Method and system for managing nested execution streams

Abstract

One embodiment of the present disclosure sets forth an enhanced way for GPUs to queue new computational tasks into a task metadata descriptor queue (TMDQ). Specifically, memory for context data is pre-allocated when a new TMDQ is created. A new TMDQ may be integrated with an existing TMDQ, where computational tasks within that TMDQ include task from each of the original TMDQs. A scheduling operation is executed on completion of each computational task in order to preserve sequential execution of tasks without the use of atomic locking operations. One advantage of the disclosed technique is that GPUs are enabled to queue computational tasks within TMDQs, and also create an arbitrary number of new TMDQs to any arbitrary nesting level, without intervention by the CPU. Processing efficiency is enhanced where the GPU does not wait while the CPU creates and queues tasks.

Description

technical field [0001] The present invention relates generally to computer architecture, and, more particularly, to methods and systems for managing nested execution flows. Background technique [0002] In a conventional computing system having both a central processing unit (CPU) and a graphics processing unit (GPU), the CPU determines which specific computing tasks are performed by the GPU and in what order. GPU computing tasks typically include highly parallel, highly similar operations across parallel datasets, such as images or sets of images. In a conventional GPU execution model, a CPU initiates a specific computing task by selecting a corresponding thread program and directing the GPU to execute a set of parallel instances of the thread program. In conventional GPU execution models, the CPU is often the only entity that can initiate execution of a thread program on the GPU. After all thread instances have finished executing, the GPU must notify the CPU and wait for...

AI Technical Summary

Problems solved by technology

If the GPU uses locking while queuing new tasks, system performance will be negatively affected

Images