HSA 模擬平台支援記憶體共享與多個運算單元

An HSA Emulation Platformwith SVM and Multiple Computing Devices Support

徐浩哲(Hsu, Hao Che)

指導教授 ： 鍾葉青 ； 金仲達

國立清華大學/電機資訊學院/資訊工程學系/碩士(2016年)

摘要

隨著計算的複雜度提高以及資料量的增加,我們需要進一步的研發下一個世代的 計算機架構以提升計算機的運算能力,異質運算是近年常被研究及提出來的解決方 案之一,特別是在行動式裝置興起後,因為 SoC 的大量應用與發展,異質架構的整 合變為重要的議題。
HSA (Heterogeneous System Architecture) 是一系列的開放標準,定義了軟體 層與硬體架構,其致力於提供一個異質架構整合的平台,使得異質運算能充份展它 的優勢,也讓應用程式開發者能夠以相對於以往更簡潔的方式運用異質運算加速。
HSA 提供了許多特性,例如:訊號的傳遞、全域共享記憶體的存取、工作分派 的機制,並定義了硬體必須支援的層面,其中有兩個特點:一是共享記憶體的機 制,希望透過記憶體的共享減少設備間的溝通負擔,其二是多個運算設備的平行運 算以達到多工的優勢。
根據 HSA 和 OpenCL 2.0 的標準我們開發了 HSAemu 2.0 ,模擬 HSA 架構並提 供給前期的應用程式開發,這篇論文使 HSAemu 2.0 更進一步模擬 HSA 記憶體管 理單元,透過此單元探討記憶體共享的機制,除此之外也探討了多個運算設備在 多工處理及共享記憶體上使用的優勢。

關鍵字

HSA ； 異質架構 ； 記憶體共享 ； 模擬器 ； 平行化運算

並列摘要

To solve the increasing computing complexity and proliferation of data, there is much research about the next-generation architectures. Heterogeneous computing is one of the solutions to achieve the goals of high performance and efficient power consumption.
Heterogeneous System Architecture (HSA) is a series of standards including the software stack and hardware architecture for heterogeneous computing. HSA has goals to build a friendly platform for multiple devices by providing Shared Virtual Memory (SVM) to reduce the overhead of the data transfer between devices. Not only does the platform improve performance on the heterogeneous architecture but it reduces complexity of programming on heterogeneous computing.
To simulate the new architecture in detail and provide a platform for developing applicationsintheearlystage,webuiltHSAemu2.0,anHSAemulationplatform. In this thesis, we enhance HSAemu 2.0 by implementing the HSA soft-MMU (Software Memory Management Unit) to achieve the goal of shared virtual memory (SVM). In addition, we make HSAemu 2.0 to support multiple computing devices including multi-core CPU and three GPU emulators. The emulation platform proves that utilizing the HSA MMU gets better performance than without HSA MMU and provides a friendly interface for multiple computing devices.

並列關鍵字

HSA ； SVM ； Multiple Computing Devices ； Emulator ； Parallel Computing

參考文獻

[1] P. Ross, "Why CPU Frequency Stalled," in IEEE Spectrum, 2008.

[2] John Nickolls and William J. Dally, "The GPU Computing Era," in IEEE Micro,

[4] G. Teodoro et al., "Coordinating the use of GPU and CPU for improving performance of compute intensive applications," in Cluster Computing and Workshops, 2009.

[5] Enhua Wu and Youquan Liu, "Emerging technology about GPGPU," in APCCAS, 2008.

[13] Fabrice Bellard, "QEMU, a fast and portable dynamic translator," in USENIX, 2005.