依賴距離主導(dǎo)的向量化方法研究
發(fā)布時(shí)間:2019-06-06 01:20
【摘要】:向量寄存器的非滿載使用方式為大量迭代次數(shù)不足的循環(huán)提供了向量化的機(jī)會(huì),但也導(dǎo)致向量化的并行寬度不固定,傳統(tǒng)的向量因子主導(dǎo)的依賴測(cè)試方法不再適用。提出了一種依賴距離主導(dǎo)的依賴測(cè)試方法,通過(guò)分析依賴圖中所有依賴環(huán)的破環(huán)關(guān)鍵邊所攜帶的依賴距離,選擇其中最小的依賴距離來(lái)決定并行寬度,破除依賴環(huán),實(shí)現(xiàn)基于向量寄存器非滿載使用方式的向量化。實(shí)驗(yàn)結(jié)果表明,該方法能夠有效增加循環(huán)向量化的機(jī)會(huì)和提高向量寄存器的使用率,測(cè)試用例的向量化加速比平均提高14.6%。
[Abstract]:The non-full load usage of vector register provides the opportunity of vector quantification for a large number of cycles with insufficient iterations, but it also leads to the unfixed parallel width of vector, and the traditional vector factor-led dependency test method is no longer applicable. In this paper, a distance-dependent dependency test method is proposed. By analyzing the dependency distance carried by the key edges of all dependency rings in the dependency graph, the minimum dependency distance is selected to determine the parallel width and break the dependency ring. The vector quantification based on the non-full load mode of vector register is realized. The experimental results show that this method can effectively increase the opportunity of cyclic Vectorization and the utilization rate of vector registers, and the Vectorization acceleration ratio of test cases is increased by 14.6% on average.
【作者單位】: 信息工程大學(xué)數(shù)學(xué)工程與先進(jìn)計(jì)算國(guó)家重點(diǎn)實(shí)驗(yàn)室;
【基金】:“核高基”國(guó)家科技重大專項(xiàng)資助項(xiàng)目(2009ZX01036-001-001-2)
【分類號(hào)】:TP332.11
本文編號(hào):2493961
[Abstract]:The non-full load usage of vector register provides the opportunity of vector quantification for a large number of cycles with insufficient iterations, but it also leads to the unfixed parallel width of vector, and the traditional vector factor-led dependency test method is no longer applicable. In this paper, a distance-dependent dependency test method is proposed. By analyzing the dependency distance carried by the key edges of all dependency rings in the dependency graph, the minimum dependency distance is selected to determine the parallel width and break the dependency ring. The vector quantification based on the non-full load mode of vector register is realized. The experimental results show that this method can effectively increase the opportunity of cyclic Vectorization and the utilization rate of vector registers, and the Vectorization acceleration ratio of test cases is increased by 14.6% on average.
【作者單位】: 信息工程大學(xué)數(shù)學(xué)工程與先進(jìn)計(jì)算國(guó)家重點(diǎn)實(shí)驗(yàn)室;
【基金】:“核高基”國(guó)家科技重大專項(xiàng)資助項(xiàng)目(2009ZX01036-001-001-2)
【分類號(hào)】:TP332.11
【相似文獻(xiàn)】
相關(guān)期刊論文 前3條
1 索維毅;趙榮彩;姚遠(yuǎn);劉鵬;;面向DSP的超字并行指令分析和冗余優(yōu)化算法[J];計(jì)算機(jī)應(yīng)用;2012年12期
2 陳向;沈立;;一種面向自動(dòng)向量化和數(shù)據(jù)置換操作的中間表示[J];計(jì)算機(jī)工程與科學(xué);2012年07期
3 ;[J];;年期
相關(guān)會(huì)議論文 前1條
1 黃君輝;劉仲;陳躍躍;;一種基于YHFT-Matrix的FFT向量化實(shí)現(xiàn)[A];第十五屆計(jì)算機(jī)工程與工藝年會(huì)暨第一屆微處理器技術(shù)論壇論文集(A輯)[C];2011年
相關(guān)碩士學(xué)位論文 前2條
1 高翔;集成眾核平臺(tái)科學(xué)計(jì)算應(yīng)用性能測(cè)評(píng)和優(yōu)化研究[D];國(guó)防科學(xué)技術(shù)大學(xué);2014年
2 夏睿杰;基于FT-Matrix2的自動(dòng)向量化關(guān)鍵技術(shù)研究與實(shí)現(xiàn)[D];國(guó)防科學(xué)技術(shù)大學(xué);2015年
,本文編號(hào):2493961
本文鏈接:http://sikaile.net/kejilunwen/jisuanjikexuelunwen/2493961.html
最近更新
教材專著