Qualcomm Incorporated

Ati Technologies Ulc

James Lyall Esliger

Wilson Kwan

Meghal Varia

Serag Gadelrab

Kyle Ernewein

Alwyn Dos Remedios

George Lee

Milivoje M Aleksic

Mark Bapst

Denis J Foley

Sumit Sur

Aris Balatsos

Kevin O'Neil

James Michael Artmeier

Robert Dreyer

Michael Douglas Sharp

Bruce Plotnick

Certain aspects of the present disclosure provide techniques for concurrently performing inferences using a machine learning model and optimizing parameters used in executing the machine learning model. An example method generally includes receiving a request to perform inferences on a data set using the machine learning model and performance metric targets for performance of the inferences. At least a first inference is performed on the data set using the machine learning model to meet a latency specified for generation of the first inference from receipt of the request. While performing the at least the first inference, operational parameters resulting in inference performance approaching the performance metric targets are identified based on the machine learning model and operational properties of the computing device. The identified operational parameters are applied to performance of subsequent inferences using the machine learning model.

Concurrent optimization of machine learning model performance

Various embodiments of methods and systems for proactive resource allocation and configuration are disclosed. An exemplary method first compiles and links a profile instrumented application with a compiler comprising a profile guided optimization feature that inserts calls to a profiler runtime. The profile instrumented application is executed on a target device using one or more workload datasets representative of probable workloads. During execution, based on recognition of the inserted calls, an instrumentation-based profile dataset is generated in association with each of the one or more workload datasets. Next, the profile instrumented application is recompiled and relinked based on the instrumentation-based profile datasets to create a set of profile guided optimizations to the source code, thereby resulting in an optimized application. The optimized application may be executed and monitored to generate a revised profile dataset useful for providing instructions to the target device for optimal workload allocation and resource configuration.

Application profile driven scheduling and configuration in a system on a chip

Systems and methods for dynamic granularity control of parallelized work in a heterogeneous multi-processor portable computing device (PCD) are provided. During operation a first parallelized portion of an application executing on the PCD is identified. The first parallelized portion comprising a plurality of threads for parallel execution on the PCD. Performance information is obtained about a plurality of processors of the PCD, each of the plurality of processors corresponding to one of the plurality of threads. A number M of workload partition granularities for the plurality of threads is determined, and a total execution cost for each of the M workload partition granularities is determined. An optimal granularity comprising a one of the M workload partition granularities with a lowest total execution cost is determined, and the first parallelized portion is partitioned into a plurality of workloads having the optimal granularity.

System and method for dynamic granularity control of parallelized work in a portable computing device (PCD)

A method and apparatus for reducing net power consumption in a computer system includes identifying a plurality of processing states operable to execute a task. A processing state and current drain pattern is selected that is most power efficient. A selected processing state may include one or more processing elements of the computer system such as one or more processors or accelerators and indicates the manner in which one or more portions of the received task may be distributed among the processing elements of the computer system. The current drain pattern selected may be a constant current drain pattern or a pulsed current drain pattern and may be selected to optimize power consumption when executing the task among the one or more processing elements.

Method and apparatus for optimizing power consumption in a multiprocessor environment

An integrated circuit () may receive a boot loader code () via a debug access port (), wherein a boot logic is operative to block, upon a reset () of the programmable processor () from the debug access port (), commands and to the programmable processor from the debug access port, while still allowing the reset () command and while allowing write access to memory () to receive the boot loader code image () written to memory (). The boot logic also blocks commands to the memory subsystem () from the debug access port and turns off write access to memory () after allowing the boot loader code image () to be written. The boot logic validates the boot loader code image () by performing a security check and jumps to the boot loader code image () if it is valid, thereby allowing it to run on the programmable processor (). The boot logic may be logic circuits, software or a combination thereof.

Integrated circuit with secure boot from a debug access port and method therefor

A portable device includes a controller that is responsive to a remaining power capacity of the battery, and a power consumption level of the portable device and based on user prioritized functional processing capability features, dynamically controls functional processing capability features of the device. The controller provides power for a higher priority feature at the expense of a lower priority functional processing capability feature consistent with the user prioritized functional processing capability features. A wireless portable device is also disclosed that includes a wireless signal strength determinator that determines a received signal strength of the wireless device and a controller that adjusts the functional processing capability feature of the wireless device based on the determined received signal strength and based on battery capacity information. In this way, received signal strength information is used as a further input to determine how to best adjust functional processing capability features of the device.

Portable device with priority based power savings control and method thereof

An electronic device having a processor powered by a power source may be operated by providing a plurality of program portions individually executable by the processor for performing the same computing function. Each program portion causes the processor to exhibit a different instantaneous power consumption profile while performing the computing function. A particular program portion is selected based on at least one characteristic of the power source and executed on the processor to perform the computing function.

Growing community of inventors

Richmond Hill, Canada

James Lyall Esliger