automatic garbage collection, of the Java language are lost. Further, the resultant interface function calls between the Java methods and the native methods can result in inefficient code. The method optimizes programs written using both Java methods and native methods, i.e. mixed mode code, by analyzing a set of interface function calls from native methods to a Java Virtual Machine. Then selects a plurality of the interface function calls to either eliminate or move within the program code by analyzing either the bytecode and the intermediate language, or both, for either inefficient code constructs or inefficient code containing native interface function calls to perform garbage collection.

Optimization of garbage collection code in the context of raw native interface function calls in the java programming language

Compilers are tools that generate efficient mappings from programs to machines A Java &ldquo;Just In Time&rdquo; runs as part of an application, and as such, it must be fast and efficient in its use of memory. To achieve good performance and further optimize code generation, the present invention introduces a method for optimizing Java performance using precompiled code. The method of the present invention first monitors the performance of program code during program execution. Then a list of program functions for possible native code compilation is created. The list may be created based upon static and dynamic analysis of the computer program. A plurality of program functions from said list of program functions is selected for optimization and native compilation. The selected program functions are precompiled into native program functions. The present invention also allows the precompiled native code reverted so that a user could explore the performance tuning until satisfactory.

Growing community of inventors

Sunnyvale, CA, United States of America

Michael Lai