Optimizing CUDA code by kernel fusion: application on BLAS

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216224%3A14330%2F15%3A00083436" target="_blank" >RIV/00216224:14330/15:00083436 - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1007/s11227-015-1483-z" target="_blank" >http://dx.doi.org/10.1007/s11227-015-1483-z</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/s11227-015-1483-z" target="_blank" >10.1007/s11227-015-1483-z</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Optimizing CUDA code by kernel fusion: application on BLAS

Popis výsledku v původním jazyce

Contemporary GPUs have significantly higher arithmetic throughput than a memory throughput. Hence, many GPU kernels are memory bound and cannot exploit arithmetic power of the GPU. Examples of memory-bound kernels are BLAS-1 (vector?vector) and BLAS-2 (matrix?vector) operations. However, when kernels share data, kernel fusion can improve memory locality by placing shared data, originally passed via off-chip global memory, into a faster, but distributed on-chip memory. In this paper, we show how kernelsperforming map, reduce or their nested combinations can be fused automatically by our source-to-source compiler. To demonstrate the usability of the compiler, we have implemented several BLAS-1 and BLAS-2 routines and show how the performance of their sequences can be improved by fusions. Compared with similar sequences using CUBLAS, our compiler is able to generate code that is up to 2.24x faster for the examples tested.

Název v anglickém jazyce

Optimizing CUDA code by kernel fusion: application on BLAS

Popis výsledku anglicky

Contemporary GPUs have significantly higher arithmetic throughput than a memory throughput. Hence, many GPU kernels are memory bound and cannot exploit arithmetic power of the GPU. Examples of memory-bound kernels are BLAS-1 (vector?vector) and BLAS-2 (matrix?vector) operations. However, when kernels share data, kernel fusion can improve memory locality by placing shared data, originally passed via off-chip global memory, into a faster, but distributed on-chip memory. In this paper, we show how kernelsperforming map, reduce or their nested combinations can be fused automatically by our source-to-source compiler. To demonstrate the usability of the compiler, we have implemented several BLAS-1 and BLAS-2 routines and show how the performance of their sequences can be improved by fusions. Compared with similar sequences using CUBLAS, our compiler is able to generate code that is up to 2.24x faster for the examples tested.

Druh

J_x - Nezařazeno - Článek v odborném periodiku (Jimp, Jsc a Jost)

CEP obor

IN - Informatika

OECD FORD obor

—

Projekt

<a href="/cs/project/EE2.3.30.0037" target="_blank" >EE2.3.30.0037: Zaměstnáním nejlepších mladých vědců k rozvoji mezinárodní spolupráce</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2015

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

The Journal of Supercomputing

ISSN

0920-8542

e-ISSN

—

Svazek periodika

71

Číslo periodika v rámci svazku

10

Stát vydavatele periodika

US - Spojené státy americké

Počet stran výsledku

24

Strana od-do

3934-3957

Kód UT WoS článku

000361531500013

EID výsledku v databázi Scopus

—