Performance drawbacks for matrix multiplication using set associative cache in GPU devices
Leonid Djinevski, Sime Arsenovski, Sasko Ristov, Marjan Gušev
Abstract
Leonid Djinevski, Sime Arsenovski, Sasko Ristov, Marjan Gušev
Abstract
Performance of shared memory processors show negative performance impulses (drawbacks) in certain regions for execution of the basic matrix multiplication algorithm. In this paper we continue with analysis of GPU memory hierarchy and corresponding cache memory organization. We give a theoretical analysis why a negative performance impulse appears for specifics problem sizes. The main reason is the cache storage organization, i.e. the negative performance peak appears caused by mapping of matrix elements onto one cache set, instead of using the whole cache. The obtained experimental results prove our theoretical analysis. We also propose a method to avoid situations where performance drawbacks appear.
OpenAlex reports 8 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Performance of shared memory processors show negative performance impulses (drawbacks) in certain regions for execution of the basic matrix multiplication algorithm. In this paper we continue with analysis of GPU memory hierarchy and corresponding cache memory organization. We give a theoretical analysis why a negative performance impulse appears for specifics problem sizes. The main reason is the cache storage organization, i.e. the negative performance peak appears caused by mapping of matrix elements onto one cache set, instead of using the whole cache. The obtained experimental results prove our theoretical analysis. We also propose a method to avoid situations where performance drawbacks appear.
Key concepts: Computer science, Parallel computing, Cache-oblivious algorithm, Cache, Matrix multiplication, Memory hierarchy, Cache algorithms, Cache coloring