Microsoft Technology Licensing, LLC

Paul F Ringseth

Genevieve Fernandes

Niklas Gustafsson

Donald James McCrady

Yosseff Levanoni

Weirong Zhu

William Robert Messmer

Bimal Kumar Mehta

Lingli Zhang

Rick Molloy

Marko Radmilac

Jonathan David Goldstein

Gueorgui Chkodrov

Roger Shane Barga

Charles David Callahan, Ii

Paul Maybee

Momin Mahmoud Al-Ghosien

Tihomir Tsvetkov Tarnavski

Philip Lucido

Andy Shen

Rahul V Patil

Joseph L Hellerstein

Satish Ramchandra Thatte

David S Callahan

Sanjib Saha

Johannes Klein

Lee B Graber

Jasjit Singh Grewal

Jonathan E Caves

Sriram Balasubramanian

John Joseph Duffy

Stephen Harris Toub

Vance Palmer Morrison

Roland Lucas Fernandez

Eric Lynn Eilebrecht

Huseyin Serkan Yildiz

Michael L Chu

Christian Beaumont

Jason J Shirk

Stephen S Toub

Bill Messmer

Sridhad S Madhugiri

Jasjit S Grewel

A high level programming language provides a read-only communication operator that prevents a computational space from being written. An indexable type with a rank and element type defines the computational space. For an input indexable type, the read-only communication operator produces an output indexable type with the same rank and element type as the input indexable type but ensures that the output indexable type may not be written. The read-only communication operator ensures that any attempt to write to the output indexable type will be detected as an error at compile time.

Read-only communication operator

A high level programming language provides an agile communication operator that generates a segmented computational space for distributing the computational space across compute nodes. The agile communication operator decomposes the computational space into segments, causes the segments to be assigned to compute nodes, and allows the user to centrally manage and automate movement of the segments between the compute nodes. The segment movement may be managed using either a full global-view representation or a local-global-view representation of the segments.

Agile communication operator

A high level programming language provides a nested communication operator that partitions a computational space. An indexable type with a rank and element type defines the computational space. The nested communication operator partitions a specified dimension of an index indexable type into segments specified by a segmentation vector and returns an output indexable type that represents the segments. By doing so, the nested communication operator allows data parallel algorithms to operate on the segments as individual units.

Nested communication operator

Computer systems attempt to manage resource pools of a dynamic number of similar resources and work tasks in order to optimize system performance. Work requests are received into the resource pool having a dynamic number of resources instances. An instance-throughput curve is determined that relates a number of resource instances in the resource pool to throughput of the work requests. A slope of a point on the instance-throughput curve is estimated with stochastic gradient approximation. The number of resource instances for the resource pool is selected when the estimated slope of the instance-throughput curve is zero.

Managing pools of dynamic resources

A high level programming language provides extensible data parallel semantics. User code specifies hardware and software resources for executing data parallel code using a compute device object and a resource view object. The user code uses the objects and semantic metadata to allow execution by new and/or updated types of compute nodes and new and/or updated types of runtime libraries. The extensible data parallel semantics allow the user code to be executed by the new and/or updated types of compute nodes and runtime libraries.

Extensible data parallel semantics

A runtime environment allows a scheduler in a process of a computer system to be finalized prior to the process completing. The runtime environment causes execution contexts that are inducted into the scheduler and execution contexts created by the scheduler to be tracked. The runtime environment finalizes the scheduler subsequent to each inducted execution context exiting the scheduler and each created execution context being retired by the scheduler.

Scheduler finalization

A high level programming language provides a tile communication operator that decomposes a computational space into sub-spaces (i.e., tiles) that may be mapped to execution structures (e.g., thread groups) of data parallel compute nodes. An indexable type with a rank and element type defines the computational space. For an input indexable type, the tile communication operator produces an output indexable type with the same rank as the input indexable type and an element type that is a tile of the input indexable type. The output indexable type provides a local view structure of the computational space that enables coalescing of global memory accesses in a data parallel compute node.

Tile communication operator

A scheduler in a process of a computer system allows an external execution context to execute on a processing resource allocated to the scheduler. The scheduler provides control of the processing resource to the external execution context. The scheduler registers for a notification of an exit event associated with the external execution context. In response to receiving the notification that the exit event has occurred, the scheduler regains control of the processing resource and causes a task associated with an execution context controlled by the scheduler to be executed by the processing resource.

Regaining control of a processing resource that executes an external execution context

A high level programming language provides an agile communication operator that generates a segmented computational space based on a resource map for distributing the computational space across compute nodes. The agile communication operator decomposes the computational space into segments, causes the segments to be assigned to compute nodes, and allows the user to centrally manage and automate movement of the segments between the compute nodes. The segment movement may be managed using either a full global-view representation or a local-global-view representation of the segments.

A concurrent data structure allows synchronization to be elided for read accesses. Processing resources that remove one or more elements of the concurrent data structure are allowed to delete the elements only after all other processing resources have reached a safe point. Each processing resource maintains an indicator that indicates whether the processing resource has reached as safe point (i.e., will not access the concurrent data structure). When the indicators indicate that all processing resources have reached a safe point, elements of the data structure may be deleted.

Eliding synchronization in a concurrent data structure

An interface between a resource manager and schedulers in a process executing on a computer system allows the resource manager to manage the resources of the schedulers. The resource manager communicates with the schedulers using the interface to access statistical information from the schedulers. The statistical information describes the amount of use of the resources by the schedulers. The resource manager also communicates with the schedulers to dynamically allocate and reallocate resources among the schedulers in the same or different processes or computer systems in accordance with the statistical information.

Interface between a resource manager and a scheduler in a process

A process in a computer system creates and uses a meta-scheduler with meta-contexts that execute on meta-virtual processors. The meta-scheduler includes a set of schedulers with scheduler-contexts that execute on virtual processors. The meta-scheduler schedules the scheduler-contexts on the meta-contexts and schedules the meta-contexts on the meta-virtual processors which execute on execution contexts associated with hardware threads.

Meta-scheduler with meta-contexts

Methods and systems for statistically eliding fences in a work stealing algorithm are disclosed. A data structure comprising a head pointer, tail pointer, barrier pointer and an advertising flag allows for dynamic load-balancing across processing resources in computer applications.

Fence elision for work stealing

Growing community of inventors

Bellevue, WA, United States of America

Paul F Ringseth