Work

Texturizing PPCG: Supporting Texture Memory in a Polyhedral Compiler

Year: 2016
Type: conference-paper
Abstract: In this paper, we discuss techniques to transform \nsequential programs to texture/surface memory optimized CUDA \n \nprograms. We achieve this by using PPCG, an automatic paral- \nlelizing compiler based on the Polyhedral model. We implemented \n \na static analysis in PPCG which validates the semantics of the \ntexturized transformed program. Depending on the results of \nthe analysis, our algorithm chooses to use texture and/or surface \nmemory, and alters the Abstract Syntax Tree accordingly. We \nalso modified the code-generation phase of PPCG to take care \nof various subtleties. We evaluated the texturization algorithm \non the PolyBench (4.2.1 beta) benchmark and observed up to \n1.6x speedup with a geometric mean of 1.103X. The title and \nat many places, the paper uses term Texture memory. But, the \noptimizations are for Texture and Surface memory.
Language: English
Open Access status: green
OpenAlex ID: W2886593707
FWCI:
Cites: 6
Cited by: 1
Related to: 9