-
公开(公告)号:KR1020120052750A
公开(公告)日:2012-05-24
申请号:KR1020100114044
申请日:2010-11-16
Applicant: 삼성전자주식회사
CPC classification number: G06F15/7867 , G06F8/452 , G06F9/38 , G06F9/3853 , G06F9/3897 , G06F11/3409 , G06F9/22
Abstract: PURPOSE: An apparatus and method for dynamically determining a practice mode of a reconfigurable array are provided to practice a loop in the optimum practice mode by dynamically determining a practice mode of the loop based on performance information and an estimate value of running time. CONSTITUTION: A performance information acquisition unit(301) obtains performance information showing running time according to loop repeat count at the VLIW(Very Long Instruction Word) mode and the CGA(Coarse Grained Array) mode. A running time estimating unit(302) obtains an estimate value in regard to the running time of a loop. A code generating unit(303) creates a CGA code of the loop for the CGA mode and a VLIW cod for the VLIW mode. A mode determining unit(304) selects the CGA code or the CGA code by using the performance information and the estimate value.
Abstract translation: 目的:提供用于动态地确定可重构阵列的练习模式的装置和方法,以通过基于性能信息和运行时间的估计值动态地确定循环的练习模式来在最佳练习模式中练习循环。 构成:性能信息获取单元(301)根据VLIW(超长指令字)模式和CGA(粗粒度阵列)模式下的循环重复计数获得表示运行时间的性能信息。 运行时间估计单元(302)获得关于循环的运行时间的估计值。 代码生成单元(303)创建用于CGA模式的循环的CGA代码和用于VLIW模式的VLIW代码。 模式确定单元(304)通过使用演奏信息和估计值来选择CGA码或CGA码。
-
公开(公告)号:KR1020120041581A
公开(公告)日:2012-05-02
申请号:KR1020100103095
申请日:2010-10-21
Applicant: 삼성전자주식회사
CPC classification number: G06F9/5066 , G06F9/38 , G06F15/7892
Abstract: PURPOSE: A reconfigurable processor and a method thereof are provided to reduce processing time and code length of a superposition loop by allocating a loop to a PE. CONSTITUTION: An extracting unit(105) extracts loop execution count information from inner and outer loops in a superposition loop. A loop merger(110) merges the inner and outer loops based on the loop execution count information. A scheduler(120) allocates a command of the inner loop to a PE(Processing Element).
Abstract translation: 目的:提供可重构处理器及其方法,通过向PE分配循环来减少叠加循环的处理时间和代码长度。 构成:提取单元(105)从叠加循环中的内圈和外圈提取循环执行计数信息。 循环合并(110)根据循环执行计数信息合并内循环和外循环。 调度器(120)将内部循环的命令分配给PE(处理单元)。
-