GPU thread scheduling optimization method

Abstract

The invention discloses a GPU thread scheduling optimization method. The GPU thread scheduling optimization method includes the steps of numbering thread blocks, mapping the thread blocks to SMs with a hash method, assigning different priority levels to the thread blocks in the same SM, dividing warps in the blocks into groups with fixed sizes according to the stage number of a flow line, and scheduling the warps in the groups, the groups and the blocks with a round-robin method. According to the better GPU thread scheduling optimization method, the problem that all the warps reach the same long delay operation in the same time is solved, through the strategy of warp three-stage scheduling, FU period idling caused by long delay operation is alleviated to some degree, and the use ratio of GPU computing resources is improved.

Description

technical field [0001] The invention relates to a multi-thread scheduling method in a computer system structure, in particular to a GPU thread scheduling method in a heterogeneous system structure. Background technique [0002] Graphics processing units (GPUs) have become a popular platform for executing common parallel applications. Programming systems like CUDA, ATI, and OpenCL allow programmers to parallelize applications into thousands of threads executing the same code. Existing research has also shown that applications running on GPUs have huge speedups compared to running on CPUs. The reason why the GPU can achieve such a huge acceleration is that it has more resources with the same computing power than the CPU. Programmers can make full use of the rich computing resources in the GPU by developing parallelism between threads (TLP, thread-level parallelism). resource. Although there are a large number of computing resources in the GPU, the computing resources in the...

AI Technical Summary

Problems solved by technology

Fair round-robin allows this to happen, however the scheduling policy results in unbalanced warp progress that destroys this locality

But a pure round-robin scheduling strategy tends to make all warps reach the same long-latency operation at the same time, since all warps are stopped, there are not enough warps to hide the long-latency, and the result is that some cycles of FU are idle

Images